Ii?: '::'.:;..~ t~~- Country Study Series . L// /=~ , 56885 -'" - The Manufacturing Sector in Zimbabwe: Dynamics and Constraints Free University of Amsterdam University of Zimbabwe April 1994 This dOCllment is not for citation unless authorized by the RPED Progra.m Manager : The views and interpretations expressed in this study are solely those of the authors. They ! i do not necessarily represent the views of the World Bank or its member countries and . i should not be attributed to the World Bank or its affiliated organizations. I Table of Contents Page 1. Introduction 2. The survey 2.1 Sample selection 2 2.2 Characteristics of the sample 5 2.3 Data collection 8 3. Entrepreneurship: who is a successful entrepreneur'? 3.1 Introduction 9 3.2 Structure of ownership 10 3.3 Theoretical background 11 3.4 Seedbed for entrepreneurial formation 13 3.5 Determinants of entrepreneurial success 17 3.6 Summary and conclusions 19 4. Firm growth in Zimbabwe 1981-1993 4.1 Introduction 27 4.2 Theoretical background 28 4.3 Growth of Zimbabwean firms 30 4.4 Realized growth 1981-1992 34 4.5 Factors determining growth 37 4.6 Effects of the economic crisis 41 4.7 Obstacles to growth 43 4.8 Concluding remarks 45 5. Indigenous and small scale enterprises 5.1 Introduction 48 5.2 Objectives and hypotheses 48 5.3 Characteristics of indigenous enterprises 49 5.4 Conclusion 53 6. The finance of investment by flfDlS 6.1 Theoretical background 55 6.2 Financial intermediation in Zimbabwe 58 6.3 The access of firms to formal finance 62 6.4 Anal ysis of bank credit in major investments 73 6.5 Conclusions 75 7. Investment and capacity utilisation 7.1 Introduction 77 7.2 Obstacles to expansion 77 7.3 Capacity utilisation 83 7.4 Conclusion 91 strata.:! The survey covered only enterprises with 50 or fewer employees (including the proprietor(s», marketing at least 50% of their product and engaged in an economic activity other than agriculture or primary product production. The survey covered over 15,000 households and shops were visited and identified 5,575 micro and small enterprises. For these 5,575 firms we know the number of employees (including working proprietors, unpaid family members and apprentices). Although 60% of the enterprises were engaged in manu- facturing activities, the number fitting our definition of a firm turned out to be small. Of the 5,575 firms only 561 employed at least five employees and of these only about a third (182) were engaged in manufacturing. When confining ourselves to the four selected subsectors this number falls to 107. These firms were included in our sampling frame. In order to combine the cso and the GEMINI lists blow-up factors were applied to the 3 GEMINI firms, corresponding to the sampling fractions. The cso list was then used for firms with more than 50 employees and the GEMINI I ist for enterprises with 50 or fewer (but at least five) employees. 4 GEMINI did not distinguish formal and informal firms. The cso list, however, contains only formal enterprises. Therefore we have implicitly assumed that there are no informal enterprises with more than 50 employees. Sampling was done on the basis of firm size (in terms of employment) in such a way that every worker had an equal probability of being drawn. The sampling method used was fixed interval sampling by sector. Within each sector firms were ordered by size of employment. GEMINI firms were put on the list as many times as the blow-up factor indicated. Selection was done by taking fixed steps from a random starting value. Although drawing firms, the sampling criterion was to give equal probability to each worker of being drawn. Thus, in fact, a worker was drawn and then the corresponding firm was selected. S The selection or replacements For each firm the firms preceding and following it on the list were selected as replacements. Unfortunately, cso did not provide the names of replacements as agreed. 6 As an alternative 2 ·Urban" was dcfmed as Il town with Ill1 estimated 1982 population of more thIlt1 20,000. The urban strata distinguished wcre: high density areas, low density areas. commcreial districts and industrilll areas. The four JUral .n.ta were: .mallcr toWDli. growth points. district councu. and JUral councila. GEMINI ulCd estimated growth ratca of the population of the eight Ib8ta (and the whole of Zimbabwe). baed on growth prior to 1982. to get atimatcs of the population per ItnItum in 1991. The population per ItratUm was divided by average houachold Iizc to get the number of howIcholds per ItratUm. This number of houacholds was then uICd to eaJcu1atc the fmction of enumerated houKholda. The invCrllC of this fraction was uICd as the blow-up factor. 4 Since the IIlmpling of GEMlNl flJTnl was billed on August 1991 employment daIa the minimum .izc condition is IOrnctimcS violatcd in our 1Ilmp1e: a fum which had five worken at the time of the GEMINI IUrvey might havc fewer workers in June J993. s This procedure of drawing a worker and then IeIccting the correaponding fum leada to flJTnl being lelected more thIlt1 once. The Itep Iizc in the interval Klection procedure was adjusted in IUch a _y that fmaIly 200 flJTnl were lelectcd. It must be noted that flJTnl with more employees thIlt1 the step Iizc 1m aI_ys in the 1Ilmp1e. The _ appliCi for GEMINI flJTnl if the product of employment .izc and blow-up fllCtOr is larger thIlt1 the Itep Iizc. 6 For RIlIOOII of confidcntial.ity the u.rnple aelec:tion _ done on the buis of a lilt of identifu:ation numben and fum 8izc, in order not to reveaJ combinations of name and .izc. The cso would only givc UI the names corrcaponding with the identification numbcn of replacemcntl when thia became neceaa.ry. i.c. after non- 4 the CZI Register & Buyers Guide was then used to find suitable replacements. In the case of the GEMINI list the replacements were known, but as it turned out there were not enough of them. Therefore 23 other small firms were selected, belonging to the same sector and situated in the same area as the firm that needed to be replaced . ... Recent starlers .. As the coo list is a so called living data base, it contains recent starters, provided these firms are registered. However, most recent starters will be informal. The GEMINI survey was conducted in August 1991, so by the time the RPED survey was held, almost two years had passed. We have not fully corrected for this. As it turned out, the replacements contained some recent starters. In the original sample 21 out of 67 started in the last two and a half years before they were interviewed (Le. in 1989, 1990 or in the first half of 1991). In our sample on1y 8 out of 67 started in 1991, 1992 or the first half of 1993. This suggests a selection bias if we assume that the age distribution of firms is more or less constant over time. This is a strong assumption however and does not seem very realistic in this case since the drought in 1992 is likely to have had a negative effect on the number of starters. Non-response At the start of the study non-response was expected to be a serious problem since a large number of surveys has been conducted in Zimbabwe. Ideally one would like to know why non-respondents do not want to respond, but it is hard to ask even a few questions when someone has just indicated that he does not want to cooperate. In Zimbabwe, the reasons frequently mentioned were: 1. We do not want to reveal financial information. 2. We do not have time for such a long interview. 3. We are not interested; do not expect that we shall benefit from your research. 4. Our business is going through difficult times and therefore no information can be given (restructuring, bankruptcy, etc.). Especially reasons 1) and 4) are likely to introduce a selection bias. Reason 1) was the most frequently given one. It may have caused a bias towards public enterprises and enterprises with higher gearing, as they have less difficulties with revealing financial information. 2.2 Characteristics or the sample The selected sample consisted of 133 firms from the coo list and 67 from GEMINI, distribu- ted over the four subsectors as follows: Table 2.2: Selected firms by size and sector fOOd wOOd textile metal total cso 34 16 57 26 133 GEMINI 14 10 32 11 67 Total 48 26 89 37 200 re8pOllIIe. However, after the fn three instancea further cooperation was ~fullCd. 5 Among the 67 GEMINI firms, 20 could not be found or had gone out of business. Only 5 of them refused to cooperate. Another 12 were taken out of the sample for various other reasons: three because it would take more than two days of travel to reach them; another three because the employment information from GEMINI turned out to be wrong; another firm because it was a small textile firm that produced sports clothing entirely for the shop of the same owner and the accounts of the firm and the shop could not be separated. In yet another case the owner was abroad and in two cases the interview was incomplete. All eso firms were found. However, 49 of them refused cooperation. Another 8 firms on the eso list had been taken over by a larger firm, already in the sample. and although still registered as a separate division, they turned out to be an integral part of the other firm. It is important to note that firms were selected on the basis of prior information about firm size. This information dated from 1991 for the GEMINI firms and probably longer back for some of the eso firms. 7 The actual size of the firms therefore differs from the size used in the sampling procedure. Not only did it turn out that firms were larger or smaller than expected, sometimes firms had switched to other products and belonged to another sector. If this was one of the selected subsectors the firm was kept in the sample, otherwise it was thrown out. In total we interviewed 205 firms, 107 of which were in the original sample. Four firms were rejected (two from the original sample). Three because the interview was incomplete, and one because the information given seemed highly unreliable. Therefore the final response rate turned out to be 52.2%. In total 201 firms were successfully visited. In order to compare these firms with the originally selected sample, a Table similar to Table 1.2 is presented. Table 2.3: Successfully visited firms by size and sector size food wood textile metal total >50 31 16 54 23124 < = 50 18 10 34 15 77 Total 49 26 88 38 201 In this repon we will use four size classes (in terms of employment): 10 or less ("small"), from 10 up to 100 ("medium"), from 100 up to 250 ("large"), and more than 250 ("very large"). We repeat Table 2.2 and Table 2.3 for this classification. Table 2.4: Selected firms by size (employment) and .ector size food wood textile metal total < I: 10 9 6 17 6 38 11 - 100 8 7 19 10 44 101-250 8 7 14 11 40 > 250 23 6 39 10 78 Total 48 26 89 37 200 7 Although the cso claimed that their data bue was up to date, it turned out that many company namea had changed or that rums had moved years ago. However, the list aIao contained rums that started operating in 1992. 6 Table 2.5: Successfully visited firms by size and sector size foOd woOd textile metal total < = 10 8 6 17 9 40 11 . , 00 14 9 29 14 66 101.250 10 7 21 9 47 > 250 17 4 22 5 48 Total 49 26 89 37 201 Of the 201 firms 7l were started during the colonial period, 56 during UDI and 74 after independence. Of these 74 firms, 25 were started after the beginning of the Economic Structural Adjustment Program (ESAP) in 1989. Table 2.6: Distribution of firms according to period of start-up year :s 64 65· 79 80-88 Oi!! 89 Count 71 56 49 25 Percent 35.32 27.86 24.38 12.44 Location Of the 201 firms visited ItO were located in Harare, 56 in Bulawayo and 35 in regional centres and growth points. When ranked according to size, we observe that the small and medium sized firms were mostly concentrated in the regions. Table 2.7: Size distribution of firms by location employment location 1-10 '1-100 101-250 Oi!! 251 total Harare 17 33 29 31 110 Bulawayo 8 22 15 11 56 other 15 6 35 40 " 66 3 47 48 201 total Sampling by firm We already described how the sample was selected by drawing workers rather than firms. To get some insight in how this affected the sample we applied weights in such a way that we can see how the sample of successfully visited firms looks if we give equal probability to firms 8 · We repeat Tables 2.5, 2.6 and 2.7 to show how important the sampling procedure is. Table 2.8: Firms (weighted), presented by size end sector size food wood textile metal total <= 10 47 15 40 19 1 21 " - 100 8 5 55 4 72 101 - 250 1 1 2 1 5 > 250 1 0 2 0 3 Total 57 21 99 24 201 We observe that sector totals have not changed much, but size classes change dramatically. So, if we had drawn our sample on the basis of equal probability per firm, the number of 8 Finns were given a weight equal to their blow-up factor divided by the minimum of emplomcnt size and step lize, where the step lize was taken from the interval umpling procedure we used. Then the numbers in the different categories were ICIIlcd in IUCh a _y that the total became 201 again, to facilitate compari.aon. 7 small firms would probably have tripled. In fact that was precisely the reason why we decided not to draw on the basis of equal probability per firm. The questionnaire was not designed for a small and medium enterprise survey. Comparison of Table 2.6 and 2.10 indicates that as a result of our choice the number of firms that were started before independence is much higher than if we had chosen the alternative sampling method. Size and age are definitely related. In our weighted sample only 11 firms were started during the colonial period, 37 during UDI and 153 after indepen- dence. Of these 153 firms, 67 were started after the beginning of the Economic Structural Adjustment Program (ESAP) in 1989. Table 2.9: Firms (weighted), presented by period of start-up year s 64 65· 79 80-88:!:: 89 Count ,, 37 86 67 Percent 5.5 18.4 42.8 33.3 To complete the comparison we also show the weighted sample by location, ranked according to size. As small firms are typically located outside the two large cities, giving more weight to them results in a larger number in the other regions. Of the 201 firms 57 would then be located in Harare, 32 in Bulawayo and 112 in regional centres and growth points. Table 2.10: Firms (weighted), presented by size and locationS employment loc8tion 1-10 11-100 101 -250 lit 251 total Harare 38 15 2 2 57 Bulawayo 17 13 1 32 other 66 45 0 112 total 121 73 3 4 201 2.3 Data Collection Firms were interviewed in June-July 1993, almost always by a team of two enumerators, with one a Zimbabwean. In the vast majority of cases interviews were held in English. Interviews typically took an hour and a half for small, informal enterprises and about three hours for large firms. This includes the separate interviewing of workers (a maximum of ten per flI'Ill). Questionnaires were completed on the same day and where gaps were discovered flI'IllS were contacted either in person or by telephone to complete the information. A large number of consistency checks were programmed so that after data entry reports were generated indicating questionable entries. In most cases this could be resolved after reference back to one of the enumerators. In a few cases reference back to the firm was necessary. II Differences between the column tolab and the row tolab preeented in Table 2.9 are due to rounding. 8 3 Entrepreneurship: who is a succesful entrepreneur? Hans Hoogeveen and Moses Tekere If requested to help a country in distress, one should send neither 100 soldiers, nor 100 teachers but 100 entrepreneurs. Abraham Maslow .. 3.1 Introduction Economic progress is in numerous ways related to entrepreneurship. Nevertheless little is known about the art of entrepreneurship and about the entrepreneur in Africa. To highlight the import- ance of entrepreneurship the World Bank (1989) recently noted that: "By creating an environment in which people develop their skills and talents to their full capacity, African countries can make the entrepreneur catalyst a key strategy for promoting sustainable growth with equity." The role and understanding of entrepreneurship has evolved in relation to economic develop- ment. In perhaps Adam Smith's most famous lines on capitalists he wrote: "In spite of their natural selfIShness and rapacity, though they mean only their own convenience, though the sole end which they proposed from the labour of all the thousands they employ, be the gratification of their own vain and insatiable desires ... they arc Icd by an invisible hand and without intending it, without knowing it advance the interests of the society". In the recent literature the entrepreneur is seen as the driving force behind economic growth: as innovator (Schumpeter (1961), Hisrich (1981), Evans (1989», as innovative manager, and decisionmaker (McClelland (1964), Sutton (1954), Liles (1974), Collins and Moore (1970), Hornaday, Abound (1971», as organizer and risk taker (Liles (1974), Palmer (1970), Atkinson (1957), Shapero (1971»1 and as angel of economic renaissance (Wetzel (1986». Such a diversity in interpreting the role of the entrepreneur signifies the need for continuous research. To underscore this diversity Kilby (1971) compares the entrepreneur to the heffalump, that is "A large and important animal which has been hunted by many individuals using various trapping devices.... All who claim to have caught sight of him report that he is enormous but they disagree on his particularities" . The purpose of this chapter is to investigate the specific characteristics of entrepreneurs in four sectors of Zimbabwe's manufacturing industry (food, wood, metal and textile) and to establish how these personal characteristics relate to firm growth. We will consider the contribution of education, previous work experience, apprenticeship, ownership of other businesses. Which groups (social, cultural, ethnic) produce entrepreneurs? Why do minority races in Zimbabwe seem to be more inclined to become entrepreneurs than the blacks, and how do female entrepreneurs perform in comparison to their male counterparts? To investigate these issues a sample of 134 entrepreneurs of which the personal history was recorded was taken from the original data set. The chapter has been divided into six sections. In the next section a description of the characteristics of Zimbabwean entrepreneurs is provided. Section 3 focuses on the theoretical foundation of entrepreneurial formation and growth. Section 4 presents logit estimates of the determinants of the decision to become an entrepreneur and section 5 presents results on the effects of personal characteristics on firm growth. Section 6 concludes. The contributions of these authon are found in: G.A. Kent, D. Sexton and C. Englewood (cds.), Encyclopedia of Entrepreneurship, 1982. 9 3.2 Structure of ownership General characteristicsZ The survey data identify 134 entrepreneurs, 49% of them in the textile sector, 18% in food, 16% in woodworking and 17% in metal working. Most firms are located in Harare or Bulawayo (51 % and 29% respectively), and 20% are based outside these cities. The size distribution shows that 23% of the firms are small, 36% are medium scaled and 21% is large or very large (19%). About half the firms are registered limited liability enterprises and 10% are cooperatives. The vast majority (81 %) of entrepreneurs are men. Female entrepreneurs are concentrated in textiles (27%), compared to just 9% in the metal sector. At an average firm size of 24 employees, female owned firms are substantially smaller than male owned ones whose average employment size is 194 employees. No female entrepreneur owned a very large firm. Black owned firms are much smaller than firms held by non-blacks (25 and 253 employees respectively). The average age of the entrepreneurs is 46.4 years, with the oldest being 84 and the youngest 22. Those that established their own business (74 entrepreneurs) have been in business for 11.3 years on average. 51 % of the entrepreneurs own more than one company and 25% owned a firm that was either closed, sold or went bankrupt. Most entrepreneurs, 58%, established their firms themselves, a remarkable 22 % bought the firms as a going concern and 11 % inherited the business from their parents. This is in line with the observation that 47% of the present business owners have at least one parent who was or is involved in trading or manufacturing. In terms of training, 49 % of the entrepreneurs went up to secondary school and for 23 % primary school was the highest level of education. 33% obtained vocational training while 26% went to university; 2 % did not receive any schooling at all. Apart from their formal education, a third of the entrepreneurs had additional vocational training. Apprenticeship is not popular amongst entrepreneurs; 11 % has experience having been one. After finishing their education, 35 % of the entrepreneurs attended courses (management, accounting, technical etc.). Immediately before becoming the owner of the present business, 48% was employed within the industrial sector and 24% was already self employed. A small fraction of the newborn entrepre- neurs were civil servants, unemployed or finished schoo) immediately before becoming entrepre- neurs. On average new entrepreneurs had nearly eight and a half years of experience in their kind of industry before starting a firm. Start-up investment was financed primarily from own savings (73 %). Zimbabwean banks, foreign banks and donor agencies account for a bit over 7% of the total stan-up capital; 1 % was sourced from money lenders and 12 % came from other sources. The average annual growth of the firms was 10.4% in the period 1981-1986 and 6.3% for the period 1987-1991. In terms of racial distribution 40% of the entrepreneurs are indigenous black owners. Owners of European or Asian origin account for 16 % and 41 % of the firms respectively. But in Harare and Bulawayo entrepreneurs of Asian or European origin own 74% and 78% of the firms respectively. In regions outside the two cities they own only 29% of the companies. Table 3.1: Racial origin of present owners by period of start-up of the firm African Aii., EUropean Asi., Totil Obs colonial 4,2% 25,0% 66,7% 4,2% 100% 24 UOI 29,5% 18,2% 50,0% 2,3% 100% 44 independence 59,1% 9.1% 31,8% 0,0% 100% 44 ESAP 63,6% 13,6% 13,6% 9,2% 100% 44 present distribution 40.3% 15,7% 41.0% 3,0% 100% 134 '2 Annex 1 prcacnu a table with most of the data discuucd in thia 1CCtion. 10 Table 3.1 shows that the unequal distribution of firms has its roots in the colonial and UDI period. when black Zimbabweans started only a fraction of all firms. After independence this has improved considerably, but still the number of firms started by Zimbabweans of European and Asian origin is higher than would be expected on the basis of their numerical presence. In our sub-sample just over 40% of the firms is owned by indigenous black Zimbabweans. .. 3.3 Theoretical background The entry of new entrepreneurs is positively related to the difference between expected entrepre- neurial income and income expected to be gained as an employee. For exit the reverse relationship holds. Hence: where, Y eent expected entrepreneurial income Yeemp expected income as an employee V·1 vector of personal characteristics of the ith individual. Concerning the vector of personal characteristics, the literature broadly distinguishes two sets of factors. The first set comprises psychological factors which include: · personal drive for achievement; McClelland (1961, 1965); Komives (1972) Wainer and Rubin (1969). · internal locus of control i.e. belief in personal ability to control outcomes; Rotter (1966) Brockhaus (1975, 1979). · risk taking propensity; Knight (1921), Atkinson (1951), Palmer (1971), Reuss (1970), lovannovic (1982). personal values; Gasse (1979), Horniday (1979), DeCarlo and Lyons (1979). · originality, willingness to work hard and the ability to provide leadership; Hisrich (1986), Wetge) (1986). The psychological paradigm Suggests that entrepreneurial skills are to a large extent innate. However, this first set falls outside the scope of this chapter because the data available do not capture psychological matters. The second set which is captured by the survey data and relevant for our analysis, comprises socio-environmental factors. The central issue of the sociological set of factors is that entrepre- neurial formations are the result of interacting situational and cultural factors and not innate. Perceptions of desirability, values, and feasibility are products of social environments and cultural predispositions that help determine particular action by an individual in a major shift from one life path to that of an entrepreneur (Shapero and Sokol 1982). The main social-environmental factors are ethnic, cultural and entrepreneurial background (tradition), previous work and business experience, personal skills and level of education (training) and access to finance. These issues deserve elaboration in as much as they constitute to entrepreneurial push and growth. 11 Training Education affects the entrepreneurial ability to start and operate a business and to handle and interpret the relevant information in the environment. Morris and Sommerset (1971, p. 206) in their study of indigenous Kenyan businessmen argued that ~To become an entrepreneur one needs at least a basic level of education to recognise oppoltUnities new at least in his own world of experience". However, this does not imply that any education followed after primary or secondary school increases the probability of becoming an entrepreneur. The influence of education on entrepreneurship also depends on recipient and on the kind of education. Howell (1972), Brockhaus and Nord (1979) have shown that people with an academic education have a very low probability of becoming entrepreneurs. Explanations of such a phenomenon vary from university graduates being overly theoretically educated, to belonging to a group of people who have already good job and promotion oppoltUnities and are less attracted to risky undertakings and the (partly) lowly esteemed work which is unmistakingly connected to entrepreneurship. Vocational training might be singled out as crucial for the formation of appropriate entrepre- neurial and work skills as it emphasizes practical training. Tradition Shapero and Sokol (1982) note that entrepreneurial tradition and acquaintances contribute to the probability of becoming a businessman. This factor is used to explain the high presence of Indians in the West African business community. or of Gujeratis in East-Africa, Ibos in Nigeria, Jews in Europe and America. Common features of the above communities are that they are minorities and . migrants or displaced people with limited job oppoltUnities. . In Zimbabwe belonging to an established minority group or privileged class facilitateS entrepreneurship. Rasmussen (1993, p. 184) notes: "The heterogeneous socio-cultural milieu has facilitated aolidarity and network creation within the groups but not across groups. Scing within one of the powc:rful groupings in c:uu the white bu.incn community, makes capital mobilisation ... access to information much less a problem than would be the ease in a larger, leas atructured and less wealthy grouping." Family background often builds a strong entrepreneurial personality as early life experiences shape prominent patterns of behaviour amongst entrepreneurs (Chell (1985), p.48». Thus relatives provide a stimulating environment towards entrepreneurship and serve as seedbeds for their children to become entrepreneurs either through the example set by them or through inheritance. In this connection Iliffe (1983) notes that: "most successful businessmen in Africa belong to minority groups with internal solidarity and mutual trust. Access to relevant information is accumulated and family succession is ensured as the young generation has no career alternatives (p. 70)." Harris (1970) and Little (1983) emphasize this powerful influence of acquaintances, peers and mentors on the perceptions of the nascent entrepreneur. which can be summarized as: when a workmate far down the production line breaks away to start his own firm his colleagues are likely to follow suit. Entrepreneurial traditions differ by gender. Cenain trades like selling food, being a blacksmith or carpenter are gender specific. Being a woman often is a hindrance to entrepreneurship as was underscored by Madzimbamuto at the third annual congress of the NAMACO: "aocw inequalities limit opportunities for women who are traditionally viewed as mothers and wives rather than as entrepreneurs. The notion of a woman in business is Iti1I not widely accepted... and this has tended to dampen the entrepreneurial spirit of the Zimbabwean businesswoman.· 12 Experience When deciding whether or not to become an entrepreneur, entrepreneurial opportunities are compared to the potential entrepreneur's present situation. Cooper (1973) reported that for about 40% of potential entrepreneurs there is a push factor which Shapero associated with dissatisfaction with the previous position. Brockhaus (1982) discovered that extreme dissatisfaction ... with previous work experience (distrust of everyone in position of authority, pay, opportunity for promotion and relations with co-workers) acts not only as a push factor from the current job but also convinces the worker that no other place of employment is a satisfactory alternative. In addition, to become an entrepreneur one needs industrial specific experience and skills to operate and understand the business, its production processes and markets before starting out on his own. By distinguishing those that have experience in at least two firms, one distinguishes people with an (assumed) wide range of experiences. 3.4 Seedbed for entrepreneurial rormation The decision to become an entrepreneur is only taken by a selected group of people and here the focus is on the differences between those that started their own business, and those that did not. Using logit analysis we try to estimate who became entrepreneur. Although the original data set was not designed for this use, a sub-sample could be compiled in which employees and entrepre- neurs were put together. A total of 1644 employees has been taken from the workers questionnaire. Although the data set included 1717 workers, 73 had to be eliminated because observations were missing. We have made no correction for the under-representation of workers in small firms (in firms with less than 10 employees the criterium of interviewing 10 employees could not be met) and for the dispropor- tionate presence of certain categories of workers (management is overrepresented and workers are underrepresented). 3 The entrepreneurs (72) in the sub-sample have been chosen according to the following criteria: (1) they were employee (or apprentice) before they started their business and (2) they have to be in a position to take decisions: entrepreneurs in fl11I1S with any foreign or state ownership were excluded, so that we were only left with private Zimbabwean owners. The entrepreneurs in this sample not only started a business, but, as a result of the sampling procedure, virtually all of them (95 %) have run their business for three or more years. As half of the newly established businesses in Zimbabwe go out of operation before the end of the third year (Mead (1993), p.ii), the entrepreneurs considered are not average entrepreneurs but have shown to be successful in terms of survival. As such their companies are likely to differ from those that started a business and failed. The sub-sample consists of 1572 indigenous Zimbabweans and 144 non-Africans. Of the non-Africans 40 are entrepreneur. The two groups are compared in Table 3.2. 3 In annex 2 com:ction facton that could be UlICd an: presented. 13 Table 3.2: Differences between Africans and non-Africans non-African African f -statistic probability of being equal years of education 12.2 8.4 170.0 0.0000 age 41.5 35.7 32.0 0.0000 wage 3435.2 932.5 410.0 0.0000 % male 75.9 82.7 3.5 0.0623 % city 82.4 84.9 0.6 0.4538 % primary education 6.8 41.9 59.0 0.0000 % secondary education 54.2 51.6 0.3 0.5834 % vocational tr. 16.9 2.8 63.0 0.0000 % university 22.0 2.0 150.0 0.0000 % experience 44.5 32.9 6.7 0.0097 age and wage are not presented as logarithms Because of these differences, the model tested below makes an explicit distinction between Africans and non-Africans. Model To test the entrepreneurial supply model, it is written as follows: Pi,ent = F(Ycent *Vr Yeemp *Vi) where, Pi,ent the propensity of the ith individual to become an entrepreneur Yeent the average expected income of entrepreneurs. Yeemp the average expected employee income_ . V·I vector of personal characteristics of the ith entrepreneur The average entrepreneurial income is a constant (c) to everybody: Ycent = c. The individual expected income dep~nds on the individual's characteristics times this constant. As an approxima- tion for the expected average employee income times the personal vector, the present wage (Wi) of the employee has been taken. Thus the above equation can be written as: Pi,ent= 8 1 (c*Vj ) + 8:z Wi + E Since the data set contains both employees and employers, the wage employers would have earned if they had been employees has to be estimated. The attributed wage is the average wages for employees in the upper echelons (technicians, supervisors, foremen, commercial, sales, adminis- trative and clerical staff and management), depending on the level of education. The following wages have been assigned: 14 Table 3.3: Monthly wages attributed to entrepreneurs Z$ no-education 780,00 primary school 1086.60 secondary school 2102.65 vocational Training 2870.54 · university 3809.17 in the estimations logarithms of monthly wages have been used, without incorporating any in kind payments or cash allowances. The vector Vi has been composed depending on the information available in the data set and comprises: age the logarithm of the age of a person as a proxy for working experience D-african dummy for black Zimbabweans D-gender dummy for men D-city dummy for firms located in Harare or Bulawayo education years of education: vocational training was assumed to add one year of schooling, polytechnic and apprenticeship two years and university three years D-secondary school dummy for people who spent the last year of schooling in secondary school D-vocational training dummy for those with vocational training as highest level of education D-university dummy for those who reached university j)-experience dummy for those with experience in more than one firm in the industry they are presently working in The estimations distinguishes between Africans and non-Africans for those variables that show significant differences between the two groups. Results Table 3.4 presents results of four logit estimations. The first two present outcomes for Africans and non-Africans respectively. The third estimation contains results for the whole sample. The fourth repeats the t.Ilird one. but with years of education replaced by dummies for the different types of education. Distinctions between black Zimbabwean and the whole sample have only been made in the third and the fourth estimation as far as Table 3.2 sbows that a significant difference exists between bla;;k Zimbabweans and Zimbabweans of European and Asian origins. The results are promising as between 23 (32~) and 33 (46~) of the entrepreneurs are correctly predicted (estimation #3 and #4). The results are even better for the for non-Africans correctly predicting 65 %. In the estimation for Africans, where entrepreneurs only consist 2 % of the sample, 25% of the businessmen are still identified correctly . .. 15 Table 3.4: Results of logit estimation on becoming entrepreneur Africans N.Africans full sample full sample #1 #2 #3 #4 constant -23.147 -5.45 0.430 0.10 -6.156 1.50 -0.110 -0.03 age 2.749 2.67 2.807 2.78 3.143 3.32 2.559 2.74 (afr· age) -0.302 -0.22 0.292 0.21 O-african -17.419 3.02 -23.343 -4.01 O-Gender -0.822 -1.24 3.276 3.06 2.843 2.85 3.001 3.05 O-(afr· gender) -3.943 -3.42 -3.837 -3.21 O-city' -2.225 -4.42 -3.229 -3.55 -2.289 -5.98 -2.481 -5.81 years of education 0.333 2.95 (atr·years of educ) -0.352 -2.62 O-sec. school 0.541 0.91 0.091 0.07 -0.453 0.84 O-vocational tr. 2.532 3.86 2.583 1.73 1.837 2.11 D-(afr· vocational) 0.798 0.86 O-university -0.649 -0.62 2.989 1.94 2.240 2.55 0-(a1r· uni) -2.881 -2.40 O-experience 1.503 3.22 2.041 2.88 1.610 2.85 1.761 2.87 O-(afr· experience) 0.073 0.10 -0.213 -0.28 wage 1.537 5.30 -1.765 -2.93 -1.469 -3.03 -1.530 -2.92 (afT· wage) 3.072 5.47 3.071 5.16 observations 1572 112 1716 1716 entrepreneurs 32 40 72 72 estimate 9 31 31 38 correct wrong , 8 26 5 23 8 33 5 age and wage are used as logarithms, education not wage is monthly wage With respect to the wages earned we fmd that the predictions of the entrepreneur supply model are confirmed for non-Africans (the sign is negative and significant at the 1 % level). This is not the case for indigenous Zimbabweans. This result is remarkable, especially as it is highly significant. A possible reason is that African wages are substantially lower than the entrepreneurial wages attributed (see Table 3.4). Thus entrepreneurship remains an attractive alternative, but can only be reached by those able to save enough to get started. The ones who are most likely to be in a position to save are the higher earning Africans, thus explaining the positive sign (see also section 6, on access to finance). Considering the factors constituting the personal vector we find that: · The dummy for city corrects for the underrepresentation of non-city employees in the sample. · Being a black Zimbabwean is (was) a very high obstacle to entrepreneurship · Experience, both when measured by years of experience (age) and by having experience in more than one firm, is a significant contributing factor to entrepreneurship. · The positive and significant gender dummy indicates that being a women is a hindrance to entrepreneurship. This is only the case as far as it concerns non-African women. · Level of education and entrepreneurship are positively related. Primary education (results presented in Annex 3) is insignificant as is secondary education irrespective of racial background. Vocational training is a significant contributing factor to entrepreneurship like university education. University education is not a positive factor for indigenous Zimbabweans. This result explains why the variable for years of education (estimation #3) is insignificant for Africans. 16 3.5 Determinants of entrepreneurial success It is one thing to start a business and to survive the first three years but something else to become successful that is to grow bigger. The analysis of firm growth generally concentrates on age and size of the firms (Varyam and Kraybill (1992), Wagner (1992), Evans (1987a». Sometimes more factors are incorporated like the number of plants (Evans, 1987b) or the capital-labour ratio, promotional efforts, unionization, innovations or productivity (Acs and Audretsch (1990». All these approaches have in common that they consider the firm as the unit of analysis. They overlook that - at least in firms managed by their owner - personal characteristics of entrepreneurs are the underlying factor of firm growth. Growth of (entrepreneurial) firms thus revolves around the strengths of the entrepreneur. Mead, Mukwenha, and Reed (1993, p. ii) not only found that in Zimbabwe just 50% of the newly started firms survive, they also discovered that of those that survived Jess than 20% add any workers while only 1% succeed in moving from very small into medium scale operations. In a rather descriptive analysis, they related owning a successful firm to personal characteristics of the entrepreneur, like willingness to work hard or to take calculated risks, experience etc. In this section we will also consider the significance of the entrepreneurs personal characteristics for firm growth. Model To measure the impact of different variables on firm growth we take Evans' specification (Evans 1987a,b)4: . in which G is a growth function, Art the age of firm i in year t, Sit the size of firm i in year t and e a lognormally distributed error term. Evans specifies G as a second order expansion of Log(Sit' Ait). To be explained is average annual growth rate. Hence we write: [Log(Sit)-Log(Si,t_k)]Ik= 60 + 6t Log(Si,t_k) + .6:2[Log(Si,t-kl + 63Log(Ai,t-k) + 64 [Log(Ai ,t-k)J2 + [6sLog(Si,t_k) * Log(~,t-k)] + tv.. "'iJ + p.g .. Apart from firm size and age which are enforced by the model, we employ the following -entrepreneur related- variables for Xij: firm size firm size (measured by the log of employment) is inc1uded to test Gibrat's law (firm growth is independent of firm size), firm age When considering Gibrat's law, Jovanovics contradiction of it (firm age and growth are inversely related) needs to be considered as well. Firm age has been taken as a logarithm. 4 . The aame equation is UIed in Chapter 4. To allow for eomparison we apply the dcfm.itiorul u they are ulCd in that chapter. Thus growth is measured in employment tenna and the time intcrvala ulCd are 1981-1986 and 1986-1992, when: 1992 indicalcs that for the determination of employment in 1992 the mean of the employment in 1991, 1992 and 1993 hu been taken. 17 education Like in the previous section education has been measured in two ways. A continu- ous variable (educ) measures the number of years of education, while dummy variables capture the highest level attained. gender dummy incorporated to find out if gender discrimination hinders female entrepre- neurs racial origin dummy for black indigenous entrepreneurs, incorporated to discover jf racial differences influence growth rates family Entrepreneurs with parents owning a trading or manufacturing business are expected to have learned valuable skills enabling them to grow faster firms A successful firm owner might decide not to invest in the existing firm but to diversify in order to spread his risk. Owning more than one manufacturing or trading firm is thus expected to have a negative impact on growth. A reverse relationship might also hold, as synergetical effects are attained. The number of firms held by an entrepreneur has only been recorded for the current year. The sub-sample used comprises 134 firms in which the general manager is the actual owner. However only 61 entrepreneurs provided information as far back as 1981 so we had to discard 73 observations to be able to estimate growth in the first time period (1981-1986). For the second period (1986-1992') 36 observations were eliminated, leaving us with 98 firms. Results Table 3.5: Results of OLS estimation of firm growth 198'-86 1986-92' 1986-92 coeff t-statistic coaff t-statistic coaff t-statistic constant 5.87E-Ol 2.75 -9.58E-02 -0.54 1.81 E-Ol 1.67 previous growth S.78E-02 0.65 size -7.39E-02 -4.55 -8.68E-03 -D.55 -1.59E-02 -'.28 size2 S.46E-07 0.78 -2.S3E-07 -0.84 -, .33E-07 -D.37 firm age -1.17E-07 -0.70 3.S1E-08 0.90 0.0192E-08 0.43 firm age2 -6.14E-02 -0.97 -, .48E-02 -0.30 -S.94E-D2 -3.18 (size· firmage) -6.13E-06 -0.30 3.49E-05 2.24 5.23E-05 3.41 years of educ ·3.85E-02 -2.83 2.28E-02 2.02 , .76E·02 '.66 years of educ2 2.63E-03 3.86 -, .21 E·03 -2.01 -7.51E·04 -, .36 O-sec. school 6.15E-02 1.67 5.69E-02 1.97 9.01E-02 3.26 D-african -, .OOE·O' -, .80 2.40E·02 0.54 3.03E-02 0.82 O-gender -1.49E-03 -0.04 6.30E-02 2.22 5.40E-02 1.97 O-family 1.38E-02 0.44 3.65E-02 1.50 S.30E-02 2.24 O-firms 4.56E-02 1.64 8.0SE-03 0.35 -, .76E-03 -0.08 n 61 61 98 R2 0.571 0.428 0.393 Adj R2 0.464 0.270 0.307 OW 2.05 1.75 2.22 From these results we learn with respect to firm related variables that: · Previous growth is not significant. Growth is not persistent, or, to put it differently: owners that attained high growth rates after independence were not more likely to grow faster after 1986 than those that did not grow much after independence. · For the 1981-86 period Gibrat's law has to be rejected, as a negative Oinear) relation between 18 firm size and growth is discovered. It seems to hold for the period after 1986, but has. to be rejected for two reasons. Firstly the combination of size and firm age is significant leavmg us with a positive relation between firm size and growth. Secondly, in the third regression, (firm age)2 is significantly negative indicating that growth reduces with firm age and hence size. With respect to the other owner specific variables we find that: · Years of education and the quadratic term are both significant in both periods. Remarkably, .. education changed from a quadratic convex function in the early eighties to a quadratic concave function in the late eighties, implying that after independence growth opportunities for high educated peofle were good but that they have decreased considerably. · Of the dummy's indicating the highest level of education reached, the one for secondary education was significantly positive and vocational training significantly negative. The dummies for primary education and university were not significant. · In the period just after independence being an indigenous black Zimbabwean reduced firm growth. In the second period, after 1986, this legacy of the colonial days seems to have disappeared. · Gender and having parents in business became important since 1986. Female entrepreneurs grow less, while stemming from an entrepreneurial background reveals the expected positive effect. · Owning more than one firm was insignificant in the second time interval. Apparently owning more than one firm does not ensue synergetical effects, but it doesn't hamper growth either. In the 1981-86 period a positive relation between firm growth and owning several firms is found. It is not clear if the entrepreneurs that owned more than one firm at the time of the survey, already held them in 1981. Intuitively this is unlikely, especially for the whole group. This then could indicate that high growth in the first period resulted in diversifying strategies of the owners. The explanatory power of the estimates is acceptable given the number of observations and the fairly straightforward variables used. The results compare well to those presented in Tables 4.4 and 4.9, where firm-related variables are used. So firm growth in owner-managed firms can be explained by variables such as location, sector, exports or foreign ownership but also by the entrepreneurs' personal characteristics like education, family and race. 3.6 Summary and conclusions Education determines the capability to become an entrepreneur, but also to survive the first three years in business and to let the firm grow. For the decision to start a firm what matters is not primary or secondary education but vocational training or university. The observation found in the literature (Brockbaus (1982» that university education hampers entrepreneurial formation is not confIrmed, except for black Zimbabweans. With regard to firm growth, a higher educational level of the entrepreneur increases firm growth, but with diminishing returns. The optimal level of education appears to be secondary. school. University education as such does not contribute to firm growth, while vocational training even hampers growth. The nature of vocational training (which is often practical with the specific aim that the pupils become self supponing craftsmen) thus creates a good seedbed for entrepreneursbip, but does not guarantee growth. Many possibly start because of the lack of other career opportunities. This can be derived from the number of people with vocational training that start a business, and from the fact that higber wages reduce the probability of someone becoming an entrepreneur. This last relationship does not S The I'IIIUlta on the educational dummy'. other than ICCOOda.Iy IChool arc preeentcd in annex 4. 19 hold for (generally low earning) black Zimbabweans. With increasing wages they are more inclined to start a business. One could infer from this that it is not a lack of entrepreneurial spirit, but lack of finance that prevents firm formation by indigenous Zimbabweans. This suggestion is supported by the fact that 73 % of the start-up capital comes from own savings. Working experience is obviously useful for starting a business. Both years of experience (approximated by the respondent's age) and having experience in more than one firm were significant. The relationship between firm growth and experience as an entrepreneur is less straightforward. Even when we assume that for entrepreneurs firm age is an indication of experience as an entrepreneur (something which clearly does not hold for those that inherited or bought the firm) the relation between experience (firm age) is not clear. Considering the emergence of indigenous Zimbabwean entrepreneurship one notes that colonial rule limited entrepreneurial opportunities for blacks in favour of European and Asian entrepreneurs. Segregation and limited access to markets, finance and skills prevented business diffusion into the African society: "Africans corning to town should sell only curios and baskets but not chickens and eggs ... 6 These past inequalities have a bearing on the present· composition of black and non-black entrepreneurs, not only through the direct restraints put upon blacks before independence, but also through the still observable effects of having had less access to education. Another factor discriminating indigenous Zimbabweans -but which could not be investigated-, might be that blacks have had less opportunity to gather wealth, thus blocking them from entrepreneurship. A third reason can be found in the fact that minority groups in Zimbabwe have the right social and business networks which stimulate becoming a businessman (see Chapter 4). In terms of growth the racial inequalities of the past had their influence in the period after independence but seem to have been overcome thereafter. However as the mean black enterprise is much smaller than the non-black enterprise, differences in growth rates would be expected as larger firms grow at a lower speed. This not being the case indicates that black entrepreneurs have not yet caught up completely, although the growth-gap between the two groups has narrowed to such an extent that it can not be measured directl y any more. · Being a woman reduces the probability of becoming an entrepreneur, but only as far as it concerns non·African women. Madzimbamuw's remarks (section 3.2) concerning the limited opportunities of African women are not confirmed by our study and seem to have more relevance for European and Asian women, who, generally being from a wealthier background, are less prone to economic push factors than black women. In terms of growth performance, being a women reduces growth since 1986, and this is a worrying development. Finally, the estimation of growth has shown that growth of entrepreneurial firms can just as well be estimated by using personal characteristics as by employing attributes of the firm. Although the analysis highlighted some characteristics of successful entrepreneurs, introducing policies to support them directly will be difficult because, to use Baumol's words (1986, p. 30): "the nature of the entrepreneur will forever elude us because the moment it is recognised... it is transmuted into something else. To observe the subject is to make it disappear." Policies designed to improve education (or easing access to finance), with special attention given towards blacks and women might therefore prove more beneficial. 6 O. Huggins, Prime Minister of the FedenWon of Rhodesia and NyuaJand, cited in Gray (1960). p. 154. 20 Annex 1: Description of entrepreneurs and their firms In = 134) sector firm size food 17.9 small 27.6 wood 15.7 medium 42.5 textile 49.3 large 16.4 metal 17.2 very large 13.4 racial origin legal status .. African 40.3 sole owner 31.3 Asian 15.7 limited liability ent. 49.3 European 41.0 cooperative 9.7 Other 3.0 other 9:7 education prior occupation none 2.2 private sector emp!. 48.5 primary School 23.1 self-employed 23.9 secondary School 48.5 schooling 9.0 university 26.1 other 18.6 location gender Harare 50.8 male 80.6 Bulawayo 29.1 female 19.4 Other 20.2 Annex 2: Correction factors for workers management 0.359017 adminst/clerical 0.581758 commercial/sales 1.796616 supervisorlforeman 0.396981 technicians 0.746084 equipment maintenance 0.813712 skilled production workers 0.977032 other production workers 1.2266' , apprentices/trainees 0.491979 support staff 1 .382608 21 Annex 3: Logit estimate on entrepreneurship, including primary education constant -0.450 -0.' 2 age 2.642 2.78 (atr· age) 0.266 0.19 O-african -23.375 -4.03 O-gender 3.047 3.05 O-Iafr· gender) -3.886 -3.22 O-city -2.472 -5.77 O-primary school -0.143 -0.11 O-(afr·prim) 0.480 0.33 O-vocational tr. 2.235 3.02 O-Iafr·voc. tr.) 0.799 0.84 O-university 2.661 3.64 O-Iafr· uni) -2.870 -2.38 O-exp. 1.797 2.89 O-{afr· experience) -0.250 -0.33 wage -, .584 -2.96 (afr·wage) 3.094 5.03 estimate 39 correct 33 wrong 6 22 Annex 4a: OLS estimations of firm growth including primary, secondary and university education 1981-'86 1986-'92 1986-'92 coefficient t-statistic coefficient t·statistlc coefficient t-statistic constant 6.42E-Ol 2.56 -1.78E-Ol -0.86 1.93E-Ol 1.63 previous growth 6.14E-02 0.68 size -7.79E-02 -4.64 -S.94E-03 -0.55 -1.65E-02 -1.29 size 2 1.11E-06 0.90 -2.63E-07 -0.82 -1.2SE-07 -0.35 firm age -1.40E-07 -0.83 3.S3E-OS 0.90 1.S6E-OS 0.41 firm age 2 -7.59E-02 -1.10 2.43E-03 0.05 -1.01E-01 -3.14 ISize*firmage -5.74E-06 -D.27 3.25E-05 2.02 S.23E-OS 3.35 years of educ -3.18E-02 -2.12 2.28E-02 1.86 1.S4E-02 1.56 years of educ 2 2.03E-03 2.29 -1.01E-03 -1.39 -S.50E-04 -1.18 O-prim. school 1.S1E-02 0.23 3.91E-02 0.72 -9.0SE-03 -0.24 O-sec. school 6.89E-02 1.81 5.77E-02 1.91 8.87E-02 3.01 O-university 6.98E-02 1.00 -3.19E-02 -0.59 1.10E-02 0.21 O-african -1.17E-Ol -, .91 3.89E-02 0.80 2.6SE-02 0.67 O-gender -2.21 E-03 -O.OS 6.02E-02 2.08 5.52E-02 1.97 O-family 1.49E-02 0.42 4.53E-D2 1.6S 5.16E-02 2.09 O-firms 4.31E-02 1.51 6.11E-03 0.26 -, .3SE·03 -0.06 n 61 61 98 R2 0.583 0.437 0.394 Adj R2 0.456 0.249 0.291 OW 2.04 1.73 2.20 23 Annex 4b: OLS estimations of firm growth. including vocational training 1981·'86 1986-'92 1986-192 coefficient t-statistic coefficient t-statistlc coefficient t-stetistic constant 6.05E-Ol 2.87 -S.01E-02 -0.50 1.61 E-O' 1.44 previous growth 7.38E-02 0.83 size -7.S7E-02 -5.00 -1.27E-02 -0.79 -1.09E-02 -0.85 size 2 1.14E-06 0.95 -2.31E-07 -0.73 -2.1SE-07 -0.S9 firm age -1.42E-07 -0.S7 3.13E-OS 0.80 2.9SE-OS 0.64 firm age 2 -6. 66E-02 -1.06 -1.63E-02 -0.33 -1.0SE-01 -3.24 (size· firm age) -7.2ge-D6 -0.36 3.37E-OS 2.15 5.13E-05 3.22 years of educ -2.04E-02 -1.48 3.72E-02 3.39 3.68E-D2 3.34 years of educ 2 1.61E-03 2.40 -2.03E-03 -3.65 -1.84E-03 -3.35 O-voc. training -6.54E-02 -2.02 -4.S1E-D2 -1.71 -4.81 E-02 -2.01 O-african -l.OSE-O' -2.02 l.49E-02 0.34 l.80E-02 0.47 O-gender -'.7SE-03 -0.05 5.25E-02 2.19 5.79E-02 2.04 O-tamilv 2.l7E-02 0.70 4.33E-D2 1.77 4.91E-02 1.S9 O-firms 4.32E-02 1.57 7.11E-03 0.30 -3.l3E-04 -0.01 n 51 51 98 R2 0.582 0.417 0.348 Adj R2 0.477 0.256 0.256 OW 2.06 1.74 2.20 24 References Acs, Z.l. and D.B. Audretsch, 1990, The determinants of small-firm growth in US manufacturing, Applied Economics 22: 143-153. Baumol, l.W., 1986, Toward operational models of entrepreneurship. In J. Ronen (ed.), 1986, Entrepreneurship, Lexington Books, Toronto Brockhaus, R.H .· 1988, The psychology of the entrepreneur, Journal of Small Business Management Chell, E., 1985, The entrepreneurial personality: A few ghosts laid to rest, International Small Business Journal Elias, M., 1990, Women entrepreneurs in Africa: their condition, constraints and future direction, SAPEM 3, Sapes Trust, Harare Elkan, W., 1988, Entrepreneurs and entrepreneurship in Africa, World Bank Research Observer 3: 171-188 Evans, D.S., 1987a, The relationship between firm growth. size and age: estimates for 100 manufacturing firms, Journal of Industrial Economics 4: 567-581 Evans, D.S., 1987b, Tests of Alternative Theories of Firm Growth, Journal of Political Economy 95: 657- 674 Evans, D.S. and L. Leighton, 1989, Some empirical aspects of entrepreneurs, American Economic Review: 519-535 Gilder, G., 1984, The spirit of enterprise, Simon and Schuster, New York Gray, I.R., 1960, The two nations: aspects of the development of race relations in Rhodesia and Nyasaland, Oxford University Press Horniday J.A., Research about living entrepreneurs: characteristics of successful entrepreneurs. In C.A. Kent, D. Sexton, L. Vespar L. and C. Englewood (eds.). 1982, Encyclopedia of Entrepreneurship, Prentice Hall, New York Jayachandran, N.V. and S.D. KraybiIl. 1992, Empirical evidence on determinants of firm growth. Economic Letters 38: 31-36. Jobannison, B., 1988, Business formation: a network approach, Scandinavian Journal of Management 4 Kilby, P., 1971, Hunting the beffalump. In P. Kilby (ed.). 1971, Entrepreneurial and economic develop- ment, Free Press, New York Madzimbamuto, V·· Sustainable small and medium scale enterprises through manpower development, paper presented at the 3rd annual congress of the NAMACO Marsden, K., 1990, African entrepreneurs; pioneers of development, World Bank, Washington Morris, P. and A. Sommerset, 1971, Education and training, A/rican Businessmen: 209-241 McClelland, D., 1964, The achieving socit!ty, Van Nostrand, Princeton McClelland, D., 1969, Motivating economic development, Free Press, New York 25 Mead, D.C., H.D. Mukwenha and L. Reed, 1993, Growth and transformation among small and medium entrepreneurs in Zimbabwe, working paper of the University of Zimbabwe/GEMINI, Harare Rasmussen, J., 1992, The local entrepreneurial milieu. Entrepreneurial networks in small Zimbabwean towns, Research Report (79), Roskilde University, Copenhagen Schumpeter, J., 1958, Capitalism, Socialism and Democracy, Simon and Schuster, New York Shapero, A., 1975, The displaced, uncomfortable entrepreneur, Psychology today 9: 83-88. Shapero, A. and L. Sokol, The social dimensions of entrepreneurship. In Kent, C.A., D. Sexton, L. Vespar and C. Englewood (eds.), 1982, The Encyclopedia of Entrepreneurship, Prentice Hall, New York Smith A., 1937, The Wealth of Nations, led] Edwin Cannon, Modem Literature Susbauer, J.C., 1969, lntracorporare entrepreneurship, Cleveland State University Wagner, J., Firm Size, 1992, Firm Growth, and Persistence of Change: Testing Gibrat's Law with Establishment Data from Lower Saxony, 1978-1989, Small Business Economics 4: 125-131 World Bank, 1988, Africa has plenty of entrepreneurs; what they lack are opportunities, World Bank Research News 8 (2): 5-6 Zambezia, 1992, An outline of African business history in colonial Zimbabwe, Journal of the University of Zimbabwe 19 (l) 26 4 Finn growth in Zimbabwe 1981-1993 Peter Risseeuw 4.1 Introduction The analysis of growth of individual firms is often denoted by Firm Dynamics, although this concept is primarily associated with entry and exit of firms. In a dynamic market, each replacement of one firm by another is assumed to result in a gain of efficiency at the meso- or macro-level. If in this process the new firms introduce new technologies, the process is called 'creative destruction', a term introduced by Schumpeter (1950). In this paradigm high entry and exit rates are seen as indicators of a prosperous economic development. On the other hand, an economic crisis is a push factor for new entrepreneurship (cf. chapter 3), and high entry rates can also indicate an economy in despair. Mead et aI. (1993), focusing on very small firms, in or nearby the informal sector, find very high birth rates for Zimbabwe: the estimated number of newly established firms per annum year lies between 20 and 30 percent of the number of incumbent firms. Most of the newly established firms do not live long, survival rates are extremely low. The step from small/informal to medium-sized/formal is too large for 99% of the newly founded firms: even if they survive, they do not grow at alL Given the background of the Zimbabwean economy, these high entry rates are to be interpreted as a symptom of the slump, and not as a process of creative destruction. In the current analysis, the focus is on the formal sector. Although firms from Mead's GEMINI-dataset were included in the sample, a minimum level of five employees (in 1992) was used as a lower bound. I This filter does not exclude informal firms, but the emphasis lies on firms that have already taken the step towards a more formal status, and Mead's conclusions sbould not be adopted without further inspection. Analysis of entry and exit cannot be carried out with the current dataset, because such data are not included. The firms in the sample were interviewed about their current status (in 1993) and on their recent history (from 1981 on). Since one of the main attributes these firms share is 'baving survived until date', no comparison with non-surviving firms can be made. The cross-section will be turned into a panel, by following the development of the firms in 1994 and 1995. After these rounds, we expect to have more information on t!y1Ul- mics. For the moment, firm growth will be considered. While exit and entry are disconti- nuities (events), growth and decline of firms are regarded as continuous processes. This chapter focuses on the following questions: · Can growth of firms be explained by attributes of the firms themselves (such as size, age, location) or attributes or the owners/entrepreneurs (age, race)?2 · Are there sectoral differences, and if so, to what extent do they matter? · Can the influence of the drought in 1992 be measured? Is there a different impact for different sectors or types of firms? · Can the quality of the infrastructure and the system of regulations be identified as barriers to growth? I Sec chapter 2 for the dc:talla of the sampling procedure. 2 In this chapter the unit of analylia ill the firm. In chapter 3 growth at the level of the entrepre'Mllr ill dealt with. 27 A methodological problem rises when one tries to find answers to such questions with data from an incomplete sample. We have no information on firms that quit in the period 1981-1993 so that the sample is censored. No econometric panacea is available to meet this problem. In further rounds the analysis will acquire a more longitudinal nature and the panel will become more like a cohort, though it will never become a real one. The problem will diminish then. Still, we have reason to assume that the sample as it is, is representative for the population of Zimbabwean firms in the industries under studies. From Mead et al. 's (1993) observation that the vast majority of (very) small firms does not grow at all, we can derive that entry in the informal sector is relatively easy, but that consolidation and expansion towards a more formal status is very difficult. The barriers on entry and exit in the Zimbabwean economy are apparently located on the border between the informal and the formal sector, due to location regulations, long waiting times for getting licences, foreign exchange controls (only just removed in 1993) etc. Since deregulation is a central concern of the Structural Adjustment Program (ESAP) , it will be interesting to see whether more of these barriers are removed or at least lowered in the next few years. As a consequence, buying a distressed or (nearly) bankrupt firm in the (desired) industry turns out to be cheaper than establishing a new one for many entrepreneurs.:; Besides, the class of white entrepreneurs in Zimbabwe can -within boundaries- be seen as a closed shop: entrepreneurship is a lifetime job. If there is no continuity at the level of firms, then there is at the level of entrepreneurs. Therefore, the analysis as it is carried out now, yields valuable information about factors determining firm growth, provided that firms have been able to survive and have crossed the barrier towards formality in an earlier stage. In section 4.2 some theoretical considerations are presented together with some well- known empirical findings. In section 4.3 realized growth of the firms in our sample is described, while the actual analysis is presented in sections 4.4 (bivariate analysis of growth by firm characteristics) and 4.5 (multivariate analysis). Section 4.6 discusses the effects of the drought and the impact of ESAP. In section 4.7 actual growth and perceived obstacles to growth are related. Section 4.8 gives some concluding remarks. 4.2 Theoretical background The theory we apply in this chapter is based on the empirical school of growth analysis, which is partly founded in the bounded-rationality models of Herbert Simon et al. and in the evolutionary models designed in the Schumpeterian tradition. In the empirical tradition, ana- lysis of growth basically has three components: (i) growth of existing firms, (ii) persistence of growth (growth patterns) and (iii) variability of growth. The first topic, growth itself. is the most intensively studied. This is the topic that will be dealt with in this chapter as well. In the classic micro-economic approach, firms' decisions to grow (or to downsize) are explained by the quest for the minimum of the long run cost curve. This curve is assumed to be convex, either V-shaped (offering a single optimum) or L-shaped (in which case there is a threshold above which economies of scale can be realized). A factor restraining growth or decline is cost of (capital-)adjustment. Since technology has become cheaper, and thus easier accessible for small firms, the optimum -or minimum efficient scale level- has moved to the left: small firms are as viable as large firms, and this incentive for growth has lost its weight (cf. Carlsson (1989), Acs and Audretsch (1990». Another effect is that from the 3 In 2S percent of the 112 rums in the panel when: we interviewed the ectuaI owner, heI.he had bought the fmn. S8 percent of these ownen had established the fmn thcmee1vea, the othen had inherited it. 28 capital point of view, adjustment costs are smaller and therefore firms tend to be more flexible. This flexibility can be hampered by government regulations. Analysis of firm growth at the meso-level started in the fifties. Herbert Simon and other authors showed that from the shape of firm size distributions inferences about stochastic patterns of individual firm growth could be made. 4 If for instance growth is a random walk (Le. growth rates are randomly distributed), the limiting firm size distribution is lognor- mal. s Since most observed firm size distributions within industries are skewed (Iognormal- or Paretolike), the suggestion is that growth may be explained by a random walk process. If such is the case, no systematic influence from whatever variables exists. The particular conclusion of relative growth being independent of size is known as Gibrat's law. From the seventies on, the availability of microdata and computing power made it fea- sible to analyze growth at the micro-level. A severe attack on Gibrat's law was launched by Birch (1979). He shows that the growth of small firms was an extremely important contri- butor to the generation of new jobs in the us in the period 1969-1976. Although there has been criticism on Birch's findings, nowadays most economists agree on the importance of the birth and growth of small enterprises. To explain the relation between firm growth and firm age, Jovanovic (1982) developed a theoretical model, in which firms are able to learn. This model predicts an inverse relation- ship between age and growth. Unsurprisingly, this prediction is confirmed by most empirical work. Actually, this is about the only common finding across different authors. Evans (l987a,b) introduced the modern empirical approach. Since then a vast amount of economic and econometric research on the growth of individual firms has been done, mostly on manu- facturing firms. Wagner (1992) gives a general -though incomplete- survey of empirical results. He observes that explanatory variables are seldomly sophisticated. The analysis is often limited to size and age, because no other information is available. In the majority of empirical studies, Gibrat's law is rejected. Larger firms are found to have lower growth rates, and a smaller variability of growth. There is some evidence for persistence of growth, see for instance Singh and Whittington (1975). Acs and Audretsch (1990) relate small firm growth to technology and labour-related variables, such as unionization, innovation ratios, capital intensity. They conclude: "Small firm growth is negatively related to capital-intensity, advertisement intensity and the extent of unionization, (... ) and (. .. ) positively related to the extent of human capital and the amount of innovations in the industry." When using these findings as working hypotheses in the analysis of growth patterns of 4 See for instance Hart and Praia (1956). Simon and Bonini (1958). Jjiri and Simon (1964). Lucas (1978) gives an excellent general IUrvey. S Let SiS denote the lize of fum i at moment t and "'II a random number. then a random growth process implies If random poW'lh u nonnally distributed, than the limiting distribution U lognonnal: 29 firms in Zimbabwe or other developing countries a proviso is needed: so far all testing has been done on us or European datasets, which are not necessarily representative for develo- ping African economies. 4.3 . Growth of Zimbabwean firms Data on 201 Zimbabwean firms are available, sampled in a representative way among four manufacturing industries: foodprocessing, woodworking, textile and metalwork. The ques- tionnaire used, covers nearly all aspects of business, demographic aspects (legal status, ownership structure), sales, employment structure, market behaviour and (perceived) market structures. This level of detail, a major strength of the RPED-dataset, facilitates a 'Iook behind the veil' (Wagner (1992». A major drawback is that the detailed information is largely limited to the situation in 1993. Data on some variables (turnover, employment and age) are available for earlier years, but others of interest (ownership, exports etc.) are not. Causality in growth-explaining models requires information about the starting point of the growth process, while in the current analysis only information about the endpoint is available. In the analysis, we use as many time-invariant explanatory variables as possible. Before we start the analysis, some descriptive measures of size are presented, in order to show the context within which growth will be explained. Since growth is the first derivative of size, we have to make a choice for the measurement of size. Sales and employment are the logical candidates. 6 As noted earlier, the questionnaire focuses on - -,-...--...--...--...--...--...--...--...--...--...--, current (1993) information, but some historical data were gathered as well. Data on sales and employ- ment are available (in retrospect) for the years 1981, 1986, 1991, 1992 and 1993. 7 Especially for the . ...... - ..... . , smaller firms, historical data are based on recall so ... the results may not all be reliable. Employment proved to be easier to remember than sales. Of course, not all firms in the sample were in existence _ _ _ in 1981 or 1986. and not all respondents were able Figure 4.1: Consumer Price Index to provide the interviewers with the historical data. Nevertheless, data on employment and sales in the beginning, mid and end of the eighties are available for a panel consisting of 88 of the 201 firms in the sample. In order to use sales as a measurement for size, we have to correct for inflation, which was extremely high in the early nineties, as is shown in Figure 4.1. This makes sales an unstable measure, especially when one wants to correct for measurement error, as is explained in section 4.3. Figure 4.2 shows the development of average employment and sales (in ftxed prices) for the 88 fmns we have full historical information about. This subsample is on the average 6 In the analyais, employment is defUled as the number of employCCl at the time of the interview, excluding peak ICUOI'I and cuuaI worken. Due 10 the uncler-utilisation of production c:apecity (cf. chapler 7), IUch worltcn were found in only 15 of the 201 fumtl, 10 they do not disturb the picture. In lOme CIIIIeI, the number of worlccn It the end of the last fllCal year was given inItcad of thc current number, but .iDee for moat fumtl the fllCal year cnda in M.y or June, thcac fiprea coincide rather well with the current fipla. 7 1981 _ choacn as the fll'llt year after Independence. 1991 as (about) the fll'llt year after the beginning of the atnJctuml adjUlll.meDt program, 1986 as the year euetly in between. 30 larger then the entire sample. Firms that already existed in 1981, are now larger than firms that were established after 1981. The subsample of 88 firms is neither representative for the entire sample, nor for the population of Zimbabwean manufacturing firms. Yet the trend is clear: a steady expansion in employment from 1981 to 1991, decline afterwards. 160 Average sales shows a wave pattern, drop- 140 r ·1 ~ ··· o .. i· I"' · · · I l .. =:'II ! ping astern with the development of employ- -, 120 ~ t i I ment. This may imply that the firms in this : 100 == '=1 ~ I panel decreased in productivity, allowing some .0 .= slack capacity to enter the organization. After 60 1 1990 interpretation of the data is difficult. 60 Inflation is based on annual figures, while 20 most business data are based on figures of , broken fiscal years (ending in June/July). In f-= - 1"1 1916 I~'" times of high inflation, the process of deflating Figure 4.2: mean employment/sales sales figures is highly sensitive to seasonal patterns, which we have no information about. Employment on the other hand is not sensitive to inflation, which makes it easy to handle. A drawback is hidden in the restrictive labour regulations in Zimbabwe: there are governmental restrictions on hiring (of temporary labour) and firing, creating a lag between market developments and adaptations of size (i.e. growth/decline). In times of general growth, entrepreneurs will be reluctant to hire, and in times of general decline, it is difficult to fire or layoff workers. If there is slack capacity in the organization, it is impossible to observe 'real' growth until full capacity is reached. Taking all advantages and problems into account, employment appears to be the best choice as a measure for size. On top of that, full employment is an important objective of economic policy, which makes it more relevant from a policy point of view. Firm size distributions A widespread opinion about African econo- aos...------------------, mies says that firm size distributions differ from those in more developed economies. · .. While there are many small firms (as is :1ft cornmon), and some large firms as well, the middle of the distribution (firms between 50 and 500 employees) is thought to be missing. If such is the case, Gibrat's ..... law does not apply. A bimodal distribution, without a well-filled middle, is not lognor- ... mal and thus cannot be caused by a random walk. If the middle is missing, there must be factors influencing firm growth, either Figure 4.2: Firm Size Distribution '993 in the system, or in the firms or entrepre- neurs themselves. Figure 4.2 shows the firm size distribution of the 201 firms in the sample. As can be seen, there is an overrepresentation of large firms, a phenomenon that can partly be explained by the sampling procedure. Nevertheless we find no support for the theory of 'the missing middle': there is a significant proportion of firms between 50 and 500 employees. Figures 4.3 to 4.6 show the firm size distributions of the firms in the sample of the four 31 industries under study for 1993. Woodworking and metal show the common skewed distri- bution, but food and textile have a bimodal shape, with a relatively large fraction of (very) large firms. 8 This indicates that economies of scale are reached at different levels in the four sectors, and thus that sectoral differences are important. We may assume that Gibrat's law will be rejected. 11",-,- - - - - - - - - - - - - - - , 35"'=--------------- .,..1 ,) 15'" 111'4 10 IGO 110 100 210 100 110 _ 410 _ Figure 4.3: Firm Size Distribution 1993 Food Figure 4.4: Firm Size Distribution , 993 Wood In the first stage of the analysis, average annual growth is broken down by a number of variables, some of them indicating the business environment (e.g. sector, location), some of them being attributes of the fmn itself (e.g. age, size) or its owners (e.g. gender, race). Table 4.1 shows mean sizes and standard deviations for the various subsamp)es . H"',----------------~ .... ,.....--------------... ." I'''' ii 10 100 110 ______ _ 10100110100 _____ _ Figure 4.5: Firm Size Distribution 1993 Textile Figure 4.6: Firm Size Distribution 1 993 Metal Average firm size in 1993 of the firms in the sample is 289 employees (including working owners/managers). Most of the classifications in Table 4.1 are self-explanatory. The variable type offirm is used because there is empirical evidence that independent, stand· alone fums have lower growth rates (Variyam & Kraybill (1992». The firm type-classifica· tion we use here is quadrinomial. 9 Firms are classified as cooperatives, entrepreneurial 8 The.hape of thcac diltributiona is probably influenced by umpling bias. The textile industry under Iludy for inItancc is quite heterogenous, conaiIting of _VCl1l, prmont manufacturcra, cloth printem etc. If the long run cost curve hu a minimum (or a thrcIhold), it dOC8 not need to be the IIII'Qe for aU IUbaectorI. 9 A UlCful dichotomy for type of fum is 'Cllb'Cpn:neurial' w. 'planning'. Entrepn:neurial rums ani dermcd u rums that strongly depend on the expertiac and euthu.ium of the IOle owner. Such rums ani characterized u having no staff, management team etc. While _h rums ani usually 1RIIIll, it is not difficult to fmd examp1ct of (very) l.arJe ruma fulfilling this dClCription. Yet the correlation with .i:u is too l.arJe to UIC it in the explanation of growth when .i:u itIelf is UJed already. 32 firms (i.e. stand alone firms with a working owner), independent firms that are pan/member of a group of firms (mostly family owned chains, not necessarily in the same sector or industry) and subsidiaries of Zimbabwean stock exchange funds or multinationals. 10 Table 4.1: Mean emr:!lol;:ment 1993 number mean std.dev. sig All firms 201 289 649 by sector · food 49 342 557 · wood 26 104 98 · textile 89 369 859 · metal 37 155 224 by period of founding' 1 · before 1965 71 559 777 ··· · 1965-1980 56 257 765 · 1980-1990 49 64 77 ··· · atter 1990 25 35 60 ·· by firm type · cooperative 14 96 262 · entrepreneurial 106 77 117 ··· · part of group 35 452 475 · subsidiary 46 710 1153 ··· by gender owner · female 18 23 35 · · male 80 87 130 ··· by race ownerls) · black 56 52 206 ··· · european/asian 145 380 733 ··· 'sig' denotes significance of the F-statistic of an ANOVA·test of the given subsample vs. all other firms. Significance-levels: ··· 99%, ·· 95%, · 90% source: RPED·study Zimbabwe Table 4.1 proves there are important differences in average firm size among sectors and other subsamples. Woodworking is the smallest-sca1ed and the most homogenous of the four industries involved, while food and textile are the largest. There is a strong correlation between period of founding and firm size. The oldest firms are by far the largest. As expected, entrepreneurial firms are smaller than others. There is a clear distinction in terms of size between entrepreneurial frrms, firms that belong to (unrated) groups of firms and subsidiaries (either domestic or foreign owned). Within the group of firms with a working owner. there is a large difference between male and female owned firms. Female owned frrms, mainly to be found in the textile industry (cf. chapter 3) are much smaller. The race issue remains very important in the Zimbabwean economy: white or Asian-owned firms are on average more than seven times as large as 10 The few fUllll that had a Itock exchange rating and the .ingle pa.rastata.I WCI't' classified in the last category. 11 1965 (Unilateral Declatation of Indcpcndence by Ian Smith e ··. ), 1980 (Independence, Mr. Mugabe'. ZANU-PF winning the fll'lt general elections) and 1989/1990 (start of ESAP) are tuming points in Zimbabwe'. recent (economic) biItory. 33 black-owned firms. This is partly explained by the age of these firms: black entrepreneur- ship dates back to more recent years than white/asian entrepreneurship. The expectation is that black owned firms will catch up with the incumbent white/asian owned firms. 4.4 Realized growth 1981·1992 The years immediately following independence (1981-1986) were a booming period in terms of employment growth, at least for firms that managed to survive until 1993. The average annual growth rate for the firms in the sample during these years in 7.3%. The only sub- sample with zero-growth in this period is the small group (n= 12) of woodworking firms. Between 1986 and 1991 most of the firms continued growing, although the average growth rate is lower (5.2%). 1991-1993 shows a sharp decline of employment, with hardly any sub- samples as exceptions. The definition of size contains elements of measurement error: employment has a stock nature (current employment), and are based on the closing if the most recent jiscal year, which is not the same for all firms. Further, there is a random element in the timing of measurement. If a firm decides to grow or downsize from level A to level B, this is usually is not realized overnight. Growth needs a time path, and because measurement occurs at a certain fixed moment, this path cannot be viewed completely. A common way to handle this type of error is to eliminate the influence of this 'snapshot' -effect by defining size at period t (say S,) as the mean of the levels of employment at periods t-1, t and t + 1. Thus: (4.1) There is no possibility to transform the data in or around 1981 and 1986 in this way, but for 1991, 1992 and 1993 there is. In the analysis. the mean of the employment levels in 1991. 1992 and 1993 is used as the size indicator for 1992}2 The purpose of the analysis of firm growth is to determine which firms grow faster. and under which circumstances. From the information available in the dataset. the following firm attributes are chosen to be used in the analysis: size According to Gibrat's law size (measured at the begiMing of the period under analysis) should not discriminate, but in a lot of empirical research smaller growth rates for larger firms are found. age A common finding is that growth rates decline with age. sector Sector dummies are included since the firm size distributions suggest that growth patterns differ between sectors. location Location dummies are included distinguishing between Harare, Bulawayo and the rest of the country13. Harare is the country's economic centre, Bulawayo is the capital of Matabeleland. jirm type There is evidence in the literature that stand·alooe firms usually grow more slowly than subsidiaries. 12 Note that the effcc:ts of the drought cannot be ~n:d in thiJ _y. Since thiJ iuue u too important to be neglcctcd in the Zimbabwean economy, devel0pment8 between 1991 and 1993 ~ given special atten- tion in patagnlph 4.6. 13 MOlt of the rums in the other QretJ.S ~ located in the Midlands (Kwe Kwe, GweN, Kadoma). 34 ownership Firms with foreign majority ownership are supposed to have easier access to technology, credit, etc. and thus a firmer basis for growth. gender Female entrepreneurship is a key issue in a many development programs. The gender dummy distinguishes female entrepreneurs, not female managers. race In Table 4. I it appears that black firms are smaller than white/asian firms, which suggests that they may have lower growth rates as well. expons Exports require a certain scale (except to immediately surrounding countries, which are part of a firm's 'natural' market). Successful exports can act as a flywheel for a firm's development. In Table 4.2 a breakdown analysis of the average annual growth rates 1981·1986 and 1986·1992' is shown for the various subsamples defined by the variables listed above. 14 Two panels are distinguished: firms with information on both intervals (the 'pure' panel, n= 113, the five left columns), and all firms with information on 1986 and 1992' (the 'mixed' panel, n= 156, the three right columns). There is some redundancy in this table, but it allows for both full and 'unbiased' comparison. Table 4.2: Average annual growth rates 1981-1986-1992' 81·B6 sig 86-92' sig nl 86-92' sig n2 panel 0.073 0.030 1'3 0.044 15 6 by size 15 · small 0.164 ··· 0.047 19 0.070 24 · medium 0.085 0.042 40 0.071 ··· 56 · large 0.043 0.019 24 0.025 37 · very large 0.025 ··· 0.011 30 0.006 ··· 39 by sector · food 0.066 0.021 30 0.040 40 · wood 0.000 ·· 0.033 12 0.021 17 · textile 0.097 · 0.038 47 0.056 70 · metal 0.074 0.023 24 0.032 29 by location · harare 0.087 0.033 62 0.046 83 · bulawayo 0.062 0.032 35 0.046 48 · other areas 0.046 0.011 16 0.031 25 by period of founding · before 1965 0.051 ·· 0.023 57 0.022 ·· 64 · 1965-1980 · after 1980 by firm type 0.082 0.191 ... 0.030 0.064 48 7 0.037 0.088 ··· 53 38 · cooperative 0.049 0.016 2 0,096 9 · entrepreneurial 0.092 · 0.029 54 0.046 79 · part of group 0.088 0.058 ·· 25 0.063 30 · subsidiary 0.032 ·· 0.008 · 32 0.011 ·· 38 continued on next page 14 1992' denotes unweighed average fum lize in 1991, 1992 and 1993. IS As boundaries between arnall, mcdium-aizcd, large and very large furns, we take 10, tOO and 250 employees. 35 Table 4.2 continued: Average annual growth rates 1981-1986-1992' 81-86 sig 86-92' sig n1 86-92' sig n2 by gender · male 0.094 0.034 40 0.056 59 · female 0.136 0.004 8 0.012 13 by ownership · domestic owned 0.084 0.039 77 0.054 ·· , '5 · foreign owned · european/asian 0.054 0.067 0.008 0.030 · 34 96 0.011 0.041 -- 38 123 · african 0.112 0.029 17 0.053 33 by exports .' not exporting 0.091 0.010 36 0.032 62 · exporting 0.065 0.034 73 0.048 89 'sig' denotes significance of the F-statistic of an AN OVA-test of the given subsample vs. all other firms. Significance-levels: ··· 99%, _. 95%, · 90% source: RPED-study Zimbabwe The following results in Table 4.2 may be noted: · The common finding of an inverse relation between size and growth is confirmed, although for the second period the difference in growth rates between size classes is less marked. · Textile firms show the highest average growth rates in both periods. This may be an indication of a growing market. Given the deviant firm size distribution in this industry, with a high share of very large firms, a rapid process of concentration might have occurred as well. · Woodworking firms are the most stable subsample: this can partly be explained by the fact that average firm size in this industry is smaller than in the other industries: small firms are known to be less sensitive to economic conjunctural waves. · Firms in the two largest towns show higher growth rates than those in the rest of the country. · The common finding that young firms grow faster is confirmed. · The relation between average growth and fum type is remarkable. In the early eighties, entrepreneurial firms did very well. In the second half the cooperatives enter the stage (most of them starting very small-scaled), while the group-embedded firms show a steady growth. Subsidiaries, being the largest firms already. are relatively stable. · There is no clear relation between exports and growth. · Firms which are largely domesticalJy owned grow faster. · Female entrepreneurs did slightly better in 1981-1986, but from 1986 on they are out- performed by their male colleagues. · Black owned firms seem to do better, but there is a lot of interaction with firm size. The main conclusion that can be derived from Table 4.2 is that average growth rates -although positive for all subsamples- differ strongly among the various groups of firms. Sti1l one should be careful when drawing conclusions from such a bivariate analysis. In the next section the combined influences of all the explanatory variables are analyzed. 36 4.5 Factors determining growth To measure the influence of all the variables together, a growth relation is specified conform Evans (1987a, b). Let Sil be the size (Le. employment) of firm i in year r, and Ail the age of firm i in year r, then the basic equation is (4.2) in which G is a growth function and e a lognormally distributed error term. Evans specifies G as a second-order expansion of Log (Sil' Ail)' with possibilities for further expansion with other variables. Proportional average annual growth is the variable to be explained in equation 4.3: Log(Sit)-Log(Si,H) 2 :: U + Po Log(Si,,_t) + PI [Log(Sil-t)] + k The variables used as regressors Xu are those listed in Table 4.2. Binomial variables are treated as dummies, the other variables are either taken as continuous variables (size, age) or converted into dummy·systems. The dummy on entrepreneurial jirm is combined with gender: there is a dummy·variable for male owned entrepreneurial firms and one for female owned. For the 'pure' panel 1981-1992' annual growth 1981-86 is taken as an explanatory variable (without success). Other variables used are: · constant · log(size) and log(size) squared · log(age) and log(age) squared · log (age) ... log (size) · two dummies for entrepreneurial jirm (male and female) · dummy entrepreneurial jirm ... log(age) · dummy pan~J-group jirm · dummy subsidiary/stock exchange jirm · dummy black owned jirm · two dummies for location (Harare, Bulawayo, other areas) · three sectoral dummies (wood, food, metal, i.e. taking textiles as base) · dummy expons · dummy foreign majority · dummy expons ... dummy foreign majority For reasons of convenience, only the regressions with significant coefficients are shown in Table 4.3 (full tables are presented in the appendix to this chapter). As the specification fails to pass White's test on heteroscedasticity until a fourth order expansion of age (1986- 92') and size (1981-86-92') is incorporated, in the process of variable selection White's heteroscedasticity-consistent robust standard errors were used, and presented in Table 4.3. 37 Table 4.3: OLS-estimatss growth rates 1981-1986-1992' 1981-86 (n-l051 1986-92' In= 147) cosH T sig coeH T sig constant 0.125 3.84 ··· 0.348 3.72 size -0.028 -3.53 ··· -0.073 -3.00 ··· size 2 0.005 2.13 age age 2 -0.125 0.022 -2.47 2.14 .. ·· ··· member of group 0.049 2.34 ·· 0.056 3.39 male entrepreneurial 0.106 2.79 ··· female entrepreneurial 0.131 2.51 -0.053 -1.69 · entrepreneurial · age -0.027 -1.93 · other areas -0.035 -1.96 · -0.054 -2.32 ·· woodworking -0.105 -4.56 ··· exporting 0.087 3.02 ··· 0.094 5.20 ··· foreign owned 0.106 2.37 ·· 0.054 2.42 export .. for. owned -0.119 -2.65 ··· -0.088 -3.25 ··· mean dependent variable 0.075 0.040 adjusted R2 0.335 0.340 significance levels: ··· 99%, ·· 95%, .. 90% source: RPED-study Zimbabwe Table 4.2 indicated that smaller firms have higher growth rates. This result is confirmed in the regression analysis (Le. controlling for other variables): the coefficient of size is negative. It is remarkable that for 1986-1992' previous growth (as an indication of success- ful business in the early eighties) does not contribute to the explanation of growth. If there is such a thing as persistence of growth, it is not confIrmed by the current data; this is how- ever a topic where the sampling bias presumably causes problems: unsuccessful businesses that didn't survive are not represented in the sample, i.e. nearly all firms have grown, so growth as such cannot discriminate. Though the significant variables are not exactly the same for both periods, the system presented in Table 4.3 looks stable at face value. Only the dummy for female entrepreneurs has changed sign, from positive to negative. This should be a matter of concern for econo- mic policies: women tend to own smaller firms, with lesser growth perspectives. ~'s are reasonably high, given the number of firms. Tests on stability of the system do not lead to the conclusion that a structural change has occurred. An F-test on stability. applied on the regression as presented in Table 4.8 leads to an F-value of 1.008 (df=20, 212), which is far from significant. 16 There is no evidence for any kind of structural shift. There was less 16 Pooling of the observations on both intcrvalll lead, to a ReSlriCled Residual Sum of SqutJres (RRSS). while acparate estimation leads to two Unrf/striaed Residual Sums of SqutJrf/1 (URSS). Let "1 and "z be the number of observations in the two pcriodJ and 1 the number of parIIIIlCIen to be estimated. then the fraction (RRSS-URSS>Ik: URSSI(n1+ft2-21) has an F-dilltribution with 1'"1+";:-21: dcgrca of fftCdom. In this exercilc. the variable previous growth was om.iUed. 38 growth in the second half of the eighties, so there remains less to explain, but the general factors mfluencing firm growth seem to have stayed the same. What else can we learn from Table 4.3? · Gibrat's law is -unsurprisingly- rejected; growth rates are larger for smaller firms, though the relation is not linear. Age does not contribute to the explanation of growth in the early eighties, which is very remarkable. It means the settled incumbents grow at the same speed as younger, more recently established firms. For entrepreneurial firms however, the common finding of decreasing growth rates with firm age is found. In the second interval things seemed to have turned more or less to 'normal', though since there is no evidence for a structural shift, this 'normal' situation should be taken as a relative concept. · The sectoral dummy-variables have disappeared, except for the relatively large negative coefficient for woodworking firms in the interval 1981-1986. This means that sectoral differences are less important than expected from viewing Figures 4.3 to 4.6. o A firm's membership of a chain or other kind of inter-firm-relationship consistently proves to be a positive contribution to growth perspectives. As is shown in chapter 6, these firms are able to provide each other with (cheap) credit and cooperate in many other ways. o Entrepreneurial firms did well in the period 1981-86 (the coefficients for male and female entrepreneurs do not differ). In the second half of the eighties, female owned firms, being smaller already, have a negative sign. · Average growth rates are not affected by the race of the owners. Black firms are on average much smaller than white or asian owned firms, but they have grown at the same rate through the 1980's. As a result, the difference in size between black and white owned firms grows larger. o Both foreign ownership and exports have a positive impact on growth, though the combi- nation of the two removes more than half of the effect. It follows that firms with either an input or an output foreign orientation fmd a better soil for growth. The causality between exports and growth remains unclear: is not certain whether growth enhances exports or the other way round. The reason the subsidiaries-dummy doesn't work out perhaps l!es in some multi-collinearity with exports: 86.7% of the subsidiaries do export. Bivariate versus multivariate analysis There are differences between Table 4.2 and Table 4.3: in the bivariate analysis, signifi- cantly different growth rates were found for subsidiaries and firms in the textile industry. When controlled for firm size these single effects disappear (both subsidiaries and textile firms are on the average the largest firms within their own classification group), thus it can be concluded that size is the dominant reason for different growth rates. In the early eighties, the dummy for woodworking firms remains significant: size alone cannot explain For the 'pure' panel (lOS ftmll with data 011 1981, 1986 and 1992') the F-tcIt failB as well, but thiI does not add much insight. Since age is one or the expbuWory variablcs. the ftmll in thiI panel are not the IIUIIC anyway. A. the IIUIIC man cannot mvel twice down to the IIUIIC river (Heraclitus). the IIUIIC fum cannot ,0 twice through the IIUIIC growth 1tAge. 39 the lower growth rates of firms in this industry. A variable that did not work out in the bivariate analysis, but does work in the mul- tivariate analysis is group membership. Firms that are in some way part of a larger conglomerate consistently do grow faster: in a regulated economy like Zimbabwe, entrepre- neurs in such a group retain a flexibility and enjoy opportunities stand-alone entrepreneurs seem to lack. Exports were not significant in the analysis of variance, but when controlled for size and age, we see exporting firms grow faster; as indicated above, the causal relation is unci ear. The influence of a majority foreign ownership is remarkable. In the bivariate analysis, domestic owned firms show higher growth rates. In the combined analysis, the impact proves to be the other way round: foreign owned firms have a positive coefficient in the regression analysis. Conclusion Most of the notes above have one general factor in common. The worldwide recognized in- verse relationship between age and growth was absent in Zimbabwe in the years following independence. This implies the incumbent fums must have had a strong competitive advan- tage, either by knowing how to handle regulations better, or by fencing off the market by mutual agreements (or by doing both). Anyway they were able to keep newcomers at a distance. Perhaps this is a heritage of the rigidity of the economy during UOI and the Marxist oriented regime after independence. Firms were compelled to work together, and cooperation dominated comretition. This 'old boys' business culture is still very much alive among Zimbabwean firms. 7 And so, when the economy is not dynamic at all, the nature of firms is not expected to change over time, and there is no reason why older firms should grow slower then younger ones. As a result of this lack of competition, about 60% of the firms in the sample claim to be able to set prices as a mark up over costs, instead of taking the market price as given. Although this may be an overestimate (if you mark up to the level of the market price, marking up is merely an illusion), but it is illustrative. A significant number of respondents, most of them found in the food industry, openly declare that the prices of their products are the outcome of a regular consultation among the members of the shared cartel. 18 Another symptom of the lack of competition is the advantage enjoyed by firms embedded in an (informal) conglomerate, which have better growth rates than stand-alone firms, though this is not a typical Zimbabwean phenomenon. These firms are able to provide each other with cheap credit and other benefits as in kind loans, and thus avoid some of the burdens of competition. On one band these firms offer continuity to the economy, on the other hand they block innovation, and restrict the opportunities for new (black, female) entrepreneurship. 19 A structure in which informal circuits, managing to preserve the market for insiders, is not a favourable circumstance for the Zimbabwean economy. A lack of internal competition 17 As an owner of a textile fum IBid during the interview: ·We ICratch each others bacia! around here.' 18 Sec Chapter 10. wbcrc more lII1Cntion g paid to competitive IltrUCWrcs in Zimbabwe. 19 As we will ICC in ICCtion 4.5. it g the mIAll. stand-alone flmlll lhat auffer most from the cc:onomic: criag, caulCd by drou,ht and the ItNctunU adjultmc:nt proJl1l.m. This g probably another cffcc:t of closed shop-nature of Zimbabwean buaincu. 40 is not a cond ition to enhance efficiency, innovation and external competitive strength. Yet adaptation to a more flexible and competitive economic system, a stated goal of ESAP, may take some time. Employment versus sales Earlier in this chapter, the choice for employment as a measure for size was weighed against sales. For completeness, an exercise as described above was carried out with sales (in constant prices) instead of employment. The results are similar, but less pronounced: there are fewer observations (smaller firms were able to recall employment, not sales), and the explanatory power of the regression is lower, as indicated both by the R1 and by the significance of the individual variables. 4.6 Effects of the economic crisis The introduction of the World Bank's Structural Adjustment Program in Zimbabwe was fol- lowed almost immediately by a severe economic crisis, caused by a drought in the summer °. of 19922 Both effects are supposed to have had a (temporary) negative impact on the economy. Although it is not possible to identify which effect has the largest impact, it is important to determine which firms (operating under which circumstances) show the best resilience in this crisis. If there are firms doing well, we have found a strength and a possible success factor for the Zimbabwean economy. Earlier inthis chapter, the develop- ment 1991-1993 has been averaged in order to remove ambiguity and to introduce a more stable base for measurement. Here, the dat are separated again, allowing some error into the measurement, but also providing for more recent insitts. We hope to gather data in the winter of 19942 in order to get a better base for the analysis of the effects of this crisis. Anticipating on the availability of those data, some attention is paid to the developments between 1991 and 1993. Here we have difficulties in coping with the problems of measurement: adjustment in size to worsening market circum- stances is usually lagged (especially for SME'S and in a regulated economy even more), and the full effect cannot be seen yet. Nevertheless, because of the importance of the subject, we will give a brief sketch of the fmdings. Employment data on 1991 and 1993 are available for nearly the entire sample. Total employment fell with more than 10 percent in these two years (from 63,852 employees to 57,532 for the 188 firms in the subsample). Especially the textile industry (-11.1 %) and the foodprocessing industry (-9.5 %) experienced a massive decline, although woodworking fIrms (-2.2%) and metalworking (-6.6%) received their share as well. Table 4.4 shows average growth/decline rates for the various subsamples. As can easily be seen, decline dominates growth. Average growth rates of the four industries differ from those mentioned above on total employment. In the woodworking and metalworking indus- tries, unweighed average growth rates are positive, while total employment decreased. The explanation lies in the fact that decline is not equally distributed among firms. As can be seen in Table 4.4. it is especially large firms that have downsized during the last two years. The positive average growth rates for woodworking and metalworking firms in Table 4.4 are the result of the averaging of small positive or zero-growth rates of small firms and large negative rates for (fewer) large firms. The unweigbed average of firm's growth rates is positive, the growth rate of the average is negative. Because in this chapter the unit of 20 41 analysis is the individual firm, at the micro level, the figures in Table 4.4 are presented as they' are. Given size, woodworking and metalworking firms, the former siowlier growing in a booming period. on the average are able to sustain size in times of slump. Table 4.4: Average growth rates 1991·1993 n maan sig n mean sig panel 188 -0.009 size · small 38 0.159 ... type of firm · cooperatIve 101 13 0.033 0.005 · medium 56 ·0.015 · entrepreneurial · large 51 -0.062 · part of group 34 -0.017 · very large 48 ·0.074 ·· · subsidiary 45 -0.042 sector gender · food 46 -0.033 · male 76 -0.004 · wood 23 0.009 · female 17 0.107 · · textile 87 -0.004 ownership · metal 37 0.002 · domestic 149 0.007 location · foreign 41 -0.057 · harare 107 0.007 · european/asian 140 -0.029 · · bulawayo 54 -0.035 · african 53 0.049 · · other areas 32 -0.011 exports period of founding · not exporting 87 0.014 -0.026 101 -0.031 · before 1965 · 1965-1980 " 55 -0.039 · exporting · 1980-1990 47 -0.018 · 1990-1991 20 0.168 ··· 'sig' denotes significance of the F-statistic of an ANOvA-test of the given subsample vs. all other firms. Significance-levels: ··· 99%, ·· 95%, · 90% source: RPED-study Zimbabwe Table 4.4 shows, it is just the very small, informal type of firm that seems to realize positive growth: very young firms, cooperatives, very small firms (in 1991 less than 10 employees) female owned firms, black owned firms. A lot of these categories of firms overlap each other. The actual number of growing firms is relatively small. Analogous to the exercise presented in section 4.5, the combined effects of all variables are analyzed by means of regression. In Table 4.5 the results of an ordinary least squares regression are given, limited to significant coefficients. As in Table 4.3, T-values are based on White's heteroscedasticity-consistent standard errors. Table 4.5: OLS results growth 1991-1993 (n-1SS) coeff t sig constant 0.902 5.19 ··· size ...Q.278 -4.81 ·· · size 2 0.020 4.06 ··· age -0.151 -2.24 ·· age 2 0.031 2.25 ·· dummy entrepreneurial -0.071 -1.75 · black owned -0.176 -2.43 ·· exports 0 . 139 2.88 ··· mean dependent variable -0.010 adjusted R2 0.305 significance levels: ··· 99%, ·· 95%. · 90% 42 source: RPED-study Zimbabwe Except for the common findings for size and age (the crisis does not affect these common influence patterns), Table 4.5 shows that it is mainly the small, stand-alone firms that suf- fered from the crisis. Entrepreneurial firms, not being members of groups or subsidiaries have no network to fall back on. Especially the decline rates of black owned firms are stri· king. Exports prove to be a means to maintain a market position when domestic demand declines. The dummies on sectors and locations do not have significant coefficients. Since large firms are usually more flexible, size dominates the explanation of decline rates. Apparently, there is no evidence that, when controlled for firm characteristics such as siZe and owner- ship, the South (i.e firms in or near Bulawayo) or firms depending directly on a successful harvest (food, cotton) have suffered more from the drought. 4.7 Obstacles to growth From the above it can be concluded that there are considerable differences in growth rates among Zimbabwean firms. The general decline of firms during the last two years can be explained by the crisis in general. Yet one may wonder whether factors as regulations and market imperfections (labour, credit) are obstacles as well. In the survey, firms were asked to rank 15 potential obstacles to expansion in terms of their importance to the firm. The most important of these are lack of credit (and its price), lack of demand (which coincides with slump in general), quality of infrastructure and foreign exchange controls. The responses were measured on a five-point scale, varying between 'no obstacle at all' and 'very severe obstacle'. To determine whether the perceived obstacles are related to realized growth or decline and other firm characteristics, the fifteen potential obstacles were clustered in three scale variables: · regulations (ownership, labour, licenses, foreign exchange, location); · market conditions (lack of demand, competition from imports, quality of infrastructure and business suppon services); < · financial matters (credit, taxes, prices for public utilities). The scale variables were constructed by summing the individual scores and standardizing them by dividing the sum variable by the (theoretical) maximum. As a result all three variables vary between zero and one. Table 4.6 shows that the regulation scale is perceived to be the least hampering. The burden of market conditions and financial restrictions are perceived as about equal. Table 4.6: Perceived obstacles to growth mean atd.dev. nlzero) regulations O. '61 0.145 22% market conditions 0.300 0.171 7% financial restrictions 0.297 0.197 12% source: RPED-study Zimbabwe There are two reasons for not using these variables in the explanation of growth: (i) in the survey the current perception of obstacles was measured, while growth is measured in retrospect, with the current size as endpoint, (ii) since the perception of obstacles is a 43 subjective matter. it is impossible to identify whether there are real barriers, or that the entrepreneurs and managers have inadequate information, and perceive obstacles because of lack of good insight. Because of these identification problems, we try to determine which factors cause the perception of obstacles. We do this by regressing the three constructed scales on the common variables used earlier in this chapter as regressors. Because of the high share of zeroes, the effect of firm characteristics and realized growth is analyzed by means of a Tobit analysis. 21 Table 4.7 shows the results of the Tobit-estimates. Table 4.7: Tobit-estimates of perceived obstacles to growth regulations market financial c08ff T sig cosff T sig coeff T sig constant 0.233 5.26 ··· 0.273 6.06 ··· 0.185 3.78 ··· black owned -0.180 -4.61 ··· subsidiary 0.052 1.72 exporting 0.091 3.03 ··· bulawayo 0.078 2.32 ·· size 0.029 2.97 ··· 0.042 4.11 ··· previous growth 0.323 2.26 ·· age -0.042 -2.84 ··· -0.043 -2.55 ··· -0.038 -1.95 · sigma 2 0.024 7.'1 ··· 0.028 8.38 ··· 0.043 9.00 ··· Significance levels: ··· 99%, .... 95%, .. 90% source: RPED-study Zimbabwe Some remarkable conclusions that can be drawn from Table 4.7: · Contrary to the lack of association between growth and age, obstacles become less important when finns grow older. This is consistent with Jovanovic' model of learning effects over time. · Market and fmancial obstacles become more important when finns grow larger. This supports Mead (1993) conclusion of barriers between the informal and formal sector. · The same applies to realized growth in the period 1986-1992' (which correlates with size in 1992') of course. A test was carried out to see whether size-adjustment between 1991 and 1993 adds to the explanation: it does not. · Black owned finns feel less hampered by regulations, also when controlled for flI1Jl size. .. Though foreign exchange controls have been relaxed, exporting flI1JlS experience more obstacles from (other) kinds of regulations. · Subsidiaries (who have a need for transferring profits. in some cases to overseas) experience regulations to be larger obstacles as well. · The phenomenon of Bulawayo-firms having more problems with capital is difficult to ex- plain. There appear to be no economic reasons for it. Perhaps some explanation can be found in the organization of Zimbabwean banks: all head-offices are located in Harare, Zimbabwe's financial centre, and difficult to access for Matabeleland-based flI1JlS. 21 Actually the distribution ia tJuncatocl on both .idea (the 8CIIlod variablOl can only vary bctwcm 0 and 1). but the upper bound ia nevCl' bindin,g: there ia DO ainglc obIervation with the value 1 on either of the three lCIIc--variabIOl. 44 4.8 Concluding remarks In this chapter, we have made an attempt to explain growth and decline of Zimbabwean firms, in the context of its recent history (with UOI, planned economy and structural adjustment as important stages, and a severe drought in 1992/93 is an interfering factor). The results coincide partly with earlier empirical findings in other countries and economies: growth rates decrease with firm size, which leads to yet another rejection of Gibrat's law. Except for the small-scaled woodworking industry, sectoral differences appear not to be very important. A remarkable finding is that the racial issue plays no significant role as a determinant of firm growth: black owned firms are much smaller then white and asian firms, but they grew at the same speoo during the 1980's. As a result, black and white firms diverge in size. In the economic crisis that set in 1991, black ownoo firms appear to be less resilient. Firms with a foreign linkage, either on the input side (ownership), or on the output side (exports) show better growth rates than purely domestic firms: the home market is less able to sustain firms in their development. A remarkable phenomenon in the case of Zimbabwean firms is a weak (or in some cases even absent) relationship between firm age and firm growth. This may be explained by the country's regulated structure, in which entry barriers to the informal sector are low, but barriers on growth are high, and for the majority of small firms impossible to cross. The system of regulations is beneficiary to the incumbent firms, who have a firm grip on the market, while new, innovating small firms have difficulties in entering the market. Firms embedded in an (informal) conglomerate significantly grew better during the eighties. The drought caused most problems for small, informal (mostly black owned firms), but it did not influence the general patterns in the Zimbabwean context. The relation between perceived obstacles to expansion and realized growth can be analyzed only by means of data that have to be gathered yet. For the moment, we have shown that regulations are less of an obstacle than market conditions and financial struc- tures. Younger firms experience more obstacles (indicating a learning effect), as larger firms do. As activities become more large-scaled, more formal barriers have to be met. 45 Appendix: full estimation results Table 4.8: Full estimation results 1981-86 (105) 1986-92' 1105) 1986-92' 1147) cosff T coeff T coeff T constant 0.306 4.37 0.119 0.88 0.419 3.16 0.046 0.23 size -0.083 -2.47 -0.040 -1.15 -0.091 -2.53 size 2 0.006 1.50 0.007 2.84 0.007 2.18 age -0.037 -0.87 -0.037 -0.65 -0.138 -2.56 age 2 0.007 0.74 0.017 1.99 0.024 2.23 size · age 0.000 0.02 -0.015 -1.38 -0.002 -0.19 D-member of group 0.039 0.99 0.113 2.73 0.082 2.05 D-subsidiary -0.010 -0.24 0.073 1.73 0.039 0.93 D-male entrepreneurial 0.080 2.09 -0.014 -0.61 -0.020 -0.66 D-female entrepreneurial 0.078 1.23 -0.063 -1.45 -0.080 -2.13 D-entrepreneurial · D-age ·0.018 -1.17 0.022 1.55 0.016 1.15 D-black owned -0.069 -1.37 -0.023 -0.72 -0.041 -1 .1 1 D-bulawayo ·0.021 -1.16 -0.009 -0.56 -0.000 -0.03 D-other areas -0.054 -2.48 -0.028 -1.66 -0.048 -2.37 D-woodworking -0.085 -3.61 -0.016 -0.82 -0.027 -1.30 D-foodprocessing 0.000 -0.02 -0.013 -0.75 -0.010 -0.60 D-metal 0.006 0.21 -0.072 -0.37 -0.003 -0.18 D-exporting 0.080 2.54 0.076 3.94 0.092 4.82 D-foreign owned 0.109 2.86 0.034 1,26 0.042 1.61 D-exporting · D-for. owned -0.125 -3.02 -0.077 -2.43 -0.087 -2.87 Mean dependent variable 0.075 0.025 0.040 Adjusted R2 0.316 0.224 0.328 Variables names beginning with '0-' denote dummy-variables. T-values are White's heteroscedasticity-consistent standard errors. source: RPED-study Zimbabwe In the second regression (with previous growth as explanatory variable) sizel981 was used, because of the natura) correlation between growthl981_86 and sizel986' 46 References Acs, Z.J. and D.B. Audretsch, 1990. 'The determinants of small-firm growth in US manufactur- ing" Applied Economics 22: 143-153. Birch, D., 1979, The Job Generation Process, Cambridge MA: Centre for the Study of Neighbour- hood and Regional Change, Massachusetts Institute of Technology Carlsson, B., 1989, 'The Evolution of Manufacturing Technology and its Impact on Industrial Structure: An International Study', Small Business Economics 1: 21-37. Evans, D.S., 1987A, 'The Relationship between Firm Growth, Size. and Age: Estimates for 100 Manufacturing Industries', The Journal of Industrial Economics 25: 567-581. Evans. D.S., 1987B. 'Tests of Alternative Theories of Firm Growth" Journal of Political Economy 95(4): 657-674. Hart, P.E. and S.J. Prais, 1956, 'The Analysis of Business Concentration: a Statistical Approach', Journal of the Royal Statistical Society (series A) 119: 150-191. Ijiri Y. and H.A. Simon, 1964, 'Business Firm Growth and Size', The American Economic Review 54: 77-89. Jovanovic, B. t 1982, 'Selection and evolution of industry', Econometrica SO: 649-670. Lucas, R.E., 1978, 'On the size distribution of business firms', The Bell Journal of Economics 9: 508-523. Mead, D.C., H.O. Mukwenha and L. Reed, 1993, Growth and Transformation among Smail EnterpriSes in Zimbabwe, working paper University of Zimbabwe/GEMINI. Scbumpeter, I.A., 1950, Capitalism. Socialism end Democracy (third ed.), New York: Harper and Row. Simon, H.A. and C.P. Bonini, 1958, 'The Size Distribution of Business Firms', The American Economic Review 48: 607-617. Singb, A. and G. Whittington, 1975, 'The Size and Growth of Firms'. Review of Economic Studies 52: 15-26. Variyam, J.N. and D.S. Kraybill, 1992. 'Empirical evidence on determinants of firm growth', Economics Letters 38: 31-36. Wagner, J., 1992, 'Firm Size, Firm Growth, and Persistence of Chance: Testing Gibrat's Law with Establishment Data from Lower Saxony, 1978-1989" SmaIl Business Economics 4: 125-131. 47 5 Indigenous and smaU scale enterprises Clever Mumbengegwi 5.1 Introduction Development of indigenous (to mean black owned) enterprises has become a burning policy issue in Zimbabwe since the late 1980's. The relevance of this as a policy issue can only be appreciated against the following background. The colonial history of Zimbabwe (1890 -1980) created barriers to indigenous participation in the economy, especially in the urban manufacturing sector. This resulted in the present stark inequality between blacks and whites both in terms of ownership of resources and in the making of decisions that influence the pace and direction of the economy. The socialist policies of the government during the first decade of independence (1980-90) were not very conducive to the emergence of a black indigenous capitalist class. Pervasive controls and regulations, initially enacted by colonial governments to·protect white capital from potential indigenous competitors were embraced as aspects of socialist economic management. This controlled policy environment had an unintended bias towards existing large scale, white owned firms. Emergent, small and indigenous enterprises found themselves at the bottom of the queue in accessing government controlled resources such as foreign exchange, imported inputs and raw materials, capital goods, credit and business support services. Perhaps the most significant recent development fuelling the desire for indigenous enterprise development was the government's decision to adopt the Economic Structural Adjustment Programme ESAP in 1991. Although this programme primarily aims at improv- ing macroeconomic efficiency through market based reforms, the fact is that the political economy of market reform cannot divorce itself from distributional issues. With indigenous blacks historically marginalised, trepidation exists among indigenous black entrepreneurs and policy makers that the anticipated benefits of economic reform might just accrue to the established white firms in the same manner that the controlled policy regime of 1980-1990 had inadvertently not benefitted them. Thus, extreme inequality in the distribution of physical, financial and human resources between black and non black Zimbabweans has generated social and political tensions that make indigenisation a legitimate policy concern. Although this concern lies more in the realm of social policy than pure economic analysis, . economic analysis can inform on the design of social policy. 5.2 Objectives and hypotheses This chapter has two objectives. The fIrst is to give a brief description of the key character· istics of indigenous enterprises. These characteristics retlect the depth and breadth of indigenous entrepreneurial capacity. The premise underlying market based reforms is the existence of entrepreneurs with the ability to conceive and develop an idea into a viable business venture. The World Bank has identified lack of capacity as one of the major constraints to Africa's development efforts. African capacity building bas become one of its priorities in donor assistance programmes. If the recent policy concern with indigenous enterprise development is to succeed, indigenous entrepreneurial capacity building becomes particularly relevant. The second objective is to analyze the key constraints facing indigenous entrepreneurs in their attempts to meaningfully participate in the economy. The historical poverty and inequality noted above suggests that access to finance and credit for investment might be the 48 most important constraint. This is confirmed by indigenous firms' responses to our questionnaire. They cite lack of credit as the most severe constraint to enterprise expansion. Entrepre- neurial capacity, credit and liquid ity constraints are important issues for policy intervention because they constrain a firm's ability to take advantage of profitable opportunities, inhibit its ability to either venture into business or expand into new lines of production. Indigenous firms are small because their owners are too poor to raise the initial capital requirements to finance large projects from their own resources and because they lack the capacity to run a large scale business operation. Thus, our analysis of indigenous entrepreneurial capacity and the severity of finance, credit and liquidity constraints might inform on appropriate forms of policy intervention. 5.3 Characteristics or indigenous enterprises As noted above indigenous firm characteristics are examined as a proxy for entrepreneurial capacity, of which the more qualitative dimensions are intrinsically difficult to measure. However, there is a priori reason to believe that entrepreneurial capacity is a function of the following key firm characteristics: the owner's educational and skills level, previous business experience, f1I1D size, firm age and the ability to enter into business linkages with established large scale firms. The last characteristic can also be a measure of the degree of enterprise integration in the broader economy. Education and previous business experience Table 1 shows the distribution of entrepreneurs by formal educational status. The level of formal education of indigenous entrepreneurs is much lower than that of other ethnic groups. A large proportion (42.6%) possess no more than functional literacy. have gone only as far as primary school. As compared to entrepreneurs in other ethnic categories, with 90 percent possessing secondary education or better, only 52 per cent of indigenous entrepreneurs have this achievement. About 6 per cent have had no formal education and only 3.7 % have university education compared to zero and 41 per cent of all other firms respectively. Table 5.1: Indigenous entrepreneurs: formal education none prinwy .ec univ. indigenous count 3 23 26 2 percent 5.6 42.6 48.2 3.7 other count 0 8 39 33 percent 0 10 48.8 41.2 As shown in Table 5.2 very few, 4 or 15.38%, indigenous entrepreneurs had ever owned another business prior to establishing their current one. Thus the low educational levels and lack of business experience suggests severe lack of capacity especially to enter into the more technologically sophisticated lines of production. 49 Table 5.2: Previous businesses owned indigenous Firms non-indigenous Firms yes no yes no Count 4 22 17 39 Percent 15.4 84.6 30.4 69.6 26obs. 56 obs. Firm age The typical indigenously owned firm is relatively new and young. The majority (72 per cent) were started after independence. This contrasts with only 24 per cent of all other firms started after independence. The average age since start up (as at June 1993) was 10.4 years in contrast with 28.5 years for all other firms. The oldest indigenous firm is 45 years compared to 90 years for all other firms. This information is summarised in Table 5.3. Thus, the accumulated business experience is rather limited. It is often argued that successful entrepreneurial capacity development depends on the presence of a middle aged group of enterprises from whom new entrepreneurs can acquire sk.ills. This missing middle is particularly apparent in the indigenous business sector. Table 5.3: Firm formation by ethnic origin periOd of formation indigenous firms no % non indigenous firms no % % of tot81 after 1989 10 (56) 8 (44) (91 1980-1989 31 (55) 261551 (28) 1966-1979 13 (26) 441741 (26) before 1965 (4) 64 (961 (381 Firm size Details of firm size by employment and ethnic origin are summarized in Table 5.4. The majority (about 60%) of indigenous firms are very small with 10 employees or less compared to about 5 percent for all other firms. Another one third can be described as medium scale with only 7 per cent being either large or very large. The skewed distribution of firm size informs on the potential for capacity building. For small indigenous enterprises to grow, there is need for a corresponding set of efficient large scale indigenous enterprises with whom to network and interact. Evidence presented below shows the difficulty indigen- ous firms face in networking with the established white owned enterprises. Table 5.4: Firm size by ethnic origin iI1iPIoyment up to 10 11·100 101·250 over 250 Indigenous 34 19 2 2 (59.6%) (33.3%) 13.3.%) (3.5%1 Other Firms 6 44 35 31 (5.2%) (37.9%1 (30.2%1 (26.7%) Legal status and ownership structure The most common form of legal ownership among indigenous firms is sole proprietorships (4" .! %). Another 21 per cent are cooperatives. Although our survey did not explicitly solicit information about whether the firm was in the formal or informal sector, a large 50 proportion of these are known to be unregistered informal sector firms, Limited liability companies account for only 28 per cent of the sample. These are firms that have made a direct entry into the formal sector as opposed to the small firms that start in the informal sector. The wisdom in the literature is that the latter are expected to graduate and enter the formal sector as medium to large scale enterprises, However, evidence from elsewhere suggests that few firms that start very small ever graduate to become medium to large scale enterprises (see Liedholm(1993». The reason might be lack of entrepreneurial capacity to grow rather than lack of profitable opportunities, Table 5.5: Legel status of indigenous firms sole prop partnership limited liability coop. Count 27 2 16 12 Percent 47,37 3,51 28.07 21,05 Business linkages Business linkages not only reflect the degree of market integration and interaction but could be a source of skills acquisition for small and indigenous firms. Integration of indigenous small firms into the broader economy is hypothesised as an important vehicle to their growth. Business linkages facilitate this process through enhanced efficiency from specialisation and the entrepreneurial capacity building that it engenders. Since small firms are often considered to be the seed-bed of future large scale enterprises. linkages and networking might facilitate the emergence of such indigenous large scale enterprises through the process of graduation. However. there is little evidence of this occurring within Zimbabwe's manufacturing sector. Our main findings are that enterprise activities that can be. described as networking and business linkages are very limited especially for indigenous enterprises. The matrix of such activities allowed in the questionnaire included subcontracting, in kind borrowing and lending (of equipment and raw materials) supplier credit, and customer advance payment for inputs, goods and services. Subcontracting Only 14 enterprises were involved in a business relationship with another firm which could be described as subcontracting. Of these only two were indigenously owned firms in the food and textile sectors. Both were large rather than small. Thus there is a virtual absence of subcontracting between small indigenous firms in our sample and established white busi- nesses. Most subcontracting activities are between firms whose owners belong to the same racial group. There is not a single case of cross ethnic subcontracting. Contrary to the subcontracting literature, subcontracting is limited to a few large white owned firms subcontracting among themselves. Thus the notion of developing indigenous capacity through linkages with large firms is a goal still far from realisation. The length of this subcontracting relationship ranges from 1 to 36 years with a mean of 13 years for non indigenous firms and 1 to 10 years with a mean of 6 years for indigenous firms. For subcontracting the food sector (6 firms) was more likely to be involved than other sectors. The rest of the subcontracting activities were evenly divided among the remaining 3 sectors. 51 In kind lending and borrowing Raw material and equipment exchanges is another form of business linkage especially in a shortage economy as existed in Zimbabwe prior to the adoption of the economic reform pro· gramme. There were 31 reported cases of this practice which again was more prevalent among established white firms than indigenous enterprises. Thus, in summary, subcontracting in Zimbabwe's manufacturing sector is mainly between large and well established enterprises. Small indigenous businesses are hardly involved. Business linkages and networking is more common between firms that belong to the same group of companies. These results are consisted with what is already known (see Mead and Kunjeku (1992) and Zwizwai and Powell (1991». Zwizwai and Powell concluded that although "some linkages are currently at work joining large white owned with small black owned businesses; these are the exception than the rule." With this assessment we agree. Credit constraints Financial and credit constraints might explain why most indigenous enterprises are so small and why few are likely to grow into large scale enterprises. Table 5 shows the firm responses on the severity of lack of credit as a constraint for expansion. On a scale of 1 to 5 (with 1 indicating "no problem" and 5 a serious problem) 77 per cent (43 firms) of indigenous firms ranked credit to be an above average problem; with 55.4 per cent (31 firms) indicating that it is the most severe constraint to firm expansion. In contrast, just over half (about 52 per cent) of all other firms ranked credit above average with only 29.3 per cent (41 firms) indicating that it is the most severe constraint. Furthermore, 42 per cent of the latter did not consider credit to be a constraint as compared to 21.4 per cent of indigen- ous firms. Ranking the severity of the various possible constraints by the mean response, lack of credit is ranked as the number one constraint by indigenous entrepreneurs out of 16 possible responses listed in the questionnaire. In contrast, it is only ranked 5th by all other firms. The seven biggest constraints to firm expansion for indigenous firms are listed in Table 5.7 together with their mean ranking by all other firms. Table 5.6: Firm responses on credit as en obstacle expansion severity of the cridlt constraint firm category 1 2 3 4 5 indigenous count 12 1 7 5 31 percent 21.4 1.79 12.5 8.93 55.4 other firms count 59 8 12 20 4 percent 42.1 5.7 8.6 14.3 29.3 Table 5.7: The seven most serious constraints to firm expension ObstaCle indigenous firm other firms lack of credit mean 3.75 , rank mean 2.82 renk 5 bus. support services 3.16 2 2.08 10 lack of infrastr. 2.66 3 2.88 4 other problems 2.54 4 3.67 1 utility prices 2.41 5 2.75 6 no demand 2.23 6 3.16 2 forex controls 2.01 7 2.99 3 52 In addition to the evidence presented in Chapter 6 on how firms finance their initial and subsequent investments, the above suggests that indigenous firms face more serious credit constraints than other enterprises. This motivates further analysis as to why this is the case. 5.4 Conclusion The conclusion from the forgoing discussion is that the typical small scale enterprise is more likely to be owned by an indigenous black Zimbabwean who started in business after independence (1980). It is more likely to have started with a very narrow and shallow physical, financial and human resource base. The business is constrained to grow for lack of adequate financial resources. It finds it difficult to enter into networks and/or business linkages with established, white owned, large scale enterprises for lack of capacity, experience and/or business reputation. The co-existence of credit constraints in the formal market and virtual absence of informal credit markets might appear something of a paradox. In the Zimbabwean case, there is a perfectly reasonable explanation for this. Informal lenders face one major disadvantage. They rely mainly on their own limited savings or borrowing, to run the lending operations. The historical deprivation and current poverty of indigenous Africans might be such that few have built sufficient financial reserves to take up the informal lender role. This may have severely constrained the emergence of informal credit markets in Zimbabwe. Our findings suggest policy implications and dimensions that are not often emphasized in programmes of market based economic reform. While the Iiberalisation of interest rates, bank deregulation and elimination of government intervention are valid concerns for fmandal sector reform, they tend to neglect the broad institutional issues of credit provision. Perfect markets are not an infallible law of nature. Market based reforms, though essential, cannot guarantee access to fmance by indigenous firms. Programmes designed to reduce the information problems and asymmetries intrinsic to the nature of the credit transaction are as important to the reform process as the freeing of financial markets from excessive govern- ment regulation and controL The notion that the informal sector is complementary to the formal sector and hence needs encouragement does not seem to be a viable policy prescrip- tion in the Zimbabwean context. at least in the foreseeable future. Entrepreneurial capacity building is again an issue that market based reforms often take for granted. The presumption is that domestic deregulation and trade liberalisation will (by the invisible hand) promote indigenous participation in the economy. In a time dimension, the process may be so slow as to question the credibility of the reform process especially in an environment of acute racial inequality as exists in Zimbabwe. The lack of entrepreneurial capacity, the absence of a middle aged cadreship of indigenous entrepreneurs and the very limited existence of business linkages and networking between indigenous and white owned large scale firms suggests the need for more direct forms of policy intervention to comple- ment market reforms. Affirmative action for indigenous fmns might be needed not only to allay the trepidations mentioned earlier but also as a legitimate attempt to redress the market faiJuresand imperfections whose greatest incidence is on indigenous enterprises and entrepreneurs. Such a programme is already in place for the construction sector and needs to be expanded as a vital aspect of the economic reform process. 53 References Dessing, M., 1990, Support for Microenterprises: Lessons for Sub-Saharan Africa, World Bank Technical Paper (122) Floro, S. and P. Yotopoulos, 1991, Infonnal Credit Markets and the New Institutional Economics, Westview Press, Boulder Hansohm, D., 1991, Small Industry Development in Africa - Lessons from Sudan Helmsing, A.H.I. (00.), 1993, Small Enterprises and Changing Policies, IT Publications, Exeter LiOObolm, C., Small and Microenterprise Dynamics and the Evolving Role of Finance. In Heimsing, 1993, : 261-77. McPherson, M.A., 1992, Micro and Small Enterprises in Zimbabwe; Results of a Country-Wide Survey, GEMINI/MSU Mead, D. and P. Kunjeku, 1992, Business Linkages and Enterprise Development in Zimbabwe, mimeo, Harare Mumbengegwi, C., Structural Adjustment and Small Scale Enterprise Development in Zimbabwe. In Helmsing, 1993: 144-159 Uhlig, H., april 1993, Transition and Financial Collapse, mimeo, Princeton University ZimConsult/uNIDO, october 1992, Support to Small-Scale Industries and the Enhancement of Indigenous Ownership in Zimbabwe. UNIDO report Zwizwai, B. and 1. Powell, 1991, Small Scale Metal WorkinglLight Engineering Industries in Zimbabwe: A Sub Sector Study, mimeo 54 6 The fmance of investment by flrIlls Clever Mumbengegwi Jan ter Wengel In recent years the relationship between finance and growth has again captured the attention of economists at the micro· and macro·levels. The purpose of this study is to investigate the functioning of credit markets from the users' side: private, foreign, state and joint enter· prises. The empirical analysis is carried out on the basis of data provided by the interview of 201 firms in Zimbabwe. The ultimate goal of the study is to examine the functioning of financial markets in Zimbabwe and to examine the implications for the development of the country. This chapter is organized as follows. The first section presents the theoretical background for the ensuing analysis. The second section is devoted to a discussion of the financial market in Zimbabwe and pays special attention to the extent to which enterprises in Zimbabwe utilize both formal and informal markets to finance their investment expenditures and also considers the question of the use of overdrafts as a form of bank credit. The third section is devoted to the empirical investigation of the access of firms to formal finance. In Section 4 the use of formal external finance in the financing of the last major investment by firms in equipment is analyzed. Section 5 presents the conclusions and recommendations derived from the analysis of the access and utilization of formal finance by firms in Zimbabwe. 6.1 Theoretical background The relationship between finance and growth has drawn the attention of economists at both the micro· and macro~conomic levels. At the macro level although Keynes and many depression era economists stressed the importance of the institutional arrangements of finance, this emphasis was relegated to a second place by the interpretation of Keynes by some of his followers: they devoted most of their attention to Keynes' liquidity preference theory which emphasized the importance of money as opposed to the functioning of the credit markets. The emphasis on money was further promoted by the empirical work of Friedman and Schwartz (1963). In spite of the emphasis on money in the early years following the Second World War, a number of economists again took up the concern with the institutional arrangement of finance. Among those economists, Gurley and Shaw (1955) stand out because they put forward the following propositions: · In a highly developed country with a sophisticated financial system the government cannot affect the macroeconomic performance by modifying the quantity of money; · The levels of financial intermediation and direct finance are a function of the degree of economic development. In the sixties and early seventies Gerschenkron (1962) and Cameron (1967. 1972) posited the existence of a relationship between the growth of countries and the nature of the external finance available to firms. This hypothesis now figures prominently as the centre of a current policy debate among economists of the industrialized countries, the U.S., the U.K., Japan and Germany. Some economists (Mayer (1988); Cosh, Hughes and Singh (1990» argue that the competitive success of countries like Germany and Japan in contrast to the U.S. and the U.K. hinges on a "superior financial structure" where firms obtain their funds 55 for growth and expansion from banks that take more interest in the firms. Later in the seventies McKinnon (1973) and Shaw (1973) put forward the concepts of "financial repression" and "financial deepening" and sparked an imponant controversy relating finance and development. Nevertheless, McKinnon and Shaw made no effon to put their work in perspective and it is to be noted that they make no reference at all to either Gerschenkron or Cameron. By the middle seventies the development of the literature relating finance and develop- ment stagnated because of two other developments: · Modigliani and Miller (1958) demonstrated that under certain circumstances and assumptions the source of finance was irrelevant; and, · Macroeconomists expressed a preference to develop models on' the basis of "first principles" and proceeded to neglect differences in financial arrangements. Thus, the issue of the imponance of the institutional setting of finance and the imponance of stock markets was again abandoned. In spite of the long neglect of stock markets, these have gained a renewed interest because they have come to be seen as a too) to promote the growth of developing countries. Interest in the institutional setting of finance was rekindled in the 1980's by Bernanke (l983), Gertler (l988), Hellwig (1991) and others. Genler clearly points out the transition: "The main real/financial interaction in conventional Keynesian, Monetarist and Classical models stems from activity in the market for the medium of exchange, and not from the performance of markets for borrowing and lending. (p.l) If "Recently, interest has grown in exploring the possible links between the fmancial system and aggregate economic behaviour. This interest partly reflects the on-going beliefs of applied economists and policy makers that financial markets and institutions deserve serious attention - that they play important roles in the growth and fluctuation of output. "(P.l) At the microeconomic level, the neoclassical theory of invesonent has been unable to explain the observed firm preference for internal over externa1 sources of finance. This is because, since Modigliani and MiI1er (1958) developed the proposition of the irrelevance of fmancial structure to real invesonent, neoclassica1 theory has ignored financing issues in modelling investment (Gertler. 1988).Perfect financia1 and capitaJ markets are assumed. Finns that face liquidity constraints in the presence of profitable invesonent opportunities can always borrow. They borrow at market determined interest rates which, at the margin, equals the real rate of return on invesonent. Under these assumptions and logic,internal and external finance are perfect substitutes. Hence, the firm's capitaJ structure and financing are irrelevant to real invesonent decisions (ModigJiani and Miller, 1958; Jorgenson and Hall, 1967). Although largely developed and applied in the context of developed countries, recent advances in the theory of imperfect information can provide useful insights into the observed financing behaviour of firms, particularly in a less developed country context. This literature builds on Akerlofs (1970) semina1 article on the "market for lemons". It demonstrates that . under various types of market imperfections, -financing hierarchies", in which internal sources are preferred to external sources of finance, arise. Furthermore, if external sources are used at all, debt is preferred to equity (Myers, 1984; Myers and Majluf, 1984; Fazzari, 56 Hubbard and Petersen, 1988; Oliner and Rudebusch, 1992). Thus, by relaxing the standard neoclassical assumption of perfect markets, it is possible to demonstrate that imperfections in financial and capital markets impart a cost premium on external finance. This makes it less preferred to internal finance. Under these assumptions, internal and external finance are no longer perfect substitutes. This might explain the observed "pecking order" pattern of financing investment by firms (Myers, 1984). We apply the above theories to explain the observed firm preference for internal finance by manufacturing firms in Zimbabwe. Familiarity with Zimbabwe informs that the "frictionJess competitive markets" model is not a realistic conceptual framework for the analysis of Zimbabwe's financial sector. Instead, the theories of financial market imperfecti- ons hold" greater promise in explaining the observed firm financing outcomes. We postulate that Zimbabwean financial and capital markets are characterised by imperfections that tend to inhibit the quantity and quality of financial intermediation. Information asymmetries Akerlof (1970), Stiglitz and Weiss (1981), Myers and Majluf (1984) and others have shown that if asymmetric information exists between borrowers (firms) and lenders (eg banks) about the quality of projects, this may impair the functioning of financial markets. The asymmetry imparts a "lemons premium" on external finance. In controlled or highly regulated financial markets, as exists in Zimbabwe, the lemon's premium (higher costs of borrowing) may manifest itself in more perverse ways than higher interest rates. When this premium becomes sufficiently high, profitable projects may be rationed out of credit markets leading firms to rely on solely on internal finance. There are ample reasons to believe that asymmetries of information are very significant in Zimbabwean financial markets. While the argument is perfectly general to all firms, its applicability is most pertinent to small and indigenously owned firms. Agency costs Agency costs are often articulated in the context of large and well developed firms, with reference to external equity finance. They arise from a potential conflict of interest between firm managers and outside stockholders and bond-holders. To protect their interests, external shareholders attempt to com~")l management behaviour by use of audits, budget restrictions and strict monitoring. How agency costs might arise in a Zimbabwe specific context, especially with respect to bank finance, will be explored for both large and small firms. Transactions costs Transactions costs provide another explanation of why firms prefer internal over external finance. In countries with developed financial and capital markets, they arise from the costs of new equity registration, dealer's fees, taxes, legal, accounting and printing costs. The inspiration behind the transactions costs argument, in Zimbabwe applies in a slightly different manner. Few firms are listed on the stock exchange and hence most firms do not issue equity and debt contracts to the public. Therefore, transactions costs relate mainJy to the costs of borrowing from banks and other financial institutions. However, to the extent that they can be identified, it would be expected that small firms incur higher transactions costs per unit of external finance than larger firms. 57 Either of the problems noted above - information asymmetries, agency costs or transactions costs - lead to different cost structures for the different sources of finance and lead to the "pecking order" described by Myers (1984). Evidence of a "pecking order" would indicate a market failure in the financial system that might lead to slower aggregate growth. 6.2 Financial intermediation in Zimbabwe A common observation about the state of financial market intermediation in Sub-Saharan Africa is the severe underdevelopment of institutional, formal sector finance. Consequently, informal and semi-formal financial markets exist on a much wider scale than in other parts of the developing world. Zimbabwe is a partial exception. By Sub-Saharan African standards, she has a relatively developed banking and financial sector. Informal and semi-formal financial markets are not widely prevalent. This gives one an expectation that enterprises would have relative ease of access to formal sector finance. This expectation is not supported by our survey data. Less than half (47.7%) of the sample firms have ever received a loan from a formal financial institution. Part of the lack of access to credit may be explained by self-selection: among 183 of the firms that responded the question 42.1 % had never asked for a loan. These characteristics of the formal credit sector are summarized in Table 1 and will be considered in more detail in the next section. Table 6.1: The use of formal sector finance in Zimbabwe question yes no tota' Did you ever get count 94 103 197 a loan7 percent 47.7 52.3 100 Did you ever ask count 106 77 183 for a loan7 percent 57.9 42.1 100 source: RPED-study Zimbabwe The lack of access to formal credit mentioned above is also replicated when shorter and more actual periods of time are taken into account. Thus, considering the last five years, only 59 out of 201 firms obtained a formal bank loan .. An additional 30 enterprises obtained credit from a non-bank fmancial institution, 11 from the government and 10 from others, mostly parent companies. Narrowing the period of time to one year the number of com- panies that obtained bank credit was 37. credit from a non-bank financial institution 30, credit from the government 17 and credit from a parent company 7. These details are summarized in Table 6.2. 58 Table 6.2: The number of formal institution loans reported by the 201 firms during the last five years and the last year institution nr. of loans obtained in the nr. ot loans obtained in the last 5 years last year banks 59 37 non-bank financial institutions 30 14 government programs 11 8 others 10 7 source: RPED-study Zimbabwe That the majority of manufacturing firms in Zimbabwe rely predominantly on internal finance rather than formal sector institutional finance can also be inferred from the very limited use made of bank loans and other forms of institutional finance for the establ ishment or start -up of enterprises and the subsequent finance of expansion investment to fund investment expenditures. This is true for both start-up and subsequent investment expendi- tures. Table 6.3 presents the number of firms that made use of particular forms of finance at start up. The table also indicates how many enterprises utilized a particular source for 100% of their funding needs. Table 6.3: The sources of funds for start-up the sources of funds number of firms that use number of firms that rely this source for 1 00% on this source own savings 141 103 borrowing from friends and 15 4 relatives loan from a foreign bank or 3 donor agency loan from a Zimbabwean 19 2 bank loan from a money lender 4 1 loan from a supplier 1 o other (generally parent 34 19 company) source: RPED-study Zimbabwe The data in Table 6.3 reveal that the primary source of funds to establish a new enterprise was own savings: 141 enterprises financed a significant ponion of their investment out of own savings and 103 financed 100% of the investment this way. The use of bank loans and other forms of institutional finance do not appear to be very important sources of funds for the establishment of new enterprises. Surprisingly. informal borrowing, from friends and relatives and from the money lender, also proved to be insignificant. A significant source of funds proves to be that derived from the parent company in the establishment of subsidi. aries. The reliance on personal savings suggests some lack of depth of the financial and capital markets (in the McKinnon sense). The pattern observed with respect to financing at start-up is repeated in the funding of post start-up expansion investment in land,buildings. machinery and equipment. With respect to · land omy 26 out of the 201 firms interviewed had made a major investment in land in the foregoing five years. As depicted in Table 6.4 the greatest number of firms financed the acquisition of land with retained earnings. Among the 19 firms that employed retained earnings, 16 financed the purchase completely out of retained earnings. However, the S9 greatest expenditure on the acquisition of land (64 %) was completed by 7 firms that relied mostly on parent company funds. The role of banks in helping enterprises acquire land is minimal. Thus banks came to the aid of only 3 of the 26 companies that purchased land and their contribution to the total expenditure on land only amounted to 7 %. The table also shows that the contribution to land expenditures from informal sources was negligible. Table 6.4: Source of funds for investment in land (26 observations! source of funds nr. of companies nr. of companies expenditure by using this source using this source for source 8S a % 100% of expenditure of total expen- diture retained earnings 19 16 29 personal savings o o o from friends/family o o o bank loan 3 2 7 supplier credit o o o money lender o o o other(mostly parent co.l 7 5 64 source: RPED-study Zimbabwe The sources of funds for investment in buildings are virtually identical to those for the finance of land acquisition as can be seen from Table 6.5. This is somewhat remarkable because we are talking about a very different set of firms. Rather than 26 firms purchasing land we are dealing with 58 companies investing in buildings and relying to a lesser extent on only one source of funds for 100% of the investment. Nevertheless, with respect to the funds allocated, parent company financing accounted for 60% of the total expenditures on buildings followed by the companies that fmanced their investment from retained earnings (36%). Banks were involved in the acquisition of only 8 buildings and their contribution of funds represented only 4% of the total expenditure on buildings. Informal sources of funds made no contribution to the investment in buildings. Table 6.5: Source of funds for investment in buildings (58 observations) source of funds nr. of compenies using nr. of companies using expenditure bV source as this source this source for 100% of · % of total expenditure expenditure retained earnings 43 31 36 personal savings 2 2 0 from friends/family 0 o 0 bank loan 8 3 4 supplier credit 0 o 0 money lender 0 o 0 otherlmostly parent co.l 18 9 60 source: RPED-study Zimbabwe Bank loans, however, playa more important role in the purchase of plant and equipment. As shown in Table 6.6, the finance provided by banks to 39 companies was equal to the investment out of retained earnings of 124 companies. In spite of the more active role of banks in the finance of plant and equipment, as a per cent of total expenditures, the investment financed by parent companies remains the most important. The table also shows that the funds provided by personal savings, friends and relatives, suppliers or money · lenders are minimal. 60 Table 6.6: Source of funds for investment in plant and equipment (166 observations) source of funds nr. of companies using nr. of companies using expenditure by source as this source this source for 100% of 8 % of total expenditure expenditure retained earnings 124 94 29 personal savings 10 7 1 from friends/family 1 o o bank loan 39 17 29 supplier credit 6 2 4 money lender other(mostly parent co.) 1 22 ,,o o 37 source: RPED·study Zimbabwe Taking into account that expenditures on equipment represent 82% of the total investment against 16% for buildings and 2% for land, it must be concluded that expenditures on investment by banks are almost as important as the investment expenditures financed out of retained earnings. Therefore in Section 4 special attention is paid to the role of banks in the finance of investment in equipment. Another conclusion that may be drawn from the analysis of Tables 6.4, 6.5 and 6.6 is that the use of informal sources of finance such as money lenders and friends and relatives is not very important in Zimbabwe. The above figures may understate the role of formal, especially bank, finance as over two thirds of firms report access to bank overdraft facilities. The balances on overdrafts are of comparable orders of magnitudes to bank loans. However, the survey questionnaire did not allow for data collection on overdrafts in as much detail as for bank loans which constrains further comparisons between the two. The greater incidence of use of overdrafts can be explained in several ways: (i) Flexibility An overdraft is in fact an untied loan. It can be drawn to finance either working or physical capital or both. This flexibility is more attractive to enterprises, than bank loans, which are usually tied to financing a specific, prior approved project. Thus there is reason to suspect that some firms use overdrafts as substitutes for bank loans. This is evidenced by the fact that 68 per cent of firms received overdrafts compared to 48 per cent for bank loans. With these substitution possibilities, it is not surprising that 46 per cent of firms who did not apply for loans gave the reason as "I did not need one". (ii) Interest Costs Many firms reported their preference for overdrafts because they involve cheaper interest costs. This is so because interest on overdrafts accrues on the withdrawn amount whereas interest on loans accrues on the entire amoum effective from the date of approval. Unless the interest rate differential is substantial. it pays a firm with a substantial overdraft facility, to finance the desired investment from the former and thus avoid the transactions and interest costs associated with a bank loan application. (iii) Transactions Costs The transactions costs of loan application, processing and delays in approval are substan- tial compared to those of an overdraft. This is so because overdrafts involve a "once only" transactions costs as against bank loans where each application has a transactions cost associated with it. Thus, the transactions costs of an overdraft are fixed costs which decline with time and greater utilisation of that facility. (iv) Collateral Bank loans usually require physical collateral which some firms may be hard pressed to provide. Overdrafts are less formal since its limit is decided at the bank manager's discretion 61 rather than the size of an investment project, The firm's track'record as the bank's cus- tomer, the character of the owner or manager of the firm, and interpersonal relations with the bank manager play a more influential role in the decision to grant an overdraft than the provision of physical collateral. This might explain why some firms whose loan applications were refused, report access to overdrafts, It is anticipated that with a new round of interviews a more detailed picture of the trade-offs between formal bank loans and over- drafts may be provided. The extent of informal financial intermediation is reflected, firstly, in the use of informal markets to raise finance for investment purposes and secondly, by the magnitudes of informal financial market instruments in firms' asset and liability portfolios. Questions were posed to firms' about their informal borrowing and lending activi~ies. These may give insights into the breath and depth of informal financial markets in Zimbabwe. In only 57 of the 166 valid cases (34 per cent) did firms report informal borrowing in the previous 3 years. A substantial proportion of these (31 cases) were in-kind loans - of raw materials and equipment - during times of shortages. Strictly speaking, these are not an aspect of financial intermediation: this practice is unlikely to be linked to liquidity or financial constraints facing firms but rather to bottlenecks in real goods markets. In-kind loans are an aspect of business linkages and networking among enterprises. Of the remaining 35 cases, the majority are accounted for by small (9 cases) and medium sized (5 cases) firms borrowing from friends and relatives and informal groups (9 cases). Borrowing from money lenders, clients and suppliers were encountered in one or two odd cases each. Indigenous and new firms were more likely to borrow informally than other and older firms. By far, the largest proportion of informal lending is for consumption purposes to firm employees (138 out of 201 firms in the last 3 years) rather than investment lending to other firms. Cash loans to other enterprises were encountered in only 2 instances with the bulk (39 cases) being in-kind loans to which the caveat mentioned earlier applies. As noted above, neither suppliers nor clients feature prominently in the informal lendinglborrowing matrix (4 and 3 cases respectively). Disregarding in-kind and employee loans, small firms were more likely to lend informally than larger firms; but mainly to friends and relatives. A major conclusion that can be reached is that financial intermediaries playa limited role in the financing of investment in Zimbabwe. Government intervention in financial markets, (through instruments like interest rate controls and credit allocation) has been blamed for the state of poor financial intermediation in many developing countries. In Sub-Saharan Africa, it is commonly observed that financial markets are fractured, segmented and severely undeveloped. This is usually evidenced by (a) the limited use of formal financial market instruments to finance investment and (b) widespread reliance on informal and semi-formal finance by enterprises. The first contention seems supported by our data. However, the alleged prevalence of informal and semi-formal financial markets is not sustained. The next section is devoted to the analysis of the limited access of firms to formal financial markets. 6.3 The 8CCt'SS or firms to formal finance As pointed out in the last section, the access to formal credit in Zimbabwe appears to be fairly restrictive. Two hundred and one enterprises were asked if they had ever received a loan; 197 responded the question; 103, that is 52%, had never received a loan. The answer to the reduced number of firms that had ever had access to credit from a formal institution (bank, credit union, government projects and others; often referred to in this section as "banks" for brevity) can be attributed to two factors: first, a number of firms never applied 62 for a loan; and second, banks and other lending institutions did not lend to all applicants. The purpose of this section is to examine the characteristics of the firms that did and did not apply for credit and to examine the characteristics that banks considered most imponant in the selection of the firms to fund among all applicants. This inquiry will concentrate on two different samples or "windows": those that ever applied and obtained credit; and those that did so in the "last year". The advantages of this dual approach will be spelled out in the discussion of those that applied "last year" which will be examined in the latter pan of this section. Among the sample of 201 enterprises interviewed we found that 72 had never asked for a loan. Although one might say that this number goes a long way to explain the 103 firms that never got a loan, the reasons for not applying given by these firms indicate that a number of them would have wanted a Joan if they had had the adequate collateral or if they thought they could obtain a loan. Thus, it was found that 11 firms would have wanted a loan but in fact credit constrained themselves. Therefore, it might be said that more firms wanted loans than the number that actually applied. Employing the negative answers of the firms with respect to their not asking for a loan, it is possible to divide the group of 197 enterprises that responded the question jf they had ever received a loan into two subgroups: 136 that wanted loans - and that for ease of exposition are sometimes going to be referred to a applicants for a loan - and 61 that did not want a loan. In order to explain the borrowing decisions of firms a number of characteristics of the firms were employed as independent variables. The first of such variables is size, ldolsale. The reason for using the size of the company as a characteristic that might influence a firm's decision to apply for a loan are the following: large firms will have a more sophisticated administration that will make it easier to apply for a loan; the transaction costs for a loan may be relatively fixed. giving large firms with large loan requirements a relative cost advantage; large firms may feel that they have a comparative advantage in their bank loan appJications because of their importance as clients for the bank. The measure of size employed was the value of sales since this is deemed to be the most relevant variable to a bank because it can be easily verified and because it may give an indication of profits. The measure of size by personnel employed would not meet either of these two criteria so wen. In the probit regression the coefficient of the size variable is expected to be positive. The second variable that might be included in the probit equation is the profitability per unit of sales of the company. The profitability of a company could be computed in either of two ways: utilizing the profits assessment provided by the firms; or, calculating a profit measure for the firm by subtracting costs from the value of sales. The ratios of profits to sales are denominated profit for the measure that uses the firm's assessment and cproft for the measure that utilizes the calculated profits. The profitability of a company might affect the borrowing preferences of a company in two different ways: first, a highly profitable enterprise might generate a cash flow that could be sufficient to meet investment expendi- tures; second, a high profitability company might want to expand at a rapid rate thus leading it to apply for bank loans. The question as to the effect of profitability on the borrowing behaviour of an enterprise is an empirical one and the sign of the coefficient may not be anticipated. The age of a firm was selected as a characteristic that might influence borrowing behaviour because it could be realistically assumed that older firms migbt bave bad a longer relationsbip with the banks, thus facilitating their application for a loan. This information relationsbip, wbich would lead to the expectation of a positive coefficient for the age variable, lage, might bowever be offset by the lack of a need for a loan because of the presumably significant reserves built up by an older company. Therefore, again, it would be 63 difficult to forecast the sign of the coefficient for the age variable. Another variable that could be employed to explain the borrowing behaviour of the firm would be the level of education of the owner of the firm. This variable would be deemed to be relevant because owners with a higher level of education would have less trouble completing the necessary forms for the application of a loan. The educational variable could be a dichotomous variable indicating whether the owner had reached either a secondary or a university level of education or not. The sign of the coefficient for this variable would be expected to be positive. In the regression equation a dummy variable, coilat, was employed to indicate the firm's ownership of its premises. It is anticipated that the coefficient for this dummy will be positive because a firm with a title to its premises might more easily provide the necessary collateral to back its loan applications. Because different sectors exhibit different characteristics with respect to a number of variables not included explicitly in our regression, we also utilize dummies to indicate the sector of the firm. The dummies for the various sectors are labelled: text for the textile industry; melt for the metal working industry; and wood for the wood working industry. Because it is not exactly known what characteristics of the industries of the various sectors might make them more likely to ask for credit, it is impossible to forecast the signs for the coefficients for these dummy variables. Another factor that was considered to be important in a firm's decision to apply for credit would be its legal status. With respect to legal status we have two sources of information, one derived from the questionnaire and another from external sources. The information from the questionnaire is useful in identifying sole proprietorship or partnership companies, corporations and cooperatives. Unfortunately, because of a number of subsidiaries are also limited liability companies, the data on subsidiaries and limited liability enterprises from the questionnaire is not very reliable. Therefore, making judicious use of information provided by external sources, new variables were constructed to indicate the legal status of the firms in question, which were subdivided into the fonowing eight categories: 1) cooperative 2) firm with internal working owner(s) ('entrepreneurial firm') 3) firm with external owner(s), being a small shareholder (may be overseas) 4) firm with working owner(s), but part of a (family-structured) group of firms. not being a stock exchange fund 5) subsidiary of a Zimbabwean stock exchange fund 6) subsidiarY of a foreign farm 7) stock exchange farm 8) (para)statal Among these legal status variables special attention was paid to sole ownerships and partnerships, grouped as the variable pernat; variables indicating a subsidiary status, of a domestic or a foreign company, derived from the external sources, newsubs; corporations, corp; and firms belonging to a network of firms dominated by a family group, jlygrp. The reasons for concentrating on these legal status variables is that they permit some inference with respect to the sign of the respective coefficients. Thus, for sole proprietorship and partnership companies a positive coefficient might be expected because it is not anticipated that these companies have any other possible source of credit. For subsidiaries the opposite would be expected: thus a negative sign would be expected for the subsidiaries coefficient because it might be anticipated that the enterprises would obtain their funds from the parent company. With respect to corporations the sign of the coefficient is an empirical question of great relevance: do corporations prefer to borrow from the banks or to obtain the necessary 64 funds in the equities market? With respect to the family group enterprises, again the sign is questionable because firms might prefer to employ group resources rather than apply for a loan. Because of the multi-racial character of the Zimbabwean entrepreneurial world it was also considered important to examine the access to credit of the various ethnic groups. Therefore. the following dummies were constructed to test whether different ethnic groups had a higher propensity to ask for credit: Zimbabwean owners of african origin, black: Zimbabwean owners of European origin, while; and Zimbabwean owners of asian origin, asian. The question of the sign of the coefficients is again an empirical issue. A last variable that was included in the regression equation refers to location. Thus, a dummy was constructed to identify the firms located in Harare as compared to those located elsewhere: in Bulawayo and in the provinces. A positive coefficient would be anticipated for this variable since proximity to the headquarters of banks could facilitate the application by the borrowers and the verification by the banks. Although it would have been very interesting to include a variable indicating the financial status of a firm such as gearing, the data set available did not permit the computation of such a variable. It is hoped that after the second round of interviews such a variable may be constructed . The following list provides a brief summary of the variables tested: Idolsale the logarithm of the value of sales converted to dollars in thousands of dollars. profit the ratio of reported profits to working capital, defined as the sum of all current expenditures over the past year. cproft the ratio of the calculated profits to working capital. cproftsq the square of the ratio of calculated profits to working capital. lage the logarithm of the age of the firm educ dummy indicating the level of education attained by the owner. collar a dummy variable indicating whether the firm has a title deed to the business site that might be used as collateral. text a dummy to indicate whether the firm belongs to the textile sector metl a dummy to indicate whether the firm belongs to the metal working sector. wood a dummy to indicate whether the firm belongs to the wood working sector. pernat dummy reflecting whether the enterprise has the legal status of either sole ownership or partnership newsubs dummy variable indicating whether the enterprise was a subsidiary of a Zimbabwean stock exchange fund or a subsidiary of a foreign fJll1l corp dummy indicating whether the company is a corporation or not. jlygrp dummy indicating whether the firm belongs to a family group. black a racial dummy to indicate Zimbabwean of African origin. white a racial dummy to indicate Zimbabwean of European origin. asian a racial dummy to indicate Zimbabwean of Asian origin. city a dummy to indicate whether the firm was established in Harare The above variables were tested in a number of probit regression exercises which were not always comparable because some variables would exhibit more missing data than others. Therefore, as a general rule, it was decided to drop variables that did not prove to be significant and that constrained the number of data points for the probit exercise. Thus, for examp1e, contrary to our expectations, the education variable, educ, did not present a coefficient significantly different from zero even at the very generous 90% significance 65 level. In addition to its lack of significance, because the education variable only contained 134 valid observations - the education question was not relevant for subsidiaries, for example - it was dropped from the exercise and does not show up in Table 6.7. Table 6.7 presents the characteristics of firms that determine their demand for credit. Table 6.7: Probit estimation of the demand for credit by firms variable coefficient t -statistic signif. level constant ·1.80 -2.85 Idolsale 0.20 3.04 ··· collat 0.09 0.41 lage 0.05 0.39 text 0.43 1.88 ·· metl 0.04 0.15 pernat 0.34 1.28 · newsubs -0.44 -1.64 ·· black 1.14 3.15 ··· city 0.36 1.32 " nr. of D's 61 nr. of 1 's '36 correct rate 72% sensitivity 92% specificity 28% levels of significance for one sided tests: " for 90%; ·· for 95%; """ for 99%. source: RPED-study Zimbabwe Table 6.7 indicates which are the predominantly significant characteristics of the firms that indicated a desire for credit. The anticipated positive coefficient for the size of the company was corroborated by the exercise with a 99% level of significance. Surprisingly. neither the owner response or the calculated profitability ratios, profit and cpro/t. produced a coefficient with a minimal level of significance. The low level of significance of the coefficients obtained in the alternate regressions including either profit or cpro/t plus the fact that the sample size was reduced significantly lead to the deletion of these variables from further trials. The lack of significance of the profit and cpro/t coeffi- cients might be attributable to the offsetting effects of a higber casb flow and the desire for rapid growth. Also unexpectedly. although the coefficient for the collateral dummy, collat, is positive, it is not significant at any acceptable level of significance. This result is unexpected because it was presumed that collateral was a relevant variable and because in the defmition of those who wanted a loan, those who had credit constrained themselves because of their perception of inadequate collateral, were included as wanting a loan. Thus, the definition of a company wanting a loan was, in fact, slightly biased in favour of the obtention of a positive and significant coefficient for collat. The coefficient for the age of the firm, lage, was positive but not significant. This result migbt imply that older firms rely more on retained earnings and that their age does not provide them any informational advantages. With respect to the sectoral dummies, the coefficients for the food, wood and metal sector dummies did not prove to be significantly different frOIJl zero in alternate trials of the probit equation. In the end, the food sector was dropped to avoid a perfect correlation between the sectoral dummies and the constant term and the wood sector was deleted because of its extremely low significance level. 66 In contrast to the above named sectoral coefficients, the coefficient for the textile sector proved to be positive and significant at the 5% level. A possible reason for this result may be that investment decisions in the textile sector may be lumpier than in the other sectors. As had been anticipated, the coefficients for the legal form, pernat and newsubs, proved to be positive and negative with levels of significance of 10% and 5% respectively. The coefficients for the corporative legal status did not prove to be significantly different from zero and the variable was removed from the equation. The coefficients for the ethnic group variable did not turn out to be significant except for the entrepreneurs of african origin. Thus the coefficient for the black variable is positive and significant at the 1% level. This result may indicate either that entrepreneurs of African origin may feel more secure in formulating credit applications or that African entrepreneurs face higher internal liquidity constraints than their counterparts with a different ethnic background. Finally, the coefficient for the proximity variable turned out to be positive and significant. This result suggests that firms located in Harare are more likely to apply for credit than enterprises located elsewhere. The characteristics of the firms that desire credit can be summarized as follows: o Larger enterprises are more propense to apply for credit than the small enterprises. o Enterprises from the textile sector are more likely to apply for credit than firms from other sectors. o Enterprises characterized as sole ownerships or partnerships are more likely to apply for credit firms with a different legal status. o Subsidiaries do not apply for formal credits as much as other enterprises presumably because they rely on parent company funds. o Black entrepreneurs tend to be more propense to apply for credit than other entrepreneurs with another racial background. o Enterprises located in Harare are more likely to apply for credit than their counterparts located outside the capital. The fit of the equation can be considered fairly good. The percent of correctly predicted firm decisions with respect to credit is 72 %. The sensitivity measure which reflects the number of times that the model correctly predicts the occurrence of an event, in this case the application for a loan, is 92 %. The specificity rate which presents the ratio of accurately forecast zeroes to the observed decisions of the firms not to apply for credit is, however, only 28%. Although it would be ideal if the specificity rate were also close to 100, for a cross sectional exercise like that conducted here, a specificity rate of 28% is not altogether unsatisfactory . In order to determine the characteristics that banks thought important for the granting of loans we again applied a probit to the subset of 136 firms that wanted loans. This subset includes 7 firms that credit constrained themselves and that are retained in the subset because it may be argued that they anticipated the decision of the bank. Among the 136 firms that ever wanted a loan, 94 did obtain a loan. To explain the decision of banks to ever grant these loans we again employed a probit. The dependent variable was the bank decision to grant or not the loan. The independent variables employed to explain this bank decision include many of the variables employed to explain the firm decision to apply for a loan. With respect to the decision of banks to grant loans, the size variable. ltiolsale, was included in the regression as one of the characteristics that would influence a bank decision 67 for two principal reasons: larger firms are better known and are more easy to monitor: and, large firrns might have closer relations with the banks. Thus in the probit analysis of the credit decisions of banks the variable Idolsale is expected to yield a positive coefficient. Another variable that might be of particular relevance to the decision of a bank would be the profitability of the company, defined as the ratio of profits to working capital. A high profitability would indicate a high cash flow and the ability to repay the loan. Thus a positive coefficient would be expected for this variable. With respect to the age of the company it is anticipated that the sign of the coefficient wiJl be positive. This result is expected because a longer relationship with the bank would provide the bank with more information about the firm. Another variable that would facilitate the granting of a loan by the banks would be the availability of collateral. A positive coefficient might be expected for the variable coUal. With respect to the different sectors of the economy, the banks might favour one or another sector on the basis of characteristics different from those explicitly included in the regression like size, profitability or age. However, without a clear idea as to what these characteristics might be, it is difficult to predict the sign of the coefficients for the different sector dummies. With respect to the legal status of the applicants, it is difficult to say if any would play an important role in the decision of a bank to grant a loan. This would be an empirical question. The same might be said with respect to the ethnic background of the applicants. There is no reason to believe that the ethnic background of a client would affect a bank's decision to grant a loan. The question of the sign of the coefficients for ethnic variable dummies would be an empirical issue. Finally, the anticipated sign for the location dummy, city, would be positive. This expectation would be based on the greater ease with which banks could monitor the information provided by the applicants of Harare in contrast to those in BuJawayo and the provinces. The results of the probit estimation to investigate the determinants of the banks' decisions to grant credit are presented in Table 6.8. Table 6.S: Probit estimation of the decision of banks to grant credit variable coefficient t-Itatistic sign.level constant -0.60 -0.57 Idolsale 0.34 3.17 ··· cproft 0.24 1.02 lage -0.47 -2.05 ·· text 0.57 1.26 metl 0.70 1.18 wood 0.29 0.49 pernat -0.26 -0.67 newsubs -0.56 -1.17 black -0.07 -0.15 city -0.12 -0.25 nr. of as 32 nr. of 1s 60 correct rate 78% sensitivity 88% specificity 59% levels of significance for one sided tests: · for 90%; ·· for 95%; ···for 99% Table 6.8 shows that the size variable, /dolsaJe, is also crucial for the decision of banks to 68 grant credit. It must be inferred that this variable conveys a lot of information that the banks consider relevant. Although the coefficient for cprojt is positive it is not significant at the 90% level. Surprisingly, the coefficient for the age of the enterprise is negative and significant at the 95 % level. A large number of hypotheses could be formulated to explain this result but for the time being it remains somewhat of a paradox. Another result that deserves comment is that the coefficient for the African ethnic background dummy, black, turned out to be negative but not significant. The lack of significance of the coefficient, however, indicates that this regression does not provide any evidence of racial discrimination. In spite of the fact that the probit regression presented in Table 6.8 shows only three significant coefficients, the fit of the equation is fairly good. Thus the correct rate is 78% and the sensitivity and specificity rates are 88 and 59 percent respectively. The "last year" credit decisions of firms and banks A probit exercise similar to that used to investigate whether firms had "ever" asked and obtained loans was focused on the obtention of a loan in the "last year" for three reasons. The first reason for the duplication of the "ever" exercise on the rather smaller "window" of "one year" is that in the former exercise firms may have applied and obtained credit in periods far in the past, when their characteristics may have been very different from those recorded at the time of the interview. The advantage of the precision with respect to the characteristics of the firms offered by the smaller "window" is, however, to a certain extent, offset by the fact that a smaller number of firms will have confronted the credit decision. In the extreme, a time span of one week, for example, would assure that the characteristics of the firm were equal to those at the time of the interview but might provide no data with respect to credit decisions. The second reason for conducting the alternate exercise is to examine whether differences between the two exercises might permit some inferences about the evolution of the credit markets over time. For example, if the "ever" exercise showed the Harare location of an enterprise to be very important in contrast t.o the "last year" regression - where location might not be important - it might be inferred that the development of the banking network had eliminated the advantages of a capital city location. A third reason for duplicating the "ever" regression is to examine whether similar results are obtained in the smaller window exercise. The similarity of the results would provide an indication of the robustness of the analysis with respect to the characteristics that determine firms' and banks' decisions to apply and grant credit. In spite of the reasons presented above for conducting a comparison of two different windows, it must be mentioned that the results must be interpreted with caution because the second "window" only covers one year. The results of one year compared to those of a longer period of time may at bet imply a possible change and do not in any way demonstrate it. On the question if the fmn wanted a loan last year, only 65 responded affirmatively. This number includes 51 fIrms that effectively asked for a loan and 14 that credit constrained themselves. In the examination of the characteristics of the firms that expressed a desire for a loan the same independent variables of the "ever" exercise were tried in several exercises in order to exclude those with little explanatory power that might additionally be constrain~ ing the size of the sample. Table 6.9 presents the results of the probit exercise on the question whether fIrms had 69 requested a loan "last year". Table 6.9: Probit estimation of the decision of firms to apply for credit in the ftlast year" variable coefficient t -statistic signif. level constant -2.18 -2.69 ··· Idolsale 0.32 3.63 ··· cproft 0.40 , .57 · cproftsq -0.06 -, .67 ·· collat -0. '3 -0.47 lage text -.39 0.68 -2.22 1.85 ... ·· metl wood 0.82 1.11 1.75 2.33 ... ·· ··· newsubs black -1.10 1.25 -2.73 2.99 ..... city -0.01 -0.03 nr. of Os 77 or. of 15 36 correct rate 75% sensitivity 47% specificity 88% levels of significance for one sided tests: .. for 90%; .... for 95%; .... for 99% source: RPED-study Zimbabwe The results presented in Table 6.9 indicate that in the decision to ask for a loan the size of the firm plays an important role just like in the "ever" exercise. However, unlike the results obtained in the latter exercise, in the "one year" regression the profitability of the enterprise proved to be a significant variable; the coefficient for cproft is positive and significant at the 90% level and the coefficient of cproftsq is negative and significant at the same level. This combination of results suggests that profits have both a positive and a negative effect on the decision of a firm to apply for a loan. The positive effect would be derived from the fact that current profits may be a good indicator of future profits, which would induce firms to invest and to require credit for the financing. The negative effect would arise from a substitution of sources of finance: for a given level of investment, the high profitability firms do not require loans. The net effect of profits on the firm decision to apply for a loan is ambiguous but our results suggest that the positive effect dominates. The results with respect to collateral do not differ substantially from those presented earlier. Unlike the collateral coefficient, the coefficient for age became negative and significant at the 95 % level. Like in the previous exercise, in the "one year" regression the coefficients for the textile sector dummy proved to be positive and significant. However, additionally the wood and metal working sectors coefficients proved to be positive and significant in the "one year" exercise. The significance of the coefficient for the variable representing sole proprietorships and partnerships was found to be so small that it was deleted from the equation. In the "ever" exercise a positive coefficient had been determined for this variabJe. In full correspondence with the results obtained earlier. a negative coefficient was obtained for the subsidiaries variable indicating that subsidiaries probably obtain their funds from their parent companies and do not apply for credit. A similar correspondence was found with respect to the ethnic variable. In both exercises 70 the entrepreneurs of African origin showed a greater disposition to apply for credit than their counterparts of other ethnic backgrounds. Unlike the previous exercise the lo~ation of fi~ in Harare did not affect its propensity to apply for credit. Table 6.10 summarIZes the POInts of correspondence and disagreement between the two exercises. Table 6.10; The points of correspondence and disagree- ment between the "ever" and "last year" probits on the decisions of firms. variable "everA "last year" Idolsale + + cproft 0 + cproftsQ 0 cOllat 0 0 lage 0 text + + metl 0 + wood 0 + pernat + 0 newsubs black + + city + 0 source: RPED-study Zimbabwe Table 6.10 indicates that irrespective of the window chosen, the size of the firm is an important characteristic in the explanation of its decision to ask for credit. Similarly, it may be said that textile firms are more likely to apply for credit than their counterparts in other sectors. Also, it is shown that subsidiaries do not tend to apply for credit. Finally, the African ethnic background of entrepreneurs was found to be positively related to credit application in both windows. The differences in the results of the two exercises might suggest the occurrence of changes in the credit markets. Thus, the positive coefficient for profits may indicate a change in the analysis parameters for credit. Thus, recognizing that there may be a great deal of self selection in the application for credit, it might be inferred that banks have changed their emphasis to profitability. A similar change in the credit markets might be the cause for the drop in the relevance of the city advantage registered in the "ever" exercise. Thus, with better communications and a more widespread network of banks, firms in Bulawayo and the provinces might be just as likely to apply for credit as their counterparts in Harare, leading to the zero coefficient for city in the "last year" exercise. Finally, that the coefficient for wood and metl became significant in the "last year" exercise might imply either of two things: banks have become more flexible and lend to enterprises in the wood and metal sectors; or, the enterprises in these sectors have "grad- uated" into the credit market. Among the 65 firms that requested a loan "last year", 36 firms were granted a loan. The applications of 29 enterprises were rejected. To investigate the characteristics of the firms that the banks deemed important for the granting of credit another probit regression was run on the reduced data set. The results of the bank decision regression are presented in Table 6.11. 71 Table 6.11: Probit estimation of the decision of banks to grant credit variable coefficient t-statistic sign.level constant -1.65 -1.90 ·· Idolsale 0.15 1.60 · cproft 0.62 1.74 ·· lage 0.00 -0.01 text 0.82 1.39 · metl 0.22 0.27 wood 0.37 0.47 nr. of Os 21 nr. of 1s 24 correct rate 75% sensitivity 83% specificity 67% levels of significance for one sided tests: · for 90%; ·· for 95%; ··· for 99%. source: RPm-study Zimbabwe The results of Table 6.11 are fairly similar to those obtained for the "ever" exercise. The principal results of the two exercises are summarized in Table 6.12. Table 6.12: CompariSon of the "ever" and "last year" decisions of banks variable "ever" granted loan loan granted "last vear" coefficient significance coefficient significance Idolsale + ··· + · cproft + + ·· lage ·· o text 0 +. levels of significance for one sided tests: · for 90%; .. for 95%; ··· for 99%. source: RPm-studY Zimbabwe The banks base their credit decisions primariJy on the size of the applicant firms. However, the increased importance of profitability in the "last year" may indicate a change in the banks credit evaluation criteria. On the other hand, the age of the firms ceases to be relevant in the "last year" period. This may signify either that older firms got credit in the last year or that information problems may have changed. The positive sign for textiles in the "last year" may indicate the recovery of investment projects and their corresponding funding. The analysis of this section has shown that certain characteristics of firms play a significant role in the application of credit by firms and the granting of credit by banks. In order to reduce the problems of self selection we included firms that had - according to the interview - credit constrained themselves in the group of applicants. It was not expected that this measure would eliminate the problem of self--selection completely and this may account for the similarity of the characteristics of the firms that apply for credit and the characteris- tics used by banks as criteria in the granting of loans. Thus, the results indicate that large firms with a high level of profitability and belonging to the textile sector are more likely to apply and receive credit than companies with other characteristics. Among the variables that did not prove to be significant to the explanation of credit or that showed the wrong signs, lage and collar stand out. The lack of significance of the lage coefficient may not be very perplexing in view of the weak hypothesis that sustained it as a proxy for information. The case with the lack of significance and wrong sign of the 72 collateral dummy is more troublesome in view of the large literature that emphasizes property rights as a pre-condition for the efficient functioning of credit markets. The lack of significance of the collat coefficient is all the more relevant because of the rather even distribution of firms with respect to having title of ownership to their premises: 96 firms have no tile while 105 do. In the questionnaire presented to firms in Zimbabwe a question was also included to determine the characteristics of the last loan for which the enterprise had applied. This question asked for such characteristics as: amount of the loan, maturity, interest rate and collateral. Unfortunately only 33 firms answered the question and little information could be gleaned from the answers provided with respect to relationships between year of the loan, maturity, amount of the loan and interest rate and the characteristics of the diverse firms. The respondents did, however, in 26 cases answer the form that the collateral took: - land and buildings 9 cases - equipment 12 cases - other 5 cases The rather infrequent use of land and buildings as collateral together with the fact that only about half the firms have a title to their premises may indicate that the institution of collateral may not be working very well. This failure might in turn limit the access of many firms to credit, especially the smaller ones which do not hold a title to their work site as often as the larger enterprises as shown in Table 6.13. Table 6.13: The possession of title to the work site by size of firm size of firms collateral (%1 number of (ODDs of US$) firms no yes < =25 77 23 31 26-100 67 33 18 101-500 50 50 34 501-1000 52 48 23 1001-2500 51 49 33 2501-5000 23 77 22 5001-10000 21 79 14 >10000 20 80 25 source: RPED-study Zimbabwe 6.4 Analysis of bank credit in major investments In the survey held among 201 firms in Zimbabwe, the respondents were also asked about the sources of f'mance for the major investment which they might have realized in the last five years. Among the 201 firms, 166 answered to have made a major investment in the last five years. However, as discussed in Section 2, only 39 of these enterprises made use of a bank loan to finance part or the whole of his investment. The purpose of this section is to examine the characteristics of the firms that employed bank credit in the finance of equipment. The use of bank credit for the acquisition of land and buildings is not considered because of the very limited role played by banks in the restricted number of cases where firms purchased land or buildings as was described in Section 2. The uses of other sources of credit are not considered because of their relative lack of relevance - for example, informal credit from friends and relatives or the money lender - or because they do not represent credit transactions, like funding by the parent company. Unlike the estimation exercises presented in the last section, with respect to the utilization of bank credit for investment it is impossible to look at the demand parameters separately 73 from the supply characteristics. This shoncoming is derived from the fact that the question whether a firm would have wanted credit or not was not explicitly formulated in the questionnaire. The only points observed are those where both firms and banks coincided on a bank loan. In addition to the characteristics examined in the last section, in the analysis of the utiliz- ation of bank loans to finance investment projects, the logarithm of the ratio of the value of investment to the yearly value of sales, lifOS, was also employed as an explanatory variable. The results of a probit regression to determine the characteristics of banks that got loans is presented in Table 6.14. Table 6.14: Problt analysis of the firms that obtained bank credit for the last major investment in equipment in the previous five years variable coefficient t-statistic sign.level constant -2.18 -2.90 ··· .o,o, Idolsale 0.25 3.31 o,o,o, litos 0.14 2.43 collat 0.00 0.00 lage -0.12 -0.69 .o, text 0.57 1.87 metl 0.13 0.35 wood 0.60 1.42 ., black 0.53 1.42 · nr. of Os 125 nr. of 1s 39 correct rate 75% sensitivity 15 % specificity 94 % levels of significance for one sided tests: · for 90%; · ., for 95%; .,.,. for 99%. source: RPED-studY Zimbabwe Table 6.14 indicates that the size of the company and the ratio of the investment to sales are important determinants for the obtention of a "bank loan. Presumably the size of the firm is an important characteristic for banks in their granting of credit and the size of the investment with respect to sales is an important element for firms to apply for credit. The results of Table 6.14 further corroborate the results obtained in Section 3 by showing: · textile and woOd firms are more likely to apply for and obtain credit than their counter- parts in other sectors; · Zimbabwean entrepreneurs of African origin are more likely to apply for and get loans than entrepreneurs with other ethnic backgrounds. Unlike the results obtained in the previous section, the ratio of calculated profits to sales, cprofts, did not show a coefficient that was significantly different from zero. Because of the lack of significance and the fact that it restricted the number of observations that could be employed in the probit, it was dropped from the regression analysis. Again, like in the last section, the coefficients for lage and collar failed to show coefficients significantly different from zero. As discussed in the previous section, these results probably point to the lack of relevance of age as an information proxy and to probable problems in the use of land and buildings as collateral. 74 6.S Conclusions The analysis presented in the foregoing sections allows a number of conclusions that can be related to the theory on credit markets presented in Section I. In spite of the fact that the informal credit markets are of little importance to the finance of firms, suggesting a developed formal financial system, it has been found that firms are significantly restricted in their access to bank and other formal financial institutions credit. Thus, less than half of all firms have ever received a loan and onJy 26 firms obtained a loan in the year prior to the interview. This would imply a lack of financial depth as defined by McKinnon. Throughout the analyses of access and use of credit it was found that large enterprises, as measured by the value of their sales, had better access to credit than their smaller counter- parts. This inequality in the distribution of credit can probably be attributed to transaction costs and to information asymmetries. Thus, larger firms with large credit needs might not find the fixed transaction costs an impediment for a loan application. With respect to information asymmetries, it might be found that the problem might be particularly relevant to small enterprises that would be credit constrained in the Stiglitz and Weiss sense. Alongside the size of the firm, other variables that proved imponant are the profitability of the firm, the sector of the firm and the ethnic background of the entrepreneur. Thus, it was found that in many of the cases analyzed, the profitability rate was an important determinant of both the firm's decision to apply for credit as well as the bank's decision to grant a loan. This finding might imply that the credit markets might be functioning well in spite of a few distonions and accomplishing the economic goal of allocating the investible resources to the most profitable firms. With respect to the significance of the positive coefficients for industries in the textile and wood sectors, it is unfortunately not clear what characteristics of these sectors determine their greater than average propensities to apply for and get loans. A greater understanding of these sectors might help to resolve this issue. The reason for the greater likelihood for entrepreneurs of African origin to apply for and obtain loans also remains an open question. It is hoped that more light might be shed on this issue after the second wave of interviews. Among the variables that did not show to be significant, the one referring to collateral may point to problems in the utilization of collateral either because of the lack of clarity of the titles to land and buildings or to the inability of the legal system to arrange for the transfer of these titles in the case of default. The generalized use of bank overdrafts in comparison to formal credit may point to the shon-term view adopted by the banking system as a result of the highly variable rates of inflation. This sbon-term approacb of the banks is completely the opposite of what bank- centred-fmancial-systems proponents would expect: the banks do not show any greater commitment to industry than could be expected from a stock exchange centred system. Leaving aside this unresolved issue, it might be suggested that greater price stability would lead to fmancial deepening. Along with the results presented above, it must be mentioned that no account was taken of the financial situation of firms. This shoncoming is derived from the fact that no plausible measures could be constructed with respect to shon and long term liabilities and the capital structure of the company. It is hoped that this situation may be remedied in the second round of interviews through a more direct approach to these questions. 75 References A1cerJof, George, 1970, -The Market for Lemons: Qualitative Uncertainty and the Market Mechanism," Quanerly Journal oj Economics 84: 488-500. Cameron, Rondo (ed.), 1972. Banking and Economic Development. New York: Oxford University Press. Bemanke. Ben S., 1983. "Non-Monetary Effects of the Financial Crisis in the Propagation of the Great Depression." American Economic Review 73: 257-276. Cosh. A., A. Hugbes and A. Singh. 1990, "Takeovers and Sbort-Termism in the U.K.: Analytical and Policy Issues in the U.K. Economy." London: Institute for Public Policy Research. Fazzri, S.M., R.G. Hubbard and B.C. Petersen, 1988, "Financing Constraints and Corporate Investment: Brookings Papers on Economic Activity: 141-195. Friedman. Milton and Anna Schwartz.. 1963, A Monetary History oj the United States: 1867-1960, Princeton: Princeton University press. Gerschenkron. Alexande.r. 1962. Economic Backwardness in Historical Perspective. Cambridge. Mass.: Harvard University Press. Gertler. Mark, 1988, "Financial Structure and Aggregate Economic Activity: an Overview: NBER Working Paper # 2559. Gurley. John and Edward Shaw, 1955, ·Financial Aspects of Economic Development." American Economic Review 45: 515-538. Keynes, J.M., 1936, The Genera/Theory of Employment Interest, and Money, New York: Harcourt, Brace and World, Inc., 1965. Mayer, Colin, 1988, "New Issues in Corporate Finance," European Economic Review 32: 1167-1189. Modigliani, Franco and Merton Miller, 1958, "The Cost of Capital, Corporation Finance and the Theory of Investment," American Economic Review 48: 261-297. McKinnon, Ronald I., 1973, Money and Capital in Economic Development, Washington, D.C.: The Brookings Institution. Myers, S.C., 1984, "The Capital Structure Puzzle,· Journal of Finance 39. Myers, S.C. and N.S. Majluf, 1984, "Corporate Financing and Investment Decisions When Firms Have Information that Investors Do Not Have: Journal of Financitll Economics 13: 184-221. Shaw, Edward S., 1973, Financial Deepening in Economic Development, New York: Oxford University Press. Stiglitz, J. and A. Weiss, 1981, "Credit Rationing in Markets with Imperfect Information." American Economic Review 71: 393-410. 76 7 Capacity utilisation and investment Jan Willem Gunning Takawira Mumvuma Marc Pomp 7.1 Introduction In this chapter we consider determinants of investment and capacity utilisation in the Zimbabwean manufacturing sector. The structure of the chapter is as follows. In Section 7.2 we consider determinants of capacity utilisation. Section 7.3 is devoted to an analysis of investment rates where we pay particular attention to rationing in the market for bank loans. Section 7.4 concludes. 7.2 Capacity utilisation Introduction Existing studies on capacity utilisation in the Zimbabwean manufacturing sector have concentrated mainly on the causes of capacity underutilisation and on identifying measures for rectifying the problem of excess capacity. In this section, however, in addition to a review of the literature on various causes of capacity underutilisation, we also make an attempt to determine the effect of some firm characteristics on capacity utilisation in the Zimbabwean manufacturing sector. This section is organised as follows. The first subsection looks at the significance of capacity utilisation with respect to economic growth and technical change plus a review of the literature. The survey evidence on capacity utilisation rates in the manufacturing sector is presented in the next subsection. We then present the model and the results of the regression analysis. The fmal subsection deals with the implications of the results. Capacity utilisation; why it matters The rate of capacity utilisation is of paramount importance in LDCS because increased usage of installed plant capacity means that existing fixed capital assets yield greater output without incurring any additional investment costs. A study by Ndulu (1990) stresses that the real rate of economic growth in a situation where capacity underutilisation prevails is the sum of capacity growth. which is driven by the investment rate, and growth in the rate of capacity utilisation. Ndulu points out that high rates of investment can co-exist with declining actual growth as long as the decline in capacity utilisation exceeds the increase in capacity growth. Hence when the installed capital stock is used less than fully then any sustained effort in increasing the rate of capacity utilisation will increase the economy's growth rate. On the other hand, capital stock jf not fully utilised has the potential to adversely affect the economy's rate of technical progress. Merhav (1971) pointed out that jf a firm has a high rate of capacity underutilisation, there will be no incentives for it to introduce any new technical innovations that require additional investments, for such additions promise to cause even greater underutilisation of capital stock. Hence the pbysical lifetime of installed production capacities will be unnecessarily prolonged thereby hindering the renewal of obsolete equipment and the introduction of new production techniques. A recent study by Kilundwa (1993) revealed that underutilisation of capacity has been the main cause of manufacturing output decline, tecbnical regress and falling productivity 77 growth rates in the Tanzanian manufacturing industry. In all except five of the manufac- turing sub-sectors studied, a low capacity utilisation level was found to have been accom- panied by a low rate of technical progress and by productivity decline. The effect of a unit increase in capacity utilisation on production costs showed a negative relationship between total costs and capacity utilisation, an indication that confirms allegations that capacity underutilisation increases unit costs of production. These results are of great relevance to Zimbabwe. Being an import dependent and capital scarce country with a persistent foreign exchange constraint, the waste of resources implicit in the underutilisation of the country's capital stock in the manufacturing industry will be detrimental to the current rehabilitating and restructuring of the economy under the ongoing economic reform programme. Efficient use of installed capacity, on the other hand, will go a long way towards the creation of more jobs resulting from increased growth without resorting to costly capital deepening investment programmes, thus optimising on the use of scarce foreign exchange earnings and borrowed foreign savings. Hence an awareness of how to make efficient use of existing capital resources through proper adjustment of policy variables must be part of the current reform programme. Determinants of capacity utilisation The factors that have been held primarily responsible for high levels of capacity underutilisation in LDCS can be classified into two groups, namely intended and unintended factors. According to Marris' rhythmic input-price model, regular rhythmic changes in input prices that are anticipated at the time of investment justify building intentionally idle capital. The model predicts that a certain amount of capacity underutilisation will always exist, depending on costs and technical factors peculiar to each firm. For example, the more capital intensive is the production process, the more important are capital costs and the greater is the incentive to economise on them through higher utilisation. As another example, if the capital stock is old and frequently breaks down, then low utilisation rates may be tolerated if this reduces costs of repair and equipment maintenance. Alternatively, unfamiliarity with the purchased new machinery and their SUbjection to stress for the first time can lead also to frequent breakdowns and repair-work can be prolonged in instances were outside help is called for. Under such circumstances low utilisation rates are unavoidable as local technicians and equipment maintenance staff take some time to acquaint themselves with the equipment. The last two issues raised are very relevant in analyzing capacity utilisation levels in the Zimbabwean manufacturing sector. The rhythmic cost model points to a relationship between underutilisation of the capital stock in LDCS and distortions in factor markets. For example, minimum wage legislation policies have tended to raise the relative price of labour. Therefore for those industrialists who use labour intensive production techniques low utilisation rates will be justified in order to avoid high labour costs on night shift workers. On the other end, the price of capital has been kept low over the years by interest rates which were pegged below the equilibrium market rate and lower tariffs on imponed capital goods which gives manufacturers very little incentive to utilise it fully. However in many LDCS, Zimbabwe included, this has started or has already fallen off. The deregulation of both the labour and financial markets implies that wages are now being determined through collective bargaining without government interference and interest rates are being left to be determined by the forces of supply and demand. Whilst the easing of labour regulations might be a welcome move, financial deregulation might further constrain capacity utilisation if the cost of obtaining working 78 capital becomes too prohibitive. . . The model was found to be relevant in explaining the factors behmd the hIgh prevalence of intentional idle capacity in LDCS (see Winston (1974), Clague (1976)). Although the rhythmic cost model adequately explains why some manufacturers choose not to fully utilise their capital stock, recent studies have tended to concentrate on supply side variables and the effects of market structure and industrial organisation factors in explaining unintended excess capacity in the manufacturing sector (Teitel (1993), Rajasopalan (1992), Kibria el al. 1986, UNIDO (1986), Ndulu (1986), Wange (1977». The most cited causes are demand deficiencies, scarcity of skilled manpower, lack of working capital, the age of the firm, shortage of imported raw materials, completion from imported finished goods and lack of vital complementary inputs such as energy due to electricity blackouts or load-shedding. . The economics of production tells us that under normal production conditions, an increase in output in the short run does not only involve an increase in the variable input, labour, but also a rise in the demand for working capital and intermediate inputs either domestically sourced or imported. The non-availability of these complementary inputs will severely constrain the firm's ability to realise its potential output capacity. Given the fact that the growth of foreign exchange receipts is insufficient to meet the rapidly growing import demands generated by economic growth, foreign exchange availability is a major determinant of capacity utilisation in LDCS. There is evidence that the shortage of inter- mediate imports, their availability which is determined by foreign exchange availability, is the most important single constraint to full capacity utilisation in the manufacturing sector in Sub-Saharan Africa (Ndulu (1990, 1986), World Bank (1991». Mlambo (1993) found that excess capacity existed in the Zimbabwean manufacturing industry. The lowest rate of capacity utilisation was in metals and the highest in textiles and clothing as well as in wood and paper. The rates of the latter sectors exceeded unity, a phenomenon that can be explained by the short-sighted investment behaviour of expanding firms (see Morrison (1985». The study attributes low rates of utilisation to foreign exchange shortages which led to a contraction of imports, especially of spare parts and raw materials. The result of the foreign exchange bottleneck is that production will be constrained below the full utilisation level. Therefore, as long as imports remain imperfect substitutes for domestic goods, a squeeze on the capacity to import will result in a decline in the capacity utilisation rate. The same effect on capacity utilisation will arise if there are some supply constraints in other complementary production factors. By the time the RPED survey was carried out, however, 50% of the country's imports were free of the licensing restrictions as a result of the expansion of the Export Retention Scheme (ERS) and the Open General Import Licence (OOJL) list which could simultaneously have eased both the imported raw materials and foreign exchange problems. Capacity utiUsation rates in Zimbabwe Capacity utilisation rates estimated from the survey data indicate that all the sectors covered by the study were operating below capacity. with the mean utilisation rate for the four subsectors estimated to be 65 % when the mode of operation is one shift. Highest rates of capacity utilisation were realised in the food and textiles sectors with an average utilisation rate of 69 %. The metal sector registered the lowest rate of capacity utilisation (59 %) followed by wood (60%). When capacity utilisation is broken down by firmsize, a distinct pattern emerges which shows that the rate of capacity utilisation increases as firmsize increases. Thus the lowest capacity utilisation rate of 55 % was estimated for small firms and 79 the highest rate of 76% for the very large firms. The utilisation rates for the medium and large firms were estimated to be 63 % and 72 % respectivel y. Regression analysis This section is devoted to the analysis of the effect of various firm characteristics on the level of capacity utilisation in the Zimbabwean manufacturing sector. The model that will be employed borrows from two studies. The first study is the one carried out by Winston (1971) to examine the structure of excess industrial capacity in West Pakistan and the second one is the study by Kibria et al. (1986) who analyze the causes of capacity underutilisation in jute spinning mills in Bangladesh. Winston's model is compatible with both Marris's rhythmic cost model and the market structure and industrial organisation hypotheses. In addition to the explanatory variables used in these two studies, we will add some additional variables deemed relevant because of policies specific to the Zimbabwean economy at the time of the survey. The model takes the following form: Log(cu)= 80 + a1Log(sal) + azLog(xpe) + a3Log(K/L) + a.a[Log(KILW + a,Log(fge) + ~[Log(fgeW + a7Log(ska)+ a.(ddem) + aq(ele) Here cu is the percentage capacity utilisation rate of the firm measured by the ratio of actual to potential output. The D variables are sectoral dummies. Firmsize (sal) is measured by the sales revenue of each firm. The variable is expected to capture the influence of scale effects on capacity utilisation. As the size of the firm increases it is expected that a division of labour characterised by specialised management functions, meant to enhance production efficiency at the shop floor, emerges. Besides management scale effects, technological scale effects also increase with firm size. Furthermore, it is expected that large firms have direct commercial, management and technical contact with firms abroad which permits them access to information and latest production and management techniques, enabling them to achieve lower unit costs. Hence we can hypothesise that the larger the firm, the higher should be the rate of capacity utilisation. The exports to sales ratio (xpe) variable is expected to be positively related to the rate of capacity utilisation. The effect of exports on capacity utilisation operates either through demand expansion or investment and supply incentives (see Winston (1971». In our case it will capture the effects of the government's expon-Ied growth strategy under the current economic reforms, which favours exponers by awarding them investment incentives through special access to foreign exchange in order to facilitate the imponation of intermediate inputs through the various expon incentive schemes. Exponing firms are therefore less constrained by the foreign exchange bottleneck and may be able to realise higher capacity utilisation. The capital intensity variable (k:pe) is measured by the capital-labour ratio. As argued above, the expected sign for this variable is positive. The capital stock is measured by the firm's sales value of equipment which takes into account both depreciation and obsolescence. The firm age variable (fge) is included as a continuous variable measured by the firm's age. The squared logarithmic variable in (fge) is introduced to examine whether firmage has a linear or non-linear effect on capacity utilisation. The age of the firm is expected to be positively correlated with the rate of capacity utilisation (see Kibria (1986». The expectation 80 is that a firm which has survived in the market for a long period of time has acquired a wealth of technical experience enabling the efficient running of the production process, whilst firms still in their infancy grapple with a lot of production bottlenecks due to inexperience. Electricity is a very important source of energy for the manufacturing sector, and power cuts may be a cause of low capacity utilization. For example, Kalindwa (1993) found electricity power cuts caused by deteriorated electric power distribution infrastructure to have negatively influenced the rate of capacity utilisation in the Tanzanian manufacturing sector. A problematic supply of electricity is measured by a dummy (ele) taking on a value of one if electricity was mentioned as the firm's biggest infrastructural problem and zero otherwise. The fall in demand for manufactured products due to falling disposable income resulted in the buildup of finished stocks putting pressure on firms to cut down on their production levels. The period in question also coincided with a world recession which contracted the global demand for finished manufactured products. The questionnaire ranks on a scale of one to five indicating whether lack of demand was a constraint to flI'Ill expansion, with "1" indicating no problem and "5" a serious problem. The demand variable is variable is defined as a dummy taking the value of unity jf the respondent's score is greater or equal to the mean score of all respondents, and zero otherwise. Skills availability (ska) is measured by the ratio of skilled personnel to all employees. Here we hypothesise that the more skilled personnel a firm has the higher is the rate of labour productivity and hence the rate of capacity utilisation. We define skilled personnel as all employees minus "other production workers", apprentices and support staff. This variable is expected to capture the availability or non·availability of the requisite skills in the manufacturing sector. A negative correlation between capacity utilisation and the skills variable can be expected when there is a serious shortage of skilled supervisory, production, technical and management personnel (see Winston (1971». Manufacturing firms in Zimbabwe are known to depend heavily on the use of imported intermediate inputs, a phenomenon that can be attributed to the import substitution in- dustrialisation process the country has been following until recently. Although we expect that with liberalisation of the economy the problem of .lack of imported intermediate inputs should wane off, in the Zimbabwean case the simultaneous depreciation of the dollar meant to restore export competitiveness and rising import tariffs should have pushed up the import costs of raw materials and spare parts. A negative correlation between the rate of capacity utilisation and dependence on imported intermediate inputs is therefore postulated. The imported inputs variable (imp) is measured by a dummy taking a value of unity if the firm imports some of its raw materials and zero otherwise. Finally, one of the most important aspect of any economic reform programme is trade liberalisation which entails the removal of all import restrictions. By the time we carried out our study, Zimbabwe had already started implementing this. The effect of the removal of these restrictions have been to cause an influx, on the domestic market, of finished consumer goods imported under OGIL and the Export Retention Scheme (ERS) facilities. Therefore it is expected that completion from imported goods could have eroded a substantial share of the domestic market of locally manufactured goods. Hence a negative association between the rate of capacity utilisation and competing imports is hypothesised. The competing imports variable is defined in a similar fashion as the demand variable (see above): the questionnaire ranks on a scale of one to five indicating whether competing imports were a constraint to flflD expansion, with "I" indicating no problem and "5" a serious problem. The competing imports variable (cim) is defined by a dummy taking on the value of unity if the respon- 81 dent's score is greater or equal to the mean of the competing impons variable and zero otherwise. Results The regression results of the model are presented in Table 7.1 below. Table 7.1: Regression results variable est.coet t-value constant 2.1972 -0.2897 exports -0.0009 -0.2532 firmsize 0.0095 3.2129 d-demand -0.2115 -2.5305 electricity -0.0028 -0.2103 (KILl 0.1295 1.7301 (K/LI 2 -0.0009 -1.5405 firmage 0.2323 1.4165 (firmage)2 -0.0039 -1.0855 skills-av -0.0061 -, .6560 imported-rw -0.1619 -1.5272 competing-imp -0.1862 ·1.0130 R2 = 0.271 adj-R 2 = 0.199 D.W. = 1.738 n = 123 Six explanatory variables are statistically significant. The regression coefficients of firm size, demand deficiency. capital intensity. firmage and imported materials have the expected signs. The skills variable, while statistically significant, is negatively related with the rate of capacity utilisation. The firmsize and demand deficiency variables are statistically significant at the 19'0 level of significance and the capital intensity and skills variables are statically significant at the 59'0 level, whilst the fmnage and imported raw materials are significant at the t09'O level of significance. This confirms our earlier hypotheses. The coefficients of electricity and competing impons dummies are properly signed but insignificant. All the four sectoral dummies are not statistically different from zero implying that there is no significant inter-sectoral variation for any coefficient of the explanatory variables in the model. Contrary to what we expected the export regression coefficient appears with a negative sign and it is not statistically different from zero. This may be due to the fact that larger firms tend to export a larger share of their output. so that the effect of export share is included in the coefficient for the size variable. The electricity variable also proved to be insignificant although negatively related to the rate of capacity utilisation. This may be due to the fact that by the time we carried out the survey electricity supply had been somewhat normalised (although the cost of electricity had risen sharply)_ The insignificance of the import competing variable can be explained in two ways. Firstly. the effects of this variable could already have been captured by the demand deficiency variable. Alternatively. domestic manufacturers must have been forced to adjust the price of their products downwards in order to enhance their competitive edge against the impons. 82 Conclusion We have found that the substantial differences in capacity utilisation rates among firms within the manufacturing industry are correlated with differences in firm size, demand conditions, capital intensity, skills availability and ability to source imported intermediate production inputs. If ability to export is a function of size, because of the advantages already mentioned, then the higher rates of capacity utilisation associated with large firms occurred partly because a greater share of their output managed to find an outlet into foreign markets. The lack of imported intermediate inputs also constitutes a major impedient towards the realisation of full plant capacity utilisation mainly due to the. constraining effects of government's trade policies. This suggests that these policies should be adjusted in a way that facilitates easy access to imported production inputs at affordable prices in order to ensure higher capacity util isation rates of the existing stock of capital in the manufacturing sector. 7.3 Investment The survey provides considerable information on investment, but the data differ in coverage, both over time and over categories of investment. Firms were asked about: start-up investment; the value of "major investments" in three selected years (1981, 1986 and 1991), when applicable, that is when the firm already existed in those years; and investment in 1992, but restricted to plant and equipment. In addition, firms were asked about the most recent "major" investment in plant and equipment and, finally, for each of three categories (land, building and equipment), about the value of the most recent investment, provided this took place in the last five years. l The design of the survey questions somewhat limits the scope for investment analysis. First, there is insufficient information on pre-1991 investment: years other than 1981 and 1986 and the firm's start-up year are covered only for firms which have not invested since 1989 and then data are available only as responses to the questions about the most recent investments. Hence, except for 1981 and 1986, there may be two biases: firms with low investment rates will be overrepresented and investments prior to the most recent one will be missed. To some extent these disadvantages can be overcome by focusing on the last few years. In this chapter the analysis is based on investment in the period 1991-1993. In- vestment rates calculated for this period will not suffer from incomplete recording except for 1993. The data for that year will be biased downwards since the survey was conducted in June so that investments in the second half of the year will not be recorded. Table 7.2: Parcentage of firms investing. 1991·93 food woOd textile metal iii 1991 58.5 36.4 50.0 57.7 53.6 1992 56.1 27.3 37.5 42.3 42.8 1993 17.1 27.3 15.6 12.7 16.3 1991·3 87.8 63.6 75.0 76.1 77.1 In the Wave I Zimbabwe questionnaire the investment quClltiOlll can be found on pp. 2, 3, 10, 15 and 23. 83 Table 7.3: Frequency distribution of investment rates investment rate N <0 10 0 38 1-10% 45 11-20% 25 21-30% 22 31-40% 8 41-50% 4 51-60% 4 61-70% 2 71-80% 2 81-90% 2 91-150% 4 >150% 11 Total 177 Tllble 7.4: Mean investment rates by sector, in % sector rllte (%\ N Food 27.3 34 Wood 10.2 22 Textiles 37.4 68 Metal 42.3 30 All sectors 32.2 154 Note: excluding firms stating negative investment rates Tabla 7.5: Meen Investment rates by location. In % location rate(%) N Harare 32.1 83 Bulawayo 19.1 44 Other 54.0 27 All sectors 32.2 154 Note: excluding firms stating negative investment rates While the chosen window of observation covers only two and a half years, Table 7.2 shows that most firms (77.1 %) invested in this period. Indeed. both for 1991 and for 1992 more than half of the firms reported having invested. There are- some differences between sectors (the wood sector invested less than other sectors in 1991 but a remarkably high number of wood firms invested in the first half of 1993) but for the period as a whole these differences are minor. . Not only did most firms continue to invest in the last few years (which in Zimbabwe were marked both by a severe drought and by the introduction of the structural adjustment program. ESAP), but investments were substantial relative to the opening value of the capital stock. Of the 177 firms for which we can calculate investment rates, a close to one quarter 84 had rates for this short period in excess of 30% (Table 7.3). The mean rate was 32%=. There are marked differences between sectors (Table 7.4): investment rates in wood firms were low (10%) while in metaJ and textiles they were rather high (about 40%). Recall from chapter 3 that the textile sector has also registered the highest growth rates. There is also a difference between locations: in recent years investment rates have been much higher in Harare than in Bulawayo, but even higher outside these two main cities (Table 7.5). It should be noted that the investment rates are unweighted averages over firms. If we use shares in the capital stock as weights, investment rates are substantially lower, since large firms tend to invest less. The rate for the four sectors combined then falls to 11.5% for the two year period, probably sufficient to maintain the real value of the capital stock, but not representing substantiaJ net investment. A notable result is that the rate in the textile sector is considerably higher (20%). The high investment rates in Zimbabwe contrast sharply with the RPED findings for Ghana: in Ghana investment virtually collapsed in the period of structural adjustment. Table 7.6: Obstacles to firm expansion: number of cases where item constitutes more then a moderate obstacle type of obstecle number of cases Credit 97 Lack of demand 80 Infrastructure (including telecom) 74 Forex controls 72 Utility prices 69 Business support ser- 54 vices Taxes 50 Obtaining licenses 39 Investment benefits 38 Import competition 35 labour regulations 31 Ownership regulations 23 Government restrictions 13 Location regulations 12 Price controls 3 Other 53 of which: · High interest rates 22 · Cost and availability of raw materials 12 Obstacles to expansion Perceived obstacles to expansion are likely determinants of investment rates. The survey collected data on obstacles as perceived by managers, Respondents were asked to rate items on a list of possible obstacles to firm expansion on a 1 to 5 scale, 1 denoting 'no problem' and 5 a 'severe problem', The responses are summarised in Table 7,6 for all cases in which an obstacle received a score of 3, 4 or 5, Credit, lack of demand and infrastructure problems were listed most often as problems. In the econometric analysis we will focus on "credit", the obstacle mentioned most often. :2 This excludes the 17 rums for which the ca1cu1atecl rate waa negative. 85 Econometric analysis of investment rates In this section we will attempt to explain differences in investment rates on the basis of various characteristics of firms. Since the data consist of a single cross-section, we will not be able to analyze the effect of changes in prices and interest rates on investment. 3 Further- more, fixed effects may be involved and once the data from the second round are available we will be able to control for this. Since we can only measure gross investment (a non-ne- gative variable) we have to take into account that this variable is truncated. We therefore use Tobit estimation. The dependent variable is the investment rate for the period 1991-1993. This rate is constructed as follows. First, all investment data for this period 4 are expressed in prices of 1993, using the CPI. Next, we allow for the substantial wedge between buying and selling prices of capital goods, assuming that the market value of new machinery falls immediately after it has been purchased. $ This is due to transaction costs, and to informational asym- metries which make the quality of second hand equipment uncertain6 · Finally, a 10% annual depreciation rate is used to arrive at the stock remaining in 1993. This stock is sub tracted from the reponed sales value in 1993 of the total stock of equipment to arrive at an estimate of the opening value of the capital stock, i.e. the value (in 1993) of pre-1991 investment. The investment rate is then simply the ratio of the stock corresponding to the 1991-1993 investments and the opening capital stock. In formula: where: = ratio of sales and replacement value for new equipment (40%) = annual depreciation factor = investment in year t in 1993 prices = reponed sales value of the 1993 capital stock 3 This implies that we cannot bue our anaIy.ia on the recent developmc:nta in theoretical and empirical raea.rch which focuaea on the intcrtcmporal upccta of investment behaviour (ace Chirinko, 1993). For an anaIy.ia of time eeriel (198()..I987) of Zimbabwean invatment data, ace Dailami and Walton (1989). " Note that theee do not alway. COmIIpOIId to the aame concepti. 1.1.1. the 1992 figu~ ia for invatment in buildings and equipment, wbile for 1991 ~ ia no IUCh RIItriction. 5 The: IUrvey colleclod data on the value of the capitailltOck. both in terms of the replacement value (replacing cummt equipment with aimiJar, but new equipment) and the eales value. For fU1lll for which the entire capita.lltock ia new (that is for recent 1t.arterI) the ratio between thcac values gives a rough mc:.uu~ of the wedge. This calculation IUggeatI that for new equipment the IaIea value ia 42$ of the purchuc value. Hcnc:c. as for Ghana, ~ ia a IUbItantial wedle. 6 What we have in mind be~ arc infonnational uymrnctriea of the type that have been analyzed in Akcrlof (1970). 86 We use the following explanatory variables: food, wood, metal sector dummies; the implied reference sector is textiles. location a dummy variable indicating that the firm is in the Bula- wayo area. profitability profits divided by sales; profits either as reported or as constructed from income and cost data expon share exports as a percentage of sales loan rationed a dummy variable indicating that the firm (a) at least once applied for a bank loan but did not get it, or (b) never applied for a bank loan because it found it unlikely that it would get one. race a dummy variable indicating that the owner is African age (squared) age of the firm employment (squared) the number of employees (permanent plus temporary workers, but excluding seasonal workers) entrepreneur a dummy variable indicating whether the firm is (partly) managed by its owner(s) labour intensity employment divided by the sales value of equipment. Several of these variables relate to possible constraints on the financing of investment. Current profitability7 is our best indicator for a firm's ability to self-finance investment. However, this indicator is flawed in more than one sense. First, current profitability may simply be a good proxy for expected future profitability. It will then measure the attrac- tiveness of investment in addition to the ability to finance it. Secondly, access to bank: finance is likely to be an increasing function of ~e fum's profitability. Hence there may be two effects: the direct effect of profitability on investment via the ability to self-finance and the indirect .effect through easier access to bank finance. However, since we are able to measure access directly we can to some extent control for this. Hence we expect profitability to have a positive effect on investment. In view of our results it is important to note that the direct effect of profitability is the main channel envisaged in the design of structural adjustment programmes, including ESAP. Reform, through changes in relative prices. affects profitability differentially between sectors. Adjustment to the new relative prices is then facilitated by increased profitability of the firms in expanding sectors, making expansion both more attractive and easier to finance. Conversely, in contracting sectors reform will reduce profitability. thereby making investment both less attractive and more difficult to finance. To the extent that firms rely on external finance banks' discretionary decisions could substitute for such induced changes in self-fmance. However, this is a poor substitute at best 7 N ate that profitability is IJ'ICII.IUJed as the ratio of profata and lila retber than as profits ~lative to the value of the c:apita1 Itock. This is to avoid apurioul eom:lation: the dependent variable already includca the c:apit.al Itock in the denominator. 87 since banks have a natural wish to continue lending to established clients. At worst, bank finance thereby puts a brake on adjustment, keeping firms in contracting sectors afloat. The "loan rationed" variable captures two aspects of rationing. First, since we know whether firms have applied for bank loans and been refused we can measure rationing in the sense of a bank refusing to agree to a loan application. But secondly, to the extent that such rationing is based on criteria which are (imperfectly) known to the applicant, there may also be selfselection: firms may have decided not to apply for a loan because they expected to be turned down. In the RPED survey such cases can be identified. We are therefore able to use a more satisfactory definition of rationing than is commonly used: our concept includes self- selection. Profits can be measured in two ways: firms were asked about profits directly but also about various income and cost items. This allows us to construct a second profits variable. B These two procedures give widely different results. The correlation coefficient for the two variables is only .49 and the correlation between the two profits rates is only .19. Given the importance of establishing the role of profits we have experimented with both variables. However, the constructed profit rate did not yield plausible results. Therefore we only repon findings for the reponed profit rate. 9 Employment is included as a proxy for the size of the firm. The remaining explanatory variables are straightforward. Life cycle effects can be captured by including the age of the firm and age squared. The race of the entrepreneur is expected to have no effect once we control for age and size of the finn and for access to externaJ finance. It is important to explore whether this is indeed the case since the question whether the banking system is biased against indigenous enterprises is currently an important politicaJ issue in Zimbabwe. Differences in labour intensity may explain differences in investment rates since a Jow labour intensity may reflect a high prior rate of investment and hence a lower need for current investment. FinaJly, we aJlow for location to affect the investment rate. We begin by a Tobit-estimation that includes the whole Jist of explanatory variables. As will be argued below. this is not a panicularly convincing specification. However. it provides a useful point of departure. Table 7.8 shows the estimation results. The only variable that is significant is the sector dummy for wood: firms in this sector repon significantly lower investment than the reference sector (textiles). Of the remaining variables, we find positive but insignificant correlations between profitability and investment. and between race and investment, and a negative (but again insignificant) correlation between loan rationing and investment. The interpretation of the last finding is that the selection effect (investing firms need more finance so are more likely to be rationed) is outweighed by the rationing effect (rationing limits investment). We also find a negative coefficient for the entrepreneuriaJ dummy: apparently, firms run by the owner(s) do not invest more than other firms. These firms are generally small, but the negative coefficient is not due to a size effect since size is included as a separate variable (proxied by a Profits an: reported after dcpn:ciation and befORl tu.. ProfJta can be CItimated a uJea (inelu.ive of income from trading and ICrviccs) minul the COlt of raw materials, the wage bill, the COlt of mit, electricity, water, telephone, liquid fuel, solid fuel and ga, and the COlt of promotion and advertiaing. (Zimbabwe question- IIIIi.n: pp. 9-10, qUCBtions 7, 8, 9, 13, 16,23 and 25.) \I The constNctcd profit rate was insigniflClUll. in aU equatiOll8 reported below, except for the two-limit Tobit model in table 7.9, where the coeffICient was negative and signiflClUll.. The difference between the two profit rates is due to depreciation, unrecorded COlt items, and reporting efron. 88 employment). !O Table 7.8: Tobit estimation results independent estimated t -statistic variable coefficient constant 259.29 .37 food -.07 .34 wood -.52 2.07 metal -.16 .75 location -.22 1.24 profitability .43 1.14 export share .0022 .42 loan rationing -.30 1.38 race .36 1.55 age -.27 .38 age squared -.000070 .39 employment .00015 .31 employment squared -.0 .17 entrepreneurial firm -.085 .47 labour intensity -8.11 .10 sigma .80 13.41 Notes: dependent variable: investment rate 1991-93 127 observations (94 positive) R2 for positive, non-rationed observations = ., 8 As already indicated, the specification that generates these results is questionable. In particular, while it is not uncommon to include variables such as "profitability" and "loan rationing" as explanatory variables in a single equation model, this is difficult to justify on theoretical grounds. For if external finance is subject to rationing then one should distinguish between a regime in which the rationing constraint is binding, and one in which it is not. The determinants of the investment rate would then differ between the two regimes. In the rationed regime the bank's lending decision effectively determines the investment rate, while in the non-rationed regime investment is determined by the firm itself. In that case the lending decision may depend on the firms' reputation, which may be correlated with its age and also with race (e.g., if black entrepreneurs belong to different social networks than loan officers). Such variables may not playa role (or a different role) in the decision to invest. The appropriate econometric model for such a micro-structure is a type of switching regression model, adjusted to allow for the fact that investment is non-negative. If the error terms in the rationing equation and the investment equation are uncorrelated, this yields a two-limit tobit model. II However, there is reason to believe that the error terms are correlated, since a firm that has a relatively high investment demand (Le., a positive shock in the investment equation) may thereby give a signal to the bank (e.g., it may signal that it is a risk lover). In addition, there may be omitted variables in both the investment and the rationing equation such as personal characteristics of the firm managers. This would also give rise to a correlation between the two disturbance terms. 10 Uaing sales rather than employment as · proxy for lizc docs not affect this mrult. \I Sec Pudney (1989). p. 275. 89 In this case, the appropriate econometric model is a simultaneous system of two Tobit equations. Such a model is described in Pudney (1989) in the context of mortage demand. Pudney also presents the likelihood function for this model. An attempt has been made to apply this model, but probably due to the small number of observations on rationed firms with positive investment demand, it proved impossible to obtain convergence. 1~ Therefore, we are forced to use an imperfect solution to the simultaneity problem, namely the two-limit Tobit model already referred to. In this case the lower limit to investment is zero, as in the standard Tobit formulation, while the upper limit is the rationing bound, which itself is variable. The likelihood function for this model has three parts, one corresponding to non- rationed observations with zero investment (in our case 26 observations), one corresponding to non-rationed observations with positive investment (80 observations), and one correspon- ding to rationed observations (21 observations). . Table 7.9 shows the results for a parsimonious specification, including only the profit rate, the race dummy sector dummies, and a location dummy as explanatory variables. All other variables where found to be insignificant at the 10% level. The coefficient on profitability is (almost) significant at the 10% level, and it somewhat higher than the estimate for the whole sample as reponed in Table 7.7. The race dummy indicates that African owners tend to invest more, even after accounting for firm characteristics. In order to get an idea of the magnitude of the effect of profitability on investment according to this specification, we have calculated the effect of a doubling in profitability on the investment rate at the mean of the explanatory variables, on the assumption that the firm is not loan constrained. 13 The result is an increase in the investment rate from .43 to .46, an increase of about 7%. Similar calculations for the effect of the race dummy indicates that being black increases the investment rate from .43 to .51, an increase of 23%. Table 7.9: Two-limlt tobit estimation results independent estimated t-statistie: yariable coefficient constant .38 2.51 food -.18 .81 wood -.49 1.89 metal -.23 1.05 BulawlYo -.30 1.63 profitability .46 , .19 race .52 2.46 sigma .84 12.58 Notes: dependent variable: investment rate 1991-93 1 27 observations R2 for positive. non-rationed observations = .26 Although the two-limit tobit model does more justice to the micro-economics of in- vestment decisions than the simple tobit in Table 7.7. a consequence is that the loan rationing variable has now been dropped from the list of explanatory variables. However. it is stm possible to assess the effect of rationing. namely by simulating investment rates for 12 Reducing the number of exp1anatory variablea did not IOlve thiI problem. 13 The appropriate fonnula for calculating the expected inveatment rue for theac observations iJ given in Greene (1993), p. 694. 90 the constrained observations on the assumption that these are not constrained (Le., that the probability of being constrained is zero). These simulated investment rates may then be compared to the actual investment rates (which, by assumption, are equal to the level of the constraint), and the difference is attributed to the loan constraint. Table 7.10 shows the results of this exercise. The results are striking: relaxing the constraint leads to a threefold increase in the average investment rate of the constrained firms. Table 7.' 0: Simulated effect of a binding loan constraint on the investment rate mean standard deviation Investment rate (actual) .18 .38 Investment rate (simulated) .61 .23 Investment ratio in whole .28 .70 sample of 127 firms Note: number of observations = 21 constrained firms If these results are to be believed, then the effect of loan constraints on the investment decisions of firms that are constrained is severe. It is therefore important to explore in what way the constrained firms differ from non-constrained enterprises. To this end, we estimate a probit equation with a dummy variable indicating whether the firm is loan constrained as the dependent variable. The probit initially includes the whole list of explanatory variables, but Table 7.11 shows results for a subset of variables of interest; all other variables were insignificant. Firms in the food sector are somewhat less likely to be rationed than firms in the other sectors, but not significantly so. At the sample mean the difference in probabilities is 11 %. The profitability variable has a positive coefficient, but it is very small: at the sample mean, the estimated coefficient implies that a doubling of the profit rate raises the probability of being constrained by onJy 0.4 percent. An interesting result is that a high expon share reduces the probability of being loan constrl",ined. Here the effect is much larger: a doubling of the expon share implies a fall in the probability of being constrained by 5.0 percent. Finally, being black increases the chance of being loan constrained by about 8.3 percent. Table 7.'1: Probit Estimation Results independent estimatid t-Itatistic variable coefficient constant -.77 2.95 food -.46 1.19 wood -.16 .37 metal -.18 .49 profitability .18 .31 export share -.02 , .57 race .36 1.09 Note: dependent variable: rationing dummy. 1 if rationed 127 observations (21 rationed) 7.4 Conclusion The survey data show that high investment rates were maintained in Zimbabwe in the 1991- 93 period. This is remarkable since the literature suggests that structura1 adjustment may 91 well have a negative effect on investment, particularly if reforms increase policy uncer- tainty.14 In addition, Zimbabwe was in this period affected by a very serious drought. We have found that capacity utilisation is strongly correlated with firm size. The analysis of investment indicated that profitability positively affects investment. This is of some imponance since the main channel through which adjustment policies can induce firms in different sectors to grow or contract is through their effect on profits. We also found evidence that investment would be considerably higher in the absence of rationing in the market for bank loans. There is also evidence that black entrepreneurs invest more, even though they are more likely to be loan rationed. Finally, banks appear to be supporting exporting firms: the more a firm exports the more likely it is to get an application for a bank loan approved. 14 Sec, for curnplc, Collier and Gunning (1992, 1994). 92 References Acheson K. et al., 1974, 'Capital Utilisation in Economic Development: a Comment', The Economic Journal, 84: 159-170. Akeriof, G., 1970, 'The Market for 'Lemons': Quality, Uncertainty and the MarketMechanism', Quanerly Journal 0/ Economics, 84: 488-500. Berndt E.R et al., 1986, 'Measuring and accessing capacity utilisation in the manufacturing sectors of nine OECD countries', European Economic Review, North Holland, (30): 961-989. Chirinko, R.S., 1993, 'Business Fixed Investment Spending: a Critical Survey of Modeling Strategies, Empirical Results, and Policy Implications" Journal of Economic Literature, 31: 1875-1911. Claugue C., 1976, 'Capital Utilisation',Journal o/Development Economics, 3: 277-287. Collier, P. and J.W. Gunning, 1992, 'Aid and Exchange Rate Adjustment in African Trade Liberalisations', Economic Journal, vol. 102, pp. 925-939. Collier, P. and J .W. Gunning, 1994, 'Portfolio Responses to Trade Policy Incredibility'. paper presented at the Royal Ecomomic Society Conference, University of Exeter. Dailami, M. and M. Walton, August 1989, 'Private Investment, Government Policy, and Foreign Capital in Zimbabwe', Country Economics Department Working Paper WPS 248. Greene, W.H., 1993, Econometric Analysis, New York: Macmillan, second edition. Harris R., et al., 1985, 'The measurement of capacity utilisation', Applied Economics, 17: 849-866. Kibria M.G et al., 1986, 'Life-time patterns of capacity utilisation by manufacturing firms in an LDe: A study of jute spinning in Bangladesh" Indian Economic Review, XXI(I): 1-19. Kulindwa K., 1993, Input substitution, Technical change, Productivity and Capacity Utilisation in the Tanzanian manufacturing sector: A Disequilibrium Factor Demand Model', Memorandum No. 189 Department of Economics. University of Goteborg, Sweden. Merhav M., 1970, 'Excess capacity-measurement, causes and uses, a case study of industry in Israel, Industrialisation and Produaivity Bulletin No.15. New York. United Nations. 93 Mlambo K., 1993, 'Total Factor Productivity Growth: An Analysis of Zimbabwe's manufacturing sector based on factor demand modelling', PhD thesis, Goteborg University, Sweden. Morrison C.J, 1985, 'On the economic interpretation and measurement of optimal capacity utilisation with anticipatory expectations', Review of Economic Studies: LII 295-310. Ndulu B.J., 1990, 'Growth and Adjustment in Sub-Saharan Africa'. Paper presented at the World Bank Africa Economic Issues Conference, Nairobi June 1990. Ndulu B.J., 1986, 'Investment, output growth and capacity utilisation in an African Economy: The Case of manufacturing sector in Tanzania', Eastern Africa Economic Review, 2(1): 14-30 Pudney, S., 1989, 'Modelling Individual Choice', The Econometrics of Corners. Kinks and Holes, Oxford: Blackwell. Rajasopalan S., 1992, 'Deterministic capacity expansion under deterioration Management Science', Journal of the Institute of Managemem Science, 38(4); 525-539. Teitel S., 1993, 'Industrial and Technological Development'. Inter-American Development Bank. Washington D.C. UNIDO, 1986b, 'The Manufacturing Sector in Zimbabwe', Vienna UNIDO, PPD/R.2. Wange S.M., 1977, 'Factors influencing capacity utiHsation in Tanzanian manufacturing, Imernarional Labour Review, US(1); 65-77. Winston G.C., 1971, 'Capital utilisation in economic development', The Economic Journal, 81: 37-60. Winston G.C, 1974, 'The theory of capital utilisation and idleness', The Journal of Economic Literature, 12; 1301-1320. World Bank, 1991, 'Macroeconomic structure and policy in Zimbabwe; Analysis and Empirical model (1965-1988),. 94 8 The labour market: wages and earnings Ann Velenchik This chapter draws on data from the labour markets section of the survey, panicularly the workers' survey, to explore a number of issues surrounding the determinants of earnings in the Zimbabwean manufacturing sector. The analysis considers patterns of differentiation across individuals, occupations, sectors and firm sizes, as well as exploring several aspects of earnings, promotion, and training within the firm. The chapter is structured as follows. Summary statistics for the entire sample are presented in Section 8.1. Estimates of a basic human capital earnings function, and attempts to correctly model the returns to schooling are contained in Section 8.2. Section 8.3 expands that analysis with a very brief discussion of issues of race and gender discrimination. Sections 8.4 and 8.5 provide detailed examinations of the relationship between wages and firm size. The first section uses a single labour market approach to explore a number of market clearing and non-market clearing explanations for the premium associated with employment in larger firms. The second section considers this premium in the context of a dual or segmented labour markets approach, in which the process of wage determination, panicularly the return to human capital, is allowed to vary by firm size. The focus moves from inter-firm differences to intra-firm issues in Section 8.4, which explores internal wage structures, internal promotion, and individual wage histories. Section 7 summarizes the results of the analysis and attempts to draw some conclusions about the nature of earnings determination and the functioning of the labour market. It should be noted that many of the results are preliminary, in the sense that further research may allow the use of more sophisticated econometric techniques and more creative incorporation of data from other section of the survey. While this is the Final Report for the first wave, it is not the last word on these issues. 1 8.1 Overview and summary statistics The dataset includes a sample of 1717 workers. A maximum of ten workers were intervie- wed in each fmn (fewer than ten in fl11llS with fewer than ten employees). The sample was drawn according to occupational categories in proportion to each occupation's weight in the firm's total labour force. A smaller sample of 1609 workers will be used for all of the analysis in this chapter. Missing data needed for the testing of some hypotheses resulted in the elimination of 108 observations.2 This was especially important because the discussion of the employer size effect requires hypothesis tests involving the stepwise addition of variables to the earnings function. Were we not to maintain a consistent sample, the results of these tests would be contaminated by changes in sample composition. In order to make all In the next round of the BUrvOY we will coll~t a ICCOIld year of earnings data for the workera in thiB IIII11plo, u well u additional inf'onnation about their previoua work experience and on the job training. This will allow UI to ICe what has happened to real wages over the year, u wen as expanding our exploration of human capital development. J alIo intend to continue working with thiB datuet to expand the analy.ia in thiB chapter, particularly with rcapcct to controlling for IClcction bias and endogcncity in lOme of the econometric reeulta. 2 Of thcIC 108, 69 were eliminated becaulC the year the individual fmiahed IChool was milling, 21 becauae of miuing leveJa of completed IChooling, 22 miasing current wagea and S miasing tenure at the cummt job. There was IOI11e overlap in the milling data. 95 of the results in the chapter comparable, it is important to use the same sample throughout. This sub-sample of 1609 is the largest for which all of the data needed for the first sections of the chapter are available. Throughout the analysis, the earnings measure used is hourly earnings as reported by the workers, and the dependent variable in all regressions is the log of hourly earnings. Earnings include both wages and the value of cash and in-kind allowances paid by the employer. These allowances included employer provided meals and protective clothing, any cash or kind payments for housing or transportation, and any bonuses, including the Christmas bonus. The wage and hours data are drawn from the workers' own reports, while the allowances are as reported by the firm. For each worker, the average allowances reported by the firm for workers in his category were added to his reported wages to produce earnings. Experience is measured as potential experience, or the number of years since the individual left school. This is equivalent in spirit to the Mincerian formulation of Experience = Age - Schooling - 6, but more appropriate to the Zimbabwean context, where many individuals start school after the age of six, and repetition of grades implies that the number of grades completed may seriously understate the number of years spent in school. Throughout the analysis, and in all tables, "experience" refers to this measure of potential experience. The tenure measure is the number of years the individual reported working in the firm. The following section will discuss a variety of ways of measuring educational attainment. The summary statistics presented here include a continuous measure of education as grades completed, based on the workers' reported school level and highest grade or form. Voca- tional education was assumed to add one year of schooling, polytechnic two years, and university three years. The analysis begins with population means. These are calculated using sampling weights, which give larger weight to workers from larger firms, each of whom "represents" a greater number of workers in generating a picture of the manufacturing labour force as a whole. Summary statistics for the whole sample of 1609 workers, as well as breakdowns by occupation, sector, race, gender 'and location, are presented in Table 8.1. As a general premise, we can discern relatively little from an examination of means, though it does serve two functions. The first is to provide a description of the data, revealing basic patterns which form a baseline. The second is to generate questions, and the dominant ones that arise from these statistics are about the large variations. in wages and worker characteristics, both across and within firm sizes, occupations and regions. The ftrst conclusion one reaches in looking at Table 8.1 is that there is substantial variation in the data. For almost all the variables, the standard deviations are quite large relative to the means. implying that examining means alone would mask much of what is interesting in the data. and therefore in the labour market. Second, there is a substantial gap between wages and earnings, indicating that allowances play an important role in total compensation. As we shall see later. their role is more important in larger ftrms. Third. the mean worker in this sample has a great deal of experience, both in this job, and potentially in the labour market over all. The mean years of schooling represents approximately 2 years of secondary school. The pattern of wages by occupation conforms with general expectations. One interesting feature is that technicians have the second highest mean earnings, exceeded only by management. These individuals generally have quite high education levels as well, for the most part having university or polytechnic education. 96 The table indicates that wages in the wood sector are much lower on average than in the other three sectors. Average firm size in this sector is also smaller than in the other sectors and, as we shall see, firm size exerts a great influence on wages. The composition of the sample by sector, firm size and occupation, along with mean wages for each group, is presented in Tables 8.1a - 8.lc. Finally, average wages in Harare are significantly higher than in the other regions. This may reflect higher costs of Jiving in Harare, or may include some compensation for greater commuting costs or the unpleasantness of urban life. Finally, the data also indicate significant earnings differences across racial groups and between men and women. Section 3 will explore these differences in some more detail. The rest of this chapter will examine hypotheses explaining the pattern of earnings indicated by Table 8.1, and explore the variation that underlies those means. Table S.1: Population summary statistics using sampling weights number of mean standard observation deviation Hourly Wage in Z$ 1609 6.87 9.58 Hourly Earnings in Z$ 1609 8.53 11.97 Worker Age in Years 1609 35.06 10.33 Potential Experience in Years 1609 17.95 11.17 Tenure in Years 1609 9.65 8.13 Number of Grades Completed 1609 9.08 3.05 earnings by occupation in Z$ Management 115 31.72 18.18 Administration 162 12.48 10.71 Commercial and Sales 55 9.06 9.78 Supervisors 152 10.35 7.68 Technicians 31 18.63 8.98 Equipment Maintenace 35 9.68 10.15 Skilled Production Workers 152 5.68 4.63 Unskilled Production Workers 767 3.24 2.09 Apprentices and Trainees 28 8.31 8.51 Support Staff 112 5.18 7.14 earnings by sector Food 392 9.61 13.56 Wood 202 4.84 7.20 Textiles and Garments 712 8.41 12.01 Metal 303 9.16 10.36 earnings by location in Z $ Harare 949 9.98 13.36 Bulawayo 477 6.61 9.46 Other 183 5.81 8.1S eamings by race In Z$ European 49 33.59 22.58 Asian 23 16.38 15.67 African 1537 7.61 10.39 earnings by gender In Z$ Men 1320 8.95 12.49 Women 289 6.29 8.35 97 Table B.la: Mean hourly earnings by firm size and sector small midium large very large food 30 117 90 155 2.51 3.38 7.17 12.29 1.35 2.61 10.24 14.72 wood 17 81 64 40 1.14 4.36 6.07 5.92 .54 6.19 9.06 7.11 metal 73 222 214 203 2.28 3.70 6.91 10.63 1.17 4.16 9.48 14.02 textile 29 136 88 50 3.22 10.34 8.51 10.06 2.51 8.51 9.23 11.09 Each cell includes: Number of observations Mean earnings Standard deviation Table 8.1b: Mean hourly earnings by occupation and firm size small medium large very large management 22 30 29 34 3.58 19.96 32.40 42.00 2.69 16.37 16.18 22.06 edministration 3 45 53 61 1.63 9.63 9.39 15.05 .75 9.43 6.11 12.31 commercial 2 23 15 15 2.13 6.03 9.68 16.64 .12 8.06 7.79 11.42 supervisor 3 41 57 51 1.78 7.16 7.08 10.37 3.8 7.47 4.76 6.94 technician 0 5 8 18 16.59 14.98 19.52 13.65 19.93 8.08 maintenance 0 6 17 12 10.89 9.59 8.95 15.26 12.56 5.87 skRled prod. 18 53 42 39 2.71 6.45 4.30 5.64 1.31 5.82 3.24 4.35 unskilled prod. 95 308 198 166 2.13 2.91 3.28 4.04 1.25 1.69 1.57 3.07 apprentice 4 4 8 12 1.39 4.21 7.16 6.67 1.15 0.69 4.52 6.74 support staff 2 41 29 40 1.96 2.67 3.67 6.23 0.26 1.60 4.20 8.73 Each cell contains: Number of observations Mean earnings Standard deviation 98 Table 8.1 c: Mean hourly earnings by occupation and sector food wood metal textile management 23 14 58 20 32.25 23.11 23.66 30.38 23.66 15.57 22.29 20.30 administration 48 18 61 35 11.14 9.62 11.02 13.54 11.45 7.86 9.79 9.71 commercial 25 4 20 6 6.12 6.30 13.45 15.13 5.79 0.99 11.84 13.85 supervisor 36 17 67 32 7.52 5.21 7.38 11.79 6.14 4.68 5.40 8.37 technician 10 1 13 7 24.75 10.60 , 3.19 17.79 16.31 8.38 10.91 maintenance 11 3 16 5 8.36 4.17 9.06 17.28 10.67 2.46 10.20 15.54 skilled prod. 27 16 70 39 5.82 2.86 3.38 9.02 5.35 1.19 1.87 5.78 unskilled prod. 171 3.35 ," 5.28 359 2.94 126 4.05 2.55 0.99 1.86 2.36 apprentice 4 3 7 14 5.97 0.87 9.85 4.59 1.07 0.62 9.43 1.03 support staff 37 15 41 19 6.18 2.40 3.41 3.41 9.14 0.43 3.62 2,1a Each cell contains: Number of observations Mean eamings Standard deviation 8.2 Returns to human capital The first step in the analysis is to estimate a basic human capital earnings function. These estimates provide a baseline for the rest of the analysis and allow for some comparison with the results from other studies, as well as permitting us to draw some conclusions about the structure of the returns to schooling and experience within Zimbabwean manufacturing. While the results are quite interesting, they should be interpreted with caution. In particular, these estimates should not be construed as providing measures of the private returns to education and experience in the Zimbabwean economy as a whole. The sample includes only wage earners in manufacturing, where wages and educational attainment are likely to be higher than among the self-employed, farmers, or employees in other sectors. Consequently, while the analysis generates reasonably precise estimates of the returns to human capital, these estimated returns should be considered as valid for the four manufactu~ ring industries comprising the sample, and not be interpreted more broadly. The simple form of the earnings function posits that earnings are a function of experience and schooling. Experience is measured as described in Section 1. Schooling is measured in two ways. The first is a continuous variable, Educate, which measures the number of grades completed as described in the previous section. The second measurement scheme uses 99 dummy variables to represent the highest level of academic education achieved, with additional dummies indicating whether the individual also attended vocational or technical school. Thus, Primary= 1 means that the last year of academic schooling the individual achieved was a primary grade. Primary = 1 does not imply that the individual completed primary school. Secondary and University are constructed the same way. The coefficients on these variables measure the increase in hourly earnings of an individual with this level of schooling relative to individuals with no formal schooling. It should be noted that only 1.3 % of the sample have no education. The Vocational and Polytechnic dummies measure the increase in hourly earnings of an individual attending either of these institutions relative to what they would have received based on their academic education alone. Finally, both specifications include a dummy variable which indicates whether the individual was ever an apprentice. The results of the estimation of both forms of the earnings function are presented in Table 8.2. 3 Table 8.2: Human capital earnings functions independent 111 variable Experience .0842 (13.79) .0812 (12.98) Experience 2 -.001' " 0.65) -.0012 ( 9.59) Educate .1627117.40) Primary .347 ( 3.65) Secondary 1.030 (9.29) University 2.627 (11.31) Vocational 0.142 ( 0.87) Polytechnic 1.208 ( 7.16) Former Ap- .6093 ( 5.71) .686 (6.04) prentice Constant -1.027 (7.12) -.323 ( 2.60) Adjusted R2 .38 .30 Dependent Variable is log Hourly Earnings Absolute value of t-statistics in parentheses Number of observations = '609 The results in Table 8.2 illustrate several interesting features of the return to human capital in Zimbabwean manufacturing. First, the return to experience is quite steep, though it does get gradually flatter, reaching a peak: at 34 to 38 years of experience, depending upon the specification. At the mean value of potential experience in the sample (18 years), an additional year of experience raises wages by 5~. The results also indicate a relatively high rate of return to an additional year of schooling at 16.3~. This is not, however, inconsistent with results produced from other African data sets. Van der Gaag and Vijverberg (1989), for example, get a rate of return of 20.7~ using the Ivorian LSMS. It is obvious from the results of the specification using dummies, however, that this average rate of return masks differences in the returns to education across schooling levels. The returns to the level of schooling indicated by each dummy variable is the antilog of the coefficient of that variable, minus one. The results indicate that individuals with at least some primary school earn 41 ~ more than their unschooled colleagues. Secondary school attendance increases earnings by 180% relative to the uneducated, while the university educated earn a remarkable 12 times 3 In this and all other regressions, standard errors have been corrected for heteroscedasticity using White's (1980) method. 100 as much as individuals with no schooling at all. Clearly, the returns to education are non linear. The results in column 2 also provide some interesting insights into the returns to technical training. Vocational school appears to provide no increment to earnings beyond that generated by the academic education the individual received. Apprenticeship, however, provides a boost of nearly 100%. The sample includes 38 vocational school graduates and 104 former apprentices, among whom 9 individuals did both types of training. While it might be tempting to use this evidence to reach conclusions about the relative quality of training provided by these two institutions, there are a number of potential selection issues which make such a conclusion inappropriate without additional research. The clear nonlinearities in the returns to schooling are explored in the estimates presented in Table 8.3. Table 8.3: Human capital earnings functions independent variable 111 Experience .1012(15.131) .0833 (13.626) Experience 2 -.0014!11.411) -.001219.918) Years Primary School .009 10.573) Years Secondary School .2386115.022) Some Primary School .3496 ( 3.626) Complete Primary School .3290 (3.244) Some Secondary School .982 (9.018) Complete Secondary School 1.726 (11.461) University .961 (4.328) .891 (4.008) Vocational -.016! 0.093) -.233 (1.268) Polytechnical .872! 5.365) .960 (4.6721 Former Apprentice .591 ( 5.290) .619 (5.450) Constant -.S32( 2.372) -.328 (2.690) Adjusted R2 .36 .33 Dependent Variable is Log Hourly Earnings Absolute value of t-statistics in parentheses Number of observations = 1609 The estimates in the first column separate measured years of schooling into years of primary and ~I!3l"S of secondary school. Unfortunately, the data do not indicate the precise number of years of university, vocational or polytechnica1 education, so these levels are measured with dummy variables. All university educated people are assumed to have completed the six years of secondary school, so the University dummy represents the returns to university education relative to completion of secondary school. The specification presented in the second column replaces the number of years of each type of school with dummies for having done some primary. complete primary, some secondary or complete secondary school. Again, university educated individuals are defined as having completed secondary school, so the coefficient on the university dummy represents the returns to university education compared with completion of secondary school only. The vocational and polytechnic dummies measure the returns to these types of education beyond that provided by the level of academic education attained. The returns per year of primary school presented in column 1 are not significantly different from zero. This does not, however, imply that attending primary school has no effect on wages. The specification in column 2 indicates that there are benefits to primary schooling. though an F-test indicates that the coefficients on the two primary school dummies are the same. This implies that the returns to primary school are the same whether 101 one completes the sequence or just attends for some years. Having attended school at all appears to increase wages by about 40%, though additional years of primary schooling have no additional value. This may indicate that the value of primary school is in the acquisition of basic literacy and numeracy skills, or in the socialization to scheduled cooperative work, but that these skills are not necessarily enhanced by additional years of primary schooling. Throughout the rest of the analysis, controls for schooling will take the form in the second column of Table 8.2, with primary education measured with a single dummy. Interestingly, the results also indicate that polytechnic training is generally as valuable as university education, as an F-test indicates no difference between the coefficients on the two dummies. This may be particular to the manufacturing sector, where polytechnic education is a route to jobs as technicians who, we have seen, are quite well paid. It is not clear that this result would hold in a broader sample. Most important, these results indicate that the returns to schooling are not only non- linear, but that they are larger at higher levels of schooling. This is different from the conventional wisdom about education in developing countries, which generally asserts that the largest returns are associated with primary school. The major source of this difference is probably our focus on the manufacturing sector, where some schooling may now be a prerequisite for employment. and where the benefits of basic education which are often posited in the agricultural and self-employed urban sectors would not be visible. What is obvious from these data, however, is that the basic predictions of human capital theory are confirmed, and education and experience have powerful effects on individual productivity and earnings. Unfortunately, labour markets in most of the world also appear to place a high value on some characteristics which are not obviously productivity enhancing. The following section will explore the evidence of discrimination on the basis of race and gender in the manufac- turing labour market in Zimbabwe. 8.3 Discrimination by race and gender This section takes a very cursory look at evidence of discrimination by race and gender in the manufacturing labour market in Zimbabwe. Although this is not one of the primary focuses of the study, controls for race and gender are included in subsequent sections, and it is therefore useful to devote some attention to a discussion of the influence of these variables. The approach taken here is a quite simple one. More detailed and thorough examinations of this issue in Africa, including some excellent analysis of the composition of earnings differen~es across racial and gender groups, can be found in the papers by Knight and Sabot and Armitage and Sabot in Birdsall and Sabot (1991). Discrimination, whether by race or by gender, can take two primary forms. The first is wage discrimination, which occurs when otherwise identical individuals, in identical occupations, earn different wages according to the group to which they belong. Job discrimination occurs when groups have unequal access, for reasons other than different economically relevant personal characteristics, to the higher paying jobs. The examination in this section will attempt to identify the existence of both types of discrimination in this dataset, but will not decompose any measured discrimination into its component parts. The data in Table 8.1 showed quite clearly that there is substantial earnings variations across races and between the genders. Since some of this difference is certainly due to differences in human capital endowments across groups, it is important to first control for these differences. 102 Table 8.4: Earnings functions including controls tor race and gender independent variable (1) Experience .0799(' 3.370) .0599(10.277) Experience 2 -.0012(10.562) -.0009( 8.575) Primary .2934( 3.227) .19B9! 2.044) Some Secondary .8650( 8.494) .5266! 5.096) Complete Secondary 1.36' 2( 9.353) .B658( 6.477) University .96S5! 4.238) .6337( 3.495) Vocational '.2027( 1.2581 -.0174( 0.156) Polytechnical .7646( 4.376) .3677( 2.0361 Former Apprentice .5a25! 5.742) .5165( 5.017) Male .2100( 3.800) .247' ( 4.675) European 1.1992110.322) .7485{ 6.356) Asian .5594( 2.535) .1 747{ 0.980) Management 1.0494( 7.628) Administration .8974(13.535) Commercial .5404( 4.410) Supervisor .5383(11.591 ) Technician 1.0956{ 7.463) Maintenance .573S( 4.8551 Apprentice .0285( 0.'73) Support Staff .09681 1.231) Constant -.3539 ( 2.869) '.161 , ( 1 .236) Adjusted R2 .38 .53 Dependent variable is log hourly earnings Absolute value of t-statistics in parentheses Number of observations = 1609 The first column of Table 8.4 presents an earnings function with three dummy variables included, one identifying the worker as male, the others identifying the worker as Asian or European. All three dummy variables are positive and significant, and their coefficients correspond to wage premia of 23% for men relative to women, and 232% and 75% respectively for Europeans and Asians relative to Africans. These dummies, however, confound the effects of wage and job discrimination. If there is significant racial or gender differentiation in occupation, then job discrimination may be an important source of earnings differences. Table 8.5 presents the distributions of the gender and race groups across occupations. Table 8.5: Occupational distributions by race Md gender efriCIIII asian european man woman Number 1537 23 49 1320 289 Management 4.8 52.2 59.2 7.0 7.6 Administration 9.2 30.4 26.5 7.3 22.5 CommerciallSales 3.4 0 4.1 3.5 3.1 Supervisor 9.8 4.3 0 10.9 2.4 Technician 1.9 0 2.1 2.3 0.3 Equipment Maintenance 2.3 0 0 2.5 0.7 Skilled Production 9.8 4.3 0 9.8 7.9 Unskilled Production 49.8 0 2.1 47.2 49.8 Apprentice 1.5 8.7 4.1 2.1 0 Support Staff 7.2 a 2.1 7.3 5.5 Numbers in cells are the percent of each group in each occupation (columns add to 100) 103 The table indicates substantial differences in the occupational distribution, both across races and across genders. While it is clear that Europeans and Asians, who constitute a very small fraction of the sample, are indeed concentrated in the higher paying occupations, the pattern of differentiation by gender is less clear. Both groups are equally likely to be unskilled production workers, the lowest paying job category, and managers, the highest paid group. There is significant differentiation within the other job categories, but it is not clear that one group is in occupations which are generally higher paying, since the high concentration of women in administrative jobs is balanced by the higher concentration of men in the supervisory and technical jobs. The importance of occupational differences, holding constant for general measures of human capital, is presented in the second column of Table 8.4, which includes controls for occupation. The inclusion of these controlS renders the coefficient on the Asian dummy insignificant, indicating that much of the effect in the previous column was due to job discrimination, rather than to wage discrimination. The introduction of occupational controls also substantially reduces the size of the coefficient on the European dummy, indicating that some of the wage difference associated with being European is due to differences in the occupational distribution. The coefficient on the male dummy, however, is actually increased, and it appears that occupational differences do not tell much of that story. It is important to be aware, however, that the pattern of occupational segregation by gender does exist, and while it does not seem to be the source of the differentiation in wages, it may merit concern for other reasons. In order to extend this analysis further, it would be advisable to allow the returns to human capital to vary across racial and gender groups. This would allow much more detailed analysis of the sources of this differentiation, and is probably a promising direction for future research. However, it is important to note that differences, for example, in the returns to schooling for Europeans and Africans may not reflect discrimination in the job market itself. Prior to 1980, the Zimbabwean school system was segregated by race, and levels of expenditure per pupil were much higher in the European and Asian schools. This difference in the quality of schooling, which resulted from discrimination in the schooling system rather than in the labour market, may be a source of some of the wage differences we see. 8.4 Employer size wage differentials - single labour market approaches The existence of positive earnings differentials associated with employment in larger flIlllS is a weU-established phenomenon, and the subject of a large theoretical and empirical literature in labour economics, including both neoclassical and institutional approaches to explaining these differentials. While many of the hypotheses in this literature have some explanatory power, none can completely explain the existence of the differentials. This section will use the RPED data to expand this literature with two goals in mind. First, as is the case in most areas of empirical labour economics, most work in this area has been based on developed country data, and the Zimbabwean RPED provides an opportunity to compare the differentials in Zimbabwe with those found in developing countries, and to evaluate the usefulness of the explanations common in the developed country literature in the African context. Second, the RPED survey is unique in that it combines a sample of over 1600 workers with a broad set of data about the firms where they are employed. Most of the empirical work on this issue is based on individual and household datasets that contain a great deal of information about the employees, but very limited information about firms. The additional firm data available in the RPED allows us to examine some additional hypotheses 104 about the source of these differentials, and to test others in different, and perhaps more compelling ways. It should also be noted, however, that the availability of additional information about firms in this dataset must be traded off against relatively sparse information about workers. Though the data do include the items necessary for the estimation of earnings functions, they do not contain as much data about the workers as is common in studies of this type. Household or individual surveys generally include, for example, information on household composition and earnings and the family background of the individual, all of which may influence labour market choices and outcomes. This is particularly relevant for attempts to include selectivity correction techniques, as we shall discuss later. One clear advantage of this dataset is that it contains precise information about firm size as measured by total employment, rather than simply the size category of firm in which the individual is employed. This permits estimation of the employer-size effect using a continuous size variable. For purposes of comparison, however, estimates based on size class dummies will also be presented. The continuous measure used is In(size), which was selected both because it is standard in studies using continuous size data (see Brown and Medoff 1989), and because the relationship between earnings and firm size appears to be nonlinear, and the log of size captures some of this nonlinearity. The discrete measure uses three firm size dummies: medium (11 to 100 workers), large (l01 to 250 workers) and very large (more than 250 workers). The dependent variable throughout the analysis, unless otherwise indicated, is the log of individual hourly earnings, as in the previous sections of this chapter. For the continuous variable, the coefficients can be interpreted as elasticities. The magnitude of the size effect can be measured as the percentage increment associated with working for a firm whose In(size) is one standard deviation above the mean relative to one whose In(size) is one s .dard deviation below the mean. The standard deviation of In(size) in this sample is 1.55. For the dummy variables, the magnitude of the size effect is the percentage increment in earnings, relative to the base category of small firm (0 to 10 workers), associated with employment in a firm of each size class. This is calculated as the antilog of the coefficient, minus one. Summary statistics by firm size class are presented in the top panel of Table 8.6. The bottom panel presents baseline size coefficients from a regression of earnings on firm size including no control variables. Both panels show a clear pattern of wages rising with firm size. The remainder of this analysis will present, and test, a number of hypotheses about the source of the differential. The presentation follows the structure of Brown and Medoff (1989), as this paner provides an excellent summary of the hypotheses presented in the literature, and a structure well worthy of emulation. The analysis begins with an exploration of market clearing hypotheses about the source of the size premium by examining the effects of differences in labour quality and job character- istics across firm sizes. While these controls do reduce the size effect, a significant premium remains. The next ~. ~~ion considers three institutional features that could generate the premium, unionizati< ; :mimum wages, and ownership structure, and finds that they cannot explain away the siz~ effect. Finally, the analysis examined new institutional hypotheses, including efficiency wages and rent sharing, and concludes that they do potentially have explanatory power. 105 Table 8.6: Summary statistics by firm size all small medium large very large Number of 1609 149 556 456 448 Workers % of Total 100 9.3 34.6 28.3 27.8 Mean Wage 5.86 2.06 4.44 5.95 8.80 /8.30) (1.56) 15.781 (7.73) 111.35) Mean Earnings 7.08 2.38 5.35 7.15 10.72 110.14) (1.561 (7.40) (9.53) (13.37) Mean Experi- 18.55 13.15 18.96 20.02 18.35 ence (11.251 (10.78) (11.90) (10.78) (10.50) Mean Grade 8.69 8.95 8.32 8.56 9.19 Completed (2.94) (2.701 (2.80) (2.931 13.13) Mean Tenure 9.58 3.95 8.85 11.02 10.89 (8.061 (4.56) (7.86) 18.26) 18.09) Earnings Dis- tribution 90 %ile 16.40 4.07 11.25 16.05 26.43 75 %ile 6.37 2.75 4.84 6.60 12.30 Median 3.37 2.02 2.79 3.56 4.88 25 %ile 2.44 1.45 2.27 2.72 3.08 10 %ile 1.94 .96 1.78 2.22 2.52 baseline size coefficients In = 1609) specification controls size measure coefficient % premium R2 (1) None In/size) .257 (12.99) 79% .205 (2) None Medium .571 (5.075) 77% .153 Large .854 135% Very Large 1.188(10.75) 228 Market clearing hypotheses Labour quality differences In a simple form, the basic labour quality difference argument is that larger firms hire "better" workers, and therefore pay higber wages. The reasons for this preference can be most easily understood with reference to the greater capital intensity of larger firms. To the extent that physical and human capital are complements, firms using more physical capital will also require more human capital, hiring workers with larger human capital endowments, and higher productivity, and paying higher wages. If this hypothesis is true, then the size wage effect should be explained, and hence the significant positive coefficient on firm size eliminated, by controlling for worker quality in the estimated earnings functions. Regressions including controls for measurable human capital are presented in columns 1 and 2 of Table 8.7. While the effects of firm size are smaller than those presented in Table 8.6, they are certainly not eliminated. It is possible that workers in larger firms are not only of better quality when hired, but that they acquire additional human capital on the job. If larger firms use more sophisticated capital equipment than smaller firms. they may provide more training for their workers, and this additional human capital would also be reflected in higher earnings. Columns 3 and 4 of- Table 8.7 present estimates including controls for whether the worker is currently receiving job-related training on or off the job. The addition of these controls further reduces, but does not eliminate, the size effect. 106 Table 8.1: Earnings regressions including size and labour quality controls (1) 121 (3) 141 Constant -.818 (5.93) -.629 14.30) -.807 (5.97) -.650 (4.5') Experience .057 (11.4) .059 (11.8) .060 (12.0) .063 (12.6) Experience 211 00 -.089 (8.90) -.095 (8.63) -.092 (9.30) -.097 (9.70) Tenure .005 (1.67) .001 (2.33) .00511.67) .006 (2.00) Primary .197 12.051 .243 (2.411 .191 (2.051 .230 (2.39) Some Secondary .663 16.371 .721 (6.79) .645 (6.321 .702 (6.68) Complete Secondary 1.01918.231 1.169 (8.47) 1.010 (7.651 1.085 (7.80) University .730 (2.911 .798 (3.59) .190 (3.28) .859 (3.95) Vocational -.14311.10) -.14211.04) -.1 12 10.88) -.109 (0.83) Polytechnical .519 12.91) .561 (3.04) .524 (2.72) .566 (2.94) Former Apprentice .564 (7.20) .572 16.81) .509 (6.44) .512 (6.02) On the job Training .208 (2.57) .226 (2.67) Outside Training .25814.22) .283 (4.71) Male .076 (1.46) .095 (1.821 .073 (1.43) .091 (1.75) European 1.251 (11.2) 1.280 (11.4) 1.216 (11.2) 1.301 (11.4) Asian .687 (3.25) .663 (3.29) .664 (3.06) .639 (3.06) Harare ., 61 (2.1 5) .199 (2.31) .174 (2.42) .207 (2.55) Bulawayo .002 (0.26) .031 (0.37) .020 (0.21) .045 (0.57) LnlSize) .118 (9.88) .159 (8.83) Medium .341 (3.85) .329 (3.73) Large .53515.88) .484 (5.56) Very Large .80418.20) .721 (7.36) Adjusted R2 .490 .472 .504 .489 Dependent Variable is log hourly earnings Absolute value of t-statistics in parentheses N = 1609 The labour quality controls necessarily omit unmeasured dimensions of quality. To the extent that these characteristics are correlated with the measured characteristics included as controls, this will bias the coefficients on the human capital variables, but should not result in the presence of a significant size effect. If, however, the unmeasured aspects of worker quality are correlated with size but not with the measured characteristics, they would generate a positive size effect even when controls for measurable quality differences are included. The appropriate mechanism for dealing with such uncorrelated components of unobserved ability would be to treat it as a case of endogenous selection, allowing for soning of workers into firm sizes non randomly, on the basis of this unobserved ability. In principle, one could use a variety of methods to control for this general endogenous selection, depending upon whether one used continuous or discrete size measures and on the nature of the unobserved ability attributes one posits. As a general rule, however, controlling for selection requires that one be able to identify the selection process, generally through the use of exclusion restrictions, including in the selection equation factors associated with the unobserved factors governing the sorting process, while including factors related to productivity in the wage equation. Because this dataset includes relatively little information about worker characteris- tics, the appropriate exclusion restrictions are not obvious. Additional information on previous work experience, household structure or family background, for example, would aid in identification of the selection equation. While further work in this area with these data will include efforts to incorporate endogenous selection, at this point the analysis excludes such considerations. A second issue surrounding controlling for selection comes from the underlying selection process itself. Techniques for controlling for selection with multiple categories or continuous selection variables include those proposed by Idson and Feaster (1990) and Garen (1984) 107 respectively. Both techniques are ordered, with Idson and Feaster using a first stage ordered probit and Garen using a first stage OLS regression. In both cases, this requires the underly- ing selection process to be related to an unobserved worker characteristics that monotoni- cally increases as firm size rises. In other words, under these approaches, workers in large firms are not simply different from those in small firms, but different in that they have more of some desirable characteristic or characteristics. It seems more likely, however, that the unobserved characteristics by which workers are desirable to different firms, and hence sorted across firm sizes, differ across size classes. A more flexible approach using Lee's (1983) method, with a multinominal logit in the first stage, would therefore be more attractive. Unfortunately, this approach includes many more parameters, generating more imprecision in the results. Continued work with these data will include attempts to incorpor- ate, and compare, all of these selection correction techniques. It is interesting to note, however, that ina study of size wage differentials in Peru, Schaffner (1994) finds the results of attempts to control for endogenous selection using both ordered probit and multi nominal logit approaches, "unable to provide trustworthy evidence about selection effects." (p. 16) Finally, it should be noted that the coefficients reported in Table 8.7 are substantially larger than those seen in the literature on developed countries. In particular, Brown and Medoff (1989), reporting results based on a large number of U.S. data sets, report two standard deviation size wage premia, controlling for measured worker quality of 6 to 15%. Our estimates are, however, roughly similar to those found by Schaffner (1994) using data from the Peruvian LSMS, and to results in Little et al. (1987) for India and Malaysia. Job quality and compensating dirrerentials On the other side of the coin from differences in labour quality across firm sizes is differen- ces in the quality of jobs. The principle of compensating differentials implies that workers who face less attractive working conditions wilJ receive higher wages in compensation. even when the quality of these workers is the same as the quality of those working in better conditions. In order for these differences to explain the size wage differential. jobs in larger firms must have more unattractive characteristics than jobs in smaller firms. Working conditions are extremely difficult to measure, as they include both physical characteristics regarding the difficulty, danger and unpleasantness of the work. as well as characteristics of the work atmosphere such as relationships in the workplace and the tedium of the tasks. The RPED data do include some data which can be used as proxies for working conditions, and this section will use them to test the hypothesis that the size wage premium can be explained by compensating differentials. A first step is to include controls for industry and occupation. Working conditions probably differ across industries and, as seen in Table 8.9, industry composition does differ by firm size in this sample. Similarly, occupational differences also may engender differ- ences in working conditions, both physical and in terms of working atmosphere. The data include only four industries and ten occupational characteristics, but Brown and Medoff (1989) find that including more detailed industry and occupation dummies has no effect on the size coefficient. Estimates of the size coefficients including these controls are presented in Table 8.9. The industry controls include three dummies (textiles is the omitted category) and occupation controls are 9 dummies (production workers are the reference group). Columns 1 and S present, for each specification, estimates without occupation and industry controls from Table 8.7. Inclusion of these controls actually increases the magnitude of the size coefficients in both specifications, indicating that even if industry and occupational differences are associated 108 with differences in working conditions, they are not a source of compensating differentials which reduce the firm size earnings premium. The data do allow measurement of, or proxies for, more specific aspects of working conditions. In order for these differences to reduce the firm size effect, working conditions must deteriorate as firm size increases. As we shall see, while addition of these controls does reduce the size coefficients, it is also generally the case that better working conditions are associated with larger employer size. One possible source of working conditions variation is hours. If longer hours of work are a source of disutility for workers, then one might expect those working longer hours to receive higher wages. The third row of Table 8.8 shows that mean usually weekly hours as reponed by the worker does vary with firm size. However, longer hours are associated with employment in smaller firms. If we restrict the sample to full time workers only (full time being more than 35 hours per week), the relationship is stronger. Table 8.8: Industry and occupation distributions by firm size aU sm81i meaium large very large industry distribution % Food 24.4 20.1 21.0 19.7 34.6 % Textile 44.2 49.0 39.9 46.9 45.3 % Wood 12.5 11.4 14.5 14.1 8.9 % Metal 18.8 19.5 24.5 19.3 11.2 occupation distribution (in percent) Management 7.2 14.7 5.4 6.4 7.6 Admin 10.1 2.0 8.1 11.6 13.6 Commercial 3.4 1.3 4.1 3.3 3.3 Supervisor 9.5 2.0 7.4 12.5 11.4 Technician 1.9 0.0 0.9 1.7 4.0 Maintenance 2.2 0.0 1.1 3.7 2.7 Production 57.1 75.8 64.9 52.6 45.7 Apprentice 1.7 2.7 0.7 1.7 2.7 Support Staff 6.9 1.3 7.4 6.4 8.9 Table 8.8 continued: Working conditions standard deviations in parentheses iii smill medium large very large usual weekly hours for all workers · Mean 45.1 (6.2) 45.9113.1) 45.7 (7.2) 44.8 (2.9) 44.5 (2.7) · Median 45 45 45 45 45 full time workers · Mean 45.5 (5.2) 48.3 (7.9) 45.9 (7.03) 45.0 (2.42) 44.5 (2.55) : Median 45 45 45 45 45 Percent in Multishift 28 6 13 35 47 Firms Percent in Firms 3 27 2 0 0 with No Electricity Percent in Firms 7 51 8 0 0 with No Phone Percent in Firms 62 41 56 68 70 where Transport is a Problem Percent who are 7 34 9 1 relative of owner or manager 109 Table 8.9: Earnings regressions Including Industry and occupation controls rn---- 12J - - [3T - - -(41- lSI (61 (11 lSI Ln(Sizel .159(8.831 .161 (9.651 .158 (9.08) .164 (9.771 Medium .329 (3.731 .322 (4.091 .380 (3.991 .367 14.191 Large .484 (5.561 .502 (6.331 .504 15.31 I .511 (5.891 Very large .721 (7.361 .750IB.311 .732 (6.991 .1690.85) Industry Controls no yes no yes no yes no yes Occupation Con- no no yes yes no no yes yes troIs Adjusted R2 .504 .526 .628 .652 .489 .515 .613 .641 Dependent Variable is log Hourlv Earnings Absolute value of t-statistics in parentheses N = 1609 All regressions include controls for experience and its square, tenure, schooling as in Table 2. training. race, gender and location 110 A second source of working conditions differences is the need to work evening or night shifts, the inconvenience of which may generate compensating differentials. Because the survey was conducted during normal business hours, none of the workers surveyed are nightshift workers. However, the firm was asked if it generally conducted multiple shifts, and a dummy variable indicating that it does so was generated. At best, this variable proxies for the possibility that workers might be asked, or forced, to work nights. The fraction of workers in each size class working in multi-shift plants is reported in Table 8.8, and is positively correlated with firm size. While the data do not include measures of the hazardousness of work, they do include some measures which can be considers as proxies for aspects of the physical comfort of the workplace. The firm data include information about use of utilities, allowing construction of two dummy variables. NOELEC takes the value 1 if the firm has no electricity, while NOPHONE takes the value 1 if the firm has no telephone. The electricity measure is obviously capturing the use of mechanized equipment. However, it also proxies for the availability of lighting, cooling or heating of the premises, and if my experience conducting the survey is any guide, the presence of indoor plumbing. Both measures can also be considered as proxies for working in "modern" facilities. Not surprisingly, the fraction of workers in facilities without phone or electricity decreases sharply as firm size rises, as seen in Table 8.8. A third commonly mentioned source of compensating differentials is commuting costs. If larger firms employ workers from a larger geographical area, these workers will have greater time and/or money costs of commuting to work, and should receive higher wages in compensation. Many of the firms and workers in Zimbabwe complained that traffic congestion and inadequate bus services generated extremely long commutes. Unfortunately, the data do not include individual measures of commuting time or costs. They do, however, include a series of questions about the firm's opinion of the quality of infrastructure. Owners and managers were asked to rank the problems posed by transport for workers on a scale from 1 to 5, where 1 signifies that transport posed no problem. From this data, define a dummy variable indicating that the firm felt that worker transport was a problem, rating it 2 or greater. 4 The firm's perception that transport is a problem may proxy for the distance and/or time that it labour force commutes, and therefore for the firm's feeling a need to pay compensation. The fourth row of Table 8.8 indicates that larger firms are indeed more likely to consider worker transport a problem. Table 8.10 presents the effects of including these measures of working conditions in the earnings regressions. Again, columns 1 and 4 reproduce results from Table 8.7 as a referen- ce. The inclusion of the physical measures does substantially reduce the size coefficients. The coefficients on the hours and electricity variables, however, indicate that "worse" working conditions are associated with lower rather than higher wages, while the effects of multishift plants and transport have the expected sign. This gives rise to the possibility that, rather than controlling for the existence of compensating differentials, these measures are picking up a distinction between some "bad" -jobs, characterized by long hours, unpleasant physical conditions, and low wages, and "good" jobs with the opposite characteristics. Regardless of the interpretation one takes, these controls certainly do not eliminate the 4 An alternative specification, in which the dummy is based on an employer mting of 3 or greater, yields qualitatively similar results. 111 premium associated with working for a large firm. 5 Table 8.10: Earnings regressions including working conditions controls (1) (2) (3) (4) Ln(Size) .159 (8.83) .110 (5.49) Medium .329 (3.73) .205 (2.24) Large .484 (5.56) .273 (2.76) Very Large .721 (7.36) .468 (4.41) Hours -.016 (4.65) -.017 (4.92) Multishift Firm .255 (3.89) .290 (4.33) No Electricity -.436 (2.45) -.412 (2.27) No Telephone .012 (0.11) -.036 (0.27) Bad Transport .037 (0.65) .064 (1.13) Adjusted R2 .504 .533 .489 .527 Dependent Variable is Log Hourly Earnings Absolute value of t-statistics in parentheses N = 1609 All regressions include controls for experience and its square. tenure. schooling as in Table 2. training. race. gender and location The hypotheses about differences in labour quality and job quality can also be examined using data on quit rates and tenure in the firm. If the labour quality hypothesis is true, then workers in larger firms would be receiving the same wage they would receive elsewhere, and would be no less inclined to quit jobs in large firms than they are to quit those in small firms. This implies that, holding earnings constant, firm level quit rates should be unrelated to firm size, as should individual tenure with a given employer. Table 8.11: Tenure and quit rate regressions dependent variable controls size measure coefficient Ln(Tenure) Union Membership. Race. Gender. Education. In (size) .150 (4.775) Training. Experience. Location. Industry Occupation Ln(Tenure) Same as Above Medium .360 (1.785) Large .650 (3.281) Very Large .695 (3.226) Ln(quitratel Unionization. share of production workers. In(size) -.328 (3.293) 1-quitrate) share of male workers. In (average hourly wage) Ln(quitratel Same as above Medium - .85 (2.009) 1-Quitrate) Large -1.39 (2.729) Very Large -1.44 (2.764) Absolute value of t-statistics in parentheses 5 One potential measure of subjective working conditions is the share of workcrs who are relatives of the owner. Such individuals may lind the work environment more familial and be willing to accept lower pay. While the share of workers who are relatives is higher in smaller linn.,. inclusion of a dummy renecting relative slatus in the earnings function raises the size coefficient. and the dummy variable is positively eom:lated with earnings. While this does not imply that work atmosphere is unimportant. we have no good means of measuring these subjective characteristics. 112 The addition of the compensating differentials argument implies that. controlling for wages but not working conditions, firm level quit rates would be positively correlated, and individual tenure negatively correlated, with firm size. Estimates of these effects are presented in Table 8.11. The top two rows examine the tenure effects, while the second two use data from 185 firms (missing data caused 16 firms to be dropped) to examine the effects of size on quit rates. The results of both tests contradict the implications of the labour quality and compensating differentiaJs hypotheses, as quit rates decrease and tenure increases as firm size rises. Despite the fact that we are unable to generate estimates including endogenous selection controls, the large magnitude of the measured size effects and the inability of other authors to eliminate them using selection controls make it reasonable ·to conclude that the employer size effect contains some premium element. The next step in the analysis is to examine some institutional explanations for the existence of this premium. Non market clearing hypotheses Traditional institutional hypotheses Unionization Unionization is measured by the employer's indication that at least some of his workers are members of labour unions. There is an extremely strong positive correlation between unionization and firm size, as seen in Table 8.12. If unions have the effect of increasing wages, then the size effects may actuaJly be picking up the higher level of unionization of larger firms. This strong correlation, however, aJso makes it difficult to disentangle the effects of unionization and firm size. This is particularly true for the very large firms, all of which are unionized. Table 8.12: Unionization, minimum wages and ownership structure by firm size (in percent) all small medium large very large Workers who are 78.6 8.7 69.9 91.2 100.0 in unionized firms Workers in firms 18.5 19.4 25.7 14.9 12.9 where minimum wage is binding Mean share of 13.7 130.3) 4.67 (1 B.55) 6.4122.9) 22.9 (3B.3) , 6.S (29.9) foreign ownership (std. dey) Workers in parast- 0.6 0 0 0 2 atal firms Workers in firms 6 0 0 8 13 with some state ownership Workers in coop- 4.5 13 8 0 2 eratives Workers in Firms 45 10 44 50 53 with Zimbabwean owners of European descent 113 Table 8.13 reports the results of including unionization dummies in both specifications. Again, columns I and 3 provide reference estimates from Table 8.7. The union dummy has no effect on the firm size coefficient in the continuous specification, and its coefficient is insignificant. While the union coefficient is also insignificant in the discrete firm size specification, the size coefficients are slightly reduced. This is not surprising if unionization has different effects for different size categories, which would be the case if, for example, larger nonunion firms raise wages to ward off unionization threats, while smaller firms do not. Table 8.13: Earnings regressions including union dummies (1) (2) (3) (4) Ln(Size) .159 (8.83) .1590.21) Medium .329 (3.73) .294 (3.08) Large .484 (5.56) .437 (3.9B) Very Large .721 (7.36) .666 (5.52) Union Dummy .001 (0.011 .067 (0.83) Adjusted R2 .504 .489 .403 .489 Dependent Variable is Log Hourly Earnings Absolute value of t-statistics in parentheses N:o 1609 All regressions include controls tor experience and its square, tenure, schooling as in Table B.2, training, race, gender and location One mechanism for controlling for this is to allow the effects of unionization to vary by firm size, which involves including the interaction of unionization and In(size) in the continuous specification, and separate union dummies for each size class in the discrete. Since all very large firms are unionized, this interaction term would be completely collinear with the very large size dummy. Therefore, for this specification the two largest size classes have been combined. Results are presented in Table 8.14. Both specifications indicate some dec! ining effect of unions at the larger firm sizes, and the discrete specification implies that unionization does generate some of the size effect in the medium size class, but this effect is not large. Table 8.14: Earnings regrassions including interactive union effects 11 I 121 LnlSize) .203 14.834) Medium .33013.75) .252 (2.33) Large & Very large .594 (6.83) .627 13.97) Union .191 (1.04) Union" Ln(Size) -.055 11.14) Small "Union .068 (0.501 Medium" Union .131 (1.31) (Large & Very Large)" Union--- -.016 (0.' 2) Adjusted R2 .504 .480 .480 Dependent Variable is Log Hourly Earnings Absolute value of t-statistics in parentheses N= 1609 All regressions include controls for experience and its square, tenure, schooling as in Table 8.2, training, race. gender and location 114 The lack of significant union effects in the continuous specification may be a result of functional form. The use of In(size) allows the size effect to be non linear, varying with firm size. Since unionization also varies with firm size, such a specification may diminish our ability to discern a union effect. Table 8.15 presents results based on a specification in which size is included linearly. In this specification, the union effect is significantly positive, and reduces the size coefficient. However, if the correct specification for size is nonlinear, then the union effect could simply be picking up these nonlinearities in the size effect. Allowing the union effect to vary by size actually increases the size effect in this specificati- on, and indicates a substantially diminished effect of unionization at larger firm sizes. The weight of this evidence indicates that even if unionization does explain part of the firm size premium, it is far from a complete explanation. Table 8.15 Earnings regressions including union effects 111 (2) (3) Size .00019 (3.' 61 .00017 (3.40) .00334 (4.07) Union .268 (3.77) .416 (5.47) Union+Size -.00317 (3.96) Adjusted R2 .463 .475 .483 Linear Specification of Size Dependent Variable is Log Hourly Earnings Absolute value of t-statistics in parentheses N= 1609 An alternative attempt to disentangle the union from the size effects can be made by restricting our attention to the 1265 workers in union firms. In a regression including the labour quality controls from Table 8.7, the coefficient on In(size) is .147 with a standard error of .025, which is slightly smaller than that estimated for the whole sample. The coefficients (and standard errors) in the discrete specification are .206 (.091), .312 (.079) and .560 (.092) respectively. Another alternative is to use only the workers in small and medium firms, where there is more variation in unionization patterns. The results of this exercise are presented in Table 8.16. Unionization does appear to have some positive effect on wages, and to reduce size coefficients. Table 8.16: Earnings regressions including union effects 111 (2) (3) (4) (5) (6) Ln(Size) .183 (4.65) .169 (.3.84 .183 (3.10) Medium .267 (3.03) .198 (2.061 193 (1.93) Union .054 (0.59) .166 10.55) .154 (1.66) Union· -.033 (0.35) Ln(Size) Small" .075 (0.61) Union Medium· .140 (1.38) Union Adjusted R2 .497 .S20 .520 .497 .503 .501 Dependent Variable is Log Hourly Earnings Small and Medium Firms Only Absolute value of t-statistics in parentheses N=70S All regressions include controls for experience and its square, tenure. schooling as in Table 8.2, training, race, gender and location 115 On the basis of the sum of this evidence, we cannot reject the hypothesis that some of the measured size effect is indeed due to greater unionization as firm size rises, and that this is particularly relevant in the smaller end of the size distribution. Although a precise estimate of the share of the size effect that is due to unionization is not possible, it seems clear that unions certainly do not explain all of the size premium. Minimum wages Another potential institutional source of this wage premium is government intervention in the form of minimum wages. Minimum wage laws have an effect on wages for those firm covered by the law only if those firms are in compliance, and then only if the level of the minimum wage is binding, in the sense of being greater than the wage the firm would offer in the absence of the legislation. If minimum wage compliance is greater in larger firms, or if the wage is binding in larger firms but not in smaller ones, minimum wages could generate the employer size wage premium. Although the data do not include direct measures of either compliance or the relationship of the minimum wage to the firm's equilibrium wage, one question posed in the survey may provide some insight. Employers were asked whether a sl ight reduction in the minimum wage would change their hiring decisions by causing them to hire slightly more or many more workers. Those employers who answered that they would hire more workers can be considered to face a binding minimum wage constraint. Employers who responded in the negative are either not in compliance or are choosing wages at or above the minimum for other reasons. The share of workers in each size class whose employer reported that the minimum wage was binding is reported in Table 8.12. The share increases as we move from the small to medium size class, perhaps indicating more compliance among medium firms, and then falls sharply for the larger size classes. Given the high rate of unionization of the largest firms, it is unlikely that the reason for this decline is non-compliance by large firms, since unions are unlikely to accept wages below the legal minimum. It is more likely that the firms are in compliance, but that the level of the minimum wage is not binding. This pattern makes it highly unlikely that the minimum wage is at the source of the size wage premium. Earnings regressions including a dummy variable indicating that the workers' firm finds the minimum wage binding are presented in Table 8.17. The table includes speCifications in which the effect of binding minimum wages is allowed to vary by firm size. The results indicate that wages received by workers in firms whose employers feel the minimum wage is binding are lower than the wages received by similar workers in other firms. Controlling for the minimum wage increases the size effect for the medium firms slightly, indicating that the minimum wage is probably not the cause of the employer size premium. This does not imply, however, that minimum wage and other labour legislation have no effect on labour markets more generally. 116 Table 8.17: Earnings regressions including controls for minimum wage effects (1) (2) (3) (4) (5) (6) Ln(Size) 1 .156 (8.28) .152 0.92) .159 (8.83) Medium .329 (3.73) .343 (3.89) .393 (3.90) Large .484 (5.56) .486 (S.62) .504 (5.33) Very Large .721 (7.36) .720 (7.35) .7230.03) Minimum Wage -- -.091 (1.47) -.20810.95) -.125 (1.93) Min Wage' .027 (0.49) Ln(Size) Small· Min -.023 (0.12) Wage Medium · Min -.234 (2.99) Wage Large · Min -.094 (1.25) Wage V Large · Min .018 (0.08) Wage Adjusted RZ .504 .505 .505 .489 .492 .493 Dependent Variable is Log Hourly Earnings Absolute value of t-statistics in parentheses N = 1609 All regressions include controls for experience and its square, tenure, schooling as in Table 8.2, training, race, gender and location. Specifications including union dummies as well produce similar results. Ownership structure, government relations and political economy If larger firms are owned by different types of organizations or individuals than are smaller firms, and if these owners have motives for paying higher wages, then this would generate an employer size wage premium. We shall consider four types of owners who may have reason to pay increased wages, and explore the hypothesis that differences in ownership structure drive the size wage premium. First, the sample includes one parastatal corporation with 2766 workers, and 10 other enterprises with some state ownership. If, for example, the parastatal enterprise is not operating on a profit-maximization principle, or is constrained to follow civil service wage guidelines, this could cause wages in this enterprise to be elevated. Similarly, the state, through its participation in joint ventures with the private sector, may require these firms to raise wages. Table 8.12 includes data on the number of workers in each size class working for firms with some state ownership. The sample also includes 14 firms identifying themselves as cooperatives, In some of these cases, the workers are members of the cooperative, in others they are not. The cooperatives are concentrated in the smaller end of the size spectrum (see Table 8.12), and to the extent that workers in these firms are accepting lower wages in order to invest in the capital stock of the cooperative, this may contribute to the pattern of lower wages in smaller firms. Third, owners may use good labour relations to maintain good government relations. This may be important if the firm wishes to continue to receive benefits from government (an issue we shall address in more detail in our discussion of rent sharing explanations), or if the firm wishes to avoid difficulties with the government. Foreign owned firms, for example, may pay higher wages to avoid accusations of worker exploitation, and hence be permitted to repatriate profits and prevent nationalization and expropriation. The share of 117 foreign ownership is indeed greater in larger firms, as seen in Table 8.12. Similarly, Zimbabwean owners of European descent may also wish to avoid trouble with government, and pay higher wages as part of this process. Firms owned by white Zimbabweans are also found in the larger end of the size spectrum, as indicated in Table 8.12. Table 8.18: Earnings regressions including ownership controls III 121 131 t41 Ln(Size) 1 ,138 (7.26) .159 (8.831 Medium .329 (3.73) .323 (3.71J Large .484 (5.56) .41614.74) Very Large .721 (7.35). .66416.85) Parastata! .600 (6.13) .801 (9.81) Some State .02310.26) .063 (0.64) Ownership Cooperative .079 (0.56) .12510.811 Percent of .004 (3.37) .004 (4.16) Foreign Ownership Owner is Zimb· ··. .059 (1.21) .075 (1.511 awean of Euro· pean Descent Adjusted R2 .504 .518 .489 .513 Dependent Variable is Log Hourly Earnings Absolute Value of t·statistics in parentheses N = 1609 All regressions include controls for experience and its sQuare, tenure, schooling as in Table 8.2, training, race, gender and location A set of dummy variables measuring these aspects of ownership structure, as well as a continuous variable measuring the share of foreign ownership, have been included in the earnings regressions presented in Table 8.18. The effects generally have the predicted signs, and are jointly significant at the 10% level. The size coefficients are also reduced by the inclusion of these controls, indicating that ownership structure has some effect, but the size premium still persists. If we include all of the controls implied by all of the hypotheses we have tested in a final earnings regression, we obtain some measure of the magnitude of the size premium left to explain. In the continuous size specification, the resulting coefficient on In(size) is .103 (standard error .023), implying that a distance of one standard deviation above and below the mean of In(size) implies a 32% wage premium. In the discrete size specification, these coefficients (standard errors) are: medium .234 (.102), large .269 (.116) and very large .479 (.120), implying premia of 26%,31 % and 61 % respectively. As a final exercise, we also estimated the size premium using a subsampJe including only the 767 unskilled production workers. This isJikely to be the most homogeneous group of workers, and potentially the one in which such differences as unobservable quality matter least. It is also the job category that is likely to involve the most similar work across size groups. A baseline estimate of the continuous size effect, including no controls, yields a coefficient of .142 (standard error .019), which is substantially smaller than that estimated for the entire sample. We conducted the same hypotheses tests with this sample as were described for the whole sample in the foregoing sections. The results were qualitatively quite similar, and in the interest of space will not be presented. In an earnings regression for the unskilled subsample, including all of these controls, the continuous In(size) coefficient falls to .044 (.021), a significantly smaller premium than we find in the sample as a whole. 118 "New" institutional hypotheses Efficiency wages The labour economics literature includes a large number of models which share the characte- ristic that under certain circumstances, raising wages may also increase firm profits. These "efficiency" wage stories are generally based on the existence of relationships between wages and such factors as worker effort, worker turnover, esprit de corps, and the number and quality of applicants per vacancy. These models can be used to explain firm size wage differentials if these potential relationships between wages, productivity and profits vary across firm size. Because the basic relationships between, for example, wages and effort or wages and applicant quality cannot be observed, direct tests of these hypotheses are generally not possible. Instead, this analysis will explore some of the implications of these hypotheses and attempt to assess their validity in the Zimbabwean manufacturing case. We will be particularly interested in arguments which might also explain why the premium received by unskilled workers is smaller than that received by workers generally. The first set of explanations focus on the hiring process, and involve the premise that raising wages increases either the size or the size and average quality of the applicant pool. In two related papers, Montgomery (1991) and Lang (1991) present models of the recruiting process in which some firms have an incentive to raise wages in order to increase the number of applications they receive per vacancy, and therefore reduce the probability that the vacancy will go unfilled. The extent to which any individual firm chooses to undertake this practice depends on two factors: the cost to the firm of an unfilled vacancy and the overall tightness of the labour market, measured as the ratio of job vacancies to potential applicants. In order for this approach to be a useful means of explaining the size premia, large firms must have higher costs of vacancy, since these models assume workers are homogeneous, so the relative tightness of the labour market is the same for all firms, though not for all occupations. Differences in the cost of vacancies can be generated by differences in capital intensity or in profitability per worker. If vacancies result in the idling of more capital or the failure to supervise more workers or a lack of accounting for larger amounts of business in larger firms, then these firms might choose to offer higher wages to avoid these vacancies. Capital intensity, as measured by the replacement value of plant and equipment per worker, does in fact increase with firm size in these data, ranging from a mean of Z$25,OOO in small firms to Z$112,OOO in very large firms. The explanation also relies on the existence of tightness in the labour market. This does not seem plausible in the case of unskilled blue collar workers> given the level of unemploy- ment and underemployment in urban Zimbabwe. One does not have the impression that any firm need fear a vacancy on the factory floor will long go unfilled. The theory does seem more plausible for skilled technical, administrative and managerial personnel, who may be in short supply. However, these workers make up a tiny fraction of the labour force of the small firms, so this argument is not likely to be useful in explaining the premium accorded by medium relative to small firms, but may explain the continued increase in wages offered by large and very large firms. The second argument focused on the hiring process is one put forth by Weiss (1980) and Weiss and Landau (1984). This formulation argues that by raising wages, firms can increase the average quality of their applicant pool by attracting a larger number of better quality applicants. The important dimension of quality in these arguments is one which cannot be easily observed by the firm (hence it must be something other than experience or schooling), 119 but which is known to the workers, whose reservation wage is an increasing function of this characteristics. Alternatively, this attribute may be easily discernible by the firm, but not by other workers. If the finn is constrained by union rules or morale concerns to pay all similar workers (in terms of obvious attributes like schooling or experience) the same wage, then it will be concerned about the average quality of the applicant pool since it cannot discriminate by ability in setting wages. This model can generate a relationship between wages and firm size in one of three ways. If smaller finns have a lower quality standard, or are able to offer seemingly similar workers different pay based on other ability measures, or are more able to discern differen- ces in worker quality, then they will be less likely to use this mechanism, and therefore pay lower wages. The first possibility, differential quality requirements across finn sizes, is similar to the endogenous selection story discussed above, and an examination of this hypothesis is forestalled by the same considerations that prevented our examination of endogenous selection. It should be noted, however, that this possibility requires not only that workers in small and large firms possess different quality attributes, but that workers in larger firms possess more of some desirable attribute, sot that they are "better" in some objective sense. The second possibility, in which smaller firms are more able to pay workers based on their total ability, and hence offer different pay to seemingly similar workers, seems unlikely. While such firms are less likely to be bound by union pay scales, morale concerns should be equally important, particularly if the familial atmosphere is important to workers. It is also more likely that workers in smaller finns will know one another's wages, making the potential morale consequence even more severe. Table 8.19: Hiring mechanisms as reported by firms and workers all small medium large very large employer responses number of observations 198.00 38.00 65.00 47.00 48.00 Relative or friend of owner 8.08 26.32 9.23 0 0 Relative or friend of employee 19.70 21.05 24.62 14.89 16.67 Suggestion of business associate 0.51 0 0 0 2.08 Subtotal 28.29 47.38 33.85 14.89 18.75 Labor Office or Employment Agency 16.16 7.89 7.69 27.66 22.92 Trade or Technical School 1.01 0 1.54 0 2.08 Subtotal 17.17 7.89 9.23 27.66 25.00 Wore of Mouth 28.79 15.79 33.85 36.17 25.00 Formal Advertising 7.58 10.53 4.62 4.26 12.50 Hired from Factory Gate 14.14 10.53 15.38 12.77 16.67 Subtotal 50.51 36.85 53.85 53.20 54.17 Other 4.03 7.89 3.08 4.26 2.08 Distribution of responses in percent continued on next page 120 Table 8.19 continued: Hiring mechanisms as reported by firms and workers all smell medium larg8very large worker responses -- all workers number of responses 1609.0 149.0 556.0 456.0 448.0 Relative or friend of owner 11.7 42.8 13.7 6.6 4.2 Relative or friend of employee 28.3 23.1 28.1 33.4 25.2 Suggestion of business associate 1.6 3.4 1.4 1.3 1.5 Subtotal 41.6 68.5 43.2 41.2 31.0 Labor Office 6.2 0.7 5.4 5.0 10.0 Trade or technical schoo! 0.9 0.0 0.0 2.2 1.1 Subtotal 7.1 0.7 5.4 7.2 11. 1 Word of Mouth 8.2 4.7 13.5 5,7 5.6 Went door to door 28.9 17.0 28.6 30.7 31.5 Forma! advertising 8.9 3.4 4.5 10.3 14.5 Subtotal 46.0 25.1 46.6 46.7 51.6 Other 5.2 4.7 4.8 4.6 6.2 Table 8.19 continued: Hiring mechanisms as reported by firms and workers all small medium large very large worker responses -unskilled pro- duction workers number of responses 767.0 95.0 308.0 198.0 166.0 Relative or friend of owner 9.3 35.8 6.5 6.6 2.4 Relative or friend of employee 31.5 29.5 31.5 35.8 27.7 Suggestion of business associate 0.5 0.0 0.3 0.5 1.2 Subtotal 41.3 65.3 38.3 42.9 31.3 Labor Office 4.3 1.0 4.2 4.5 6.0 Trade or technical school 0.5 0.0 0.0 0.5 0.8 Subtotal 4.8 1.0 4.2 5.0 6.8 Word of Mouth 9.5 4.2 15.2 5.6 6.6 Went door to door 37.7 22.1 37.0 40.9 43.9 Formal advertising 0.5 3.2 1.3 3.5 1.8 Subtotal 47.7 29.5 53.5 50.0 52.3 Other 4.0 4.2 3.9 2.0 6.6 Differential ability to screen workers across firm sizes is a more promising avenue of inquiry. Although screening costs cannot be measured, we do have information about the mechanisms firms use to hire workers, and about the way individual workers found their jobs. If we can posit a relationship between these hiring mechanisms and screening costs, and if the hiring mechanisms used by larger firms are those in which screening costs are higher, this would make this argument is plausible in this context. Employers were asked "What is the common way of finding workers?" The distribution of answers across the 198 firms who responded are presented in Table 8.19. Workers were asked "How did you know about this job?" The distribution of answers across all workers, and for the subsample of unskilled production workers, is also presented in Table 8.19. "I went door to door" from the workers' responses is equivalent to a firm response of "hiring from the factory gate.« The responses in Table 8.19 are grouped into three broad hiring mechanisms, with subtotals of responses presented for each group. These groups can be viewed as correspon- ding to monotonically increasing screening costs. The first group includes hiring one's own family members and friends, hiring the relatives and friends of one's employees, or, least common, hiring on the recommendation of a business associate. If one knows the characte- ristics of family members, if it is plausible that current employees' relatives share the 121 employees' desirable characteristics, and if business associates make trustworthy recommen- dations, then all three of these strategies involve selecting from a pool of workers about whom much is already know. This would have the effect of lowering screening costs. and generally decreasing the variance in the quality of the applicant pool. The second group includes hiring from the labour office or from trade and technical schools. Presumably, some screening of applicants is done by these institutions. However, it is probable that this screening is less reliable than that provided by the previous mecha- nisms. This group, then, can be characterized as involving moderate screening costs, and a higher variance applicant pool. The final group of strategies involve hiring from the public at large, without pre- screening by outside individuals or institutions and without explicit recommendations from trusted family members, employees or associates. Clearly, this group of applicants would be more variable in quality, and selecting among them would entail higher screening costs. The evidence in all three panels of Table 8.19 indicates that larger firms are more likely to use the higher cost hiring mechanisms, so that the notion that they are Jess able to screen workers is plausible. If quality standards are lower or less important for unskilled production workers than for the rest of the labour force, this argument is also consistent with the much smaller size premia we find in that subsample. However, this evidence certainly does not constitute a direct test of the hypothesis, since these data could also be consistent with firms desiring to increase the average size of their applicant pool by choosing those hiring mechanisms which produce the most applicants, and then perhaps also raising wages to maximize the size of that pool. While both of these hiring mechanism arguments seem plausible, we cannot distinguish between them. The second class of efficiency wage arguments are those associated with worker behaviour on the job. The most familiar of these arguments is that firms increase wages as a mechanism to decrease shirking. The expected cost of shirking to a worker is the probability of being caught (and therefore fired) multiplied by the cost of job loss. Firms can obviously increase this cost either by raising the probability of catching the worker through increased monitoring, or by increasing the cost of job loss by raising wages relative to those the worker could receive in his next best option. If monitoring workers is more costly in larger firms, they will choose to use increased wages as a means of reducing shirking. This argument requires that a worker caught shirking can indeed by dismissed. Labour legislation in Zimbabwe, however, makes it extremely difficult to fire a worker classified as permanent by having been continuously employed for three months. The fact that this law is binding is reflected by the fact that many firms respond by maintaining a stock of casual employees under fixed term (but often renewed) contracts of less than three months. The use of such casual workers increases with firm size, and when asked, most firms indicated that casual workers generally received the same wages, allowances and benefits as permanent workers, being distinguished only by the fact that they can be more easily fired. Given the existence of apparently binding firing constraints, it is unlikely that firms are using higher wages as a means of deterring shirking among -permanent workers. A more plausible argument is that firms use higher wages as a means of reducing voluntary turnover. If maintenance of a stable labour force is more important for larger firms, because of higher training costs, larger investments i.n both specific and general human capital, a greater need for teamwork and coordination or greater recruiting costs, then these firms may find it profitable to raise wages to deter quits. As we saw in Table 8.11, there is indeed a strong negative correlation between quit rates and firm size. Workers in Jarger firms are also more likely to be receiving on the job or other training and, as we have seen, the screening costs associated with recruitment may also be greater in larger 122 firms. Greater capital intensity, which is likely to be associated with the use of more specialized equipment, may increase basic training costs as well. These factors indicate that larger firms may indeed have greater need for labour force stability, and hence offer wages above market clearing levels in order to generate it. If stability is more important in the administrative, managerial and technical occupations than for unskilled production workers, this explanation would also be consistent with the different magnitudes of the size premia across occupations. This seems quite reasonable. If, for example, the training any firm provides its white collar staff, such as accounts clerks, contains a large component of general skill develop- ment, there will be a high risk of worker turnover. Since the identification and screening of job candidates and the provision of training are likely to be quite costly, raising wages to retain such workers may well be a profit maximizing strategy for large firms. The fact that higher wages may indeed increase the number and quality of applicants and reduce voluntary turnover does not necessarily imply that these are the reasons firms choose to pay them. Efficiency wage theories were developed to explain these wage premia, and are therefore designed to be consistent with the facts. These arguments are consistent with our data, and could very well be at least part of the cause of the wage premiums associated with employment in larger firms. It would certainly be difficult to prove that they are not. Rent sharing A final potential explanation for the existence of these premia is that higher wages represent workers' share in the rents earned by larger firms. If larger firms are indeed earning higher profits, then workers may be able to capture their share through bargaining. If these rents are generated through privileges bestowed by government, the firm may view sharing these rents with workers as a relatively low cost means of maintaining good relations with government and guaranteeing continued access to privileges. In order for a rent sharing argument to explain the firm size premium, profits per worker (a measure of rents) must be positively correlated with firm size, and the inclusion of this measure in the earnings function must decrease the magnitude of the size effects well as exerting a positive influence on wages. Our measure of profits is accounting profits, calculated as the value of sales less measured costs of raw materials, wages, utilities and advertising. Although firms reported these measures for a variety of time periods, they were all converted to annual measures for the analysis. As an alternative, we have also used profits as reported by the firm. The correlation coefficient between the two measures is only .46, calling into question the reliability of the data, so the results should be interpreted with caution. There is no statistically significant simple correlation between firm size and either of these profit measures, which is the first piece of evidence against a rent sharing argument. Because we have more observations for the calculated profit measure, it has been included in the earnings functions reported in Table 8.20. Missing values for profits for 15 workers reduces our sample to 1594, and baseline size coefficients for this sample are presented in columns 1 and 3. Specifications including profits per worker are presented in columns 2 and 4. Inclusion of this variable does not reduce the magnitude of the size coefficients, although the coefficient on profits per worker is positive and significant. 123 Table 8.20: Earnings functions including controls for profits per worker (1) (2) 13J (4) Ln(Size) .159 (8.67) .163 (8.92) Medium .331 (3.6l) .335 13.57) Large .488 (5.392) .507 (5.392) Very Large .724 (7.18) .744 (7.18) Profits per Worker 5.49 x e" 5.41 x e" {2.213J (2.05) Adjusted R2 .503 .508 .489 .494 Dependent Variable is Log Hourly Earnings Absolute value of t-statistics in parentheses N= 1594 All regressions include controls for experience and its square, tenure, schooling as in Table 8.2, training, race, gender and location The positive correlation between profits and wages, however, is not enough to verify the existence of rent sharing behaviour. We also need to identify a rent sharing mechanism and a motive for firms to participate in it. Given the evidence on union effects. it is unlikely that rent sharing is driven by union bargaining, though other forms of bargaining are possible. It is, however, difficult to reconcile a bargaining driven rent sharing argument with the evidence that the wage premium for unskilled workers is smaller than the premium for others, unless unskilled workers are somehow excluded from the arrangements, or receive a small share of rents. As these worker comprise the majority of the labour force in almost all firms, it is difficult to see how such an arrangement could exist. Firms may choose to engage in rent-sharing as a show of "fairness" to their employees. If, however, this desire to be perceived as fair is intended to boost morale among workers or reduce turnover, then it is not really different from the efficiency wage argument discussed above. More profitable firms may be better able to finance the increase in wages needed to reduce turnover, or face higher costs of turnover, or be more concerned about the perception of fairness, particularly if their rents are the result of concessions or benefits received from government. Conversely, firms that pay higher wages may, because of better morale and less turnover, more than proportionally increase productivity, and hence have higher profits, even if they are not deliberately engaging in rent sharing. In general, the evidence does not provide strong support for the rent-sharing hypothesis. The positive correlation between profits and wages could be explained by efficiency wages hypotheses as well, and it is difficult to posit a rent sharing motive and mechanism consist- ent with the pattern of wage premia across occupations that does not ultimately revert to a story about turnover. It is not possible to design empirical tests which can definitively distinguish among these new institutional approaches. This is particularly problematic in that all of these hypotheses were developed to explain the same set of facts, and are different only in the unobservable dimension of underlying motivation. It does, however, appear that the costs of recruiting and training workers, particularly in white collar jobs, may be quite high for large firms. Raising wages will reduce turnover and protect firms' investments in screening and training their workers, and these arguments are the most plausible of the new institutional hypotheses about the firm size wage premium. 8.5 Employer size wage differentials - segmented labour markets approach The analysis in the previous section was based on the assumption that there is a single labour market for the manufacturing sector, so that the underlying wage determination 124 process, panicuJarly the returns to human capital, is the same for all firm sizes. The firm size wage premium, then, arises outside the context of the determinants of the basic returns to human capital. Suppose, on the contrary, that labour markets are segmented. so that the fundamental wage determination process varies from segment to segment. Differences in earnings across segments, then, will reflect differences in this underlying wage setting process. As an explanation for the employer size wage premium this would require that the market be segmented by firm size, and that the returns to human capital vary across these segments. In particular, we would need to see lower returns to human capital in the smaller firms. The analysis that follows is not a general test of the segmented or dual labour markets hypothesis. Such a test, as Dickens and Lang (1985) point out, would need to include the estimation of the appropriate definition of the segments and of the process by which workers are allocated into them. This discussion focuses on the much narrower question of whether an explanation in which the segmentation is assumed to correspond with size classes generates any useful information about the relationship between earnings and firm size. That is, do we see evidence of differences in the returns to human capital across the size classes we have defined, and can these differences explain the size premium? A strong version of the segmented labour markets approach is the dual labour markets hypothesis. This theory, which was put forth in part to explain the relatively poor perfor- mance of female and minority workers in U.S. labour markets, divides jobs into two types. Primary jobs have high wages, pleasant working conditions, stable employment, opportunity for advancement and positive returns to human capital, while secondary jobs have low wages, bad working conditions, unstable employment, no opportunity for advancement and zero returns to human capital. The proponents of this view argue that primary jobs are rationed, so that individuals may become trapped in secondary jobs even if they possess the human capital necessary to obtain employment in the primary sector. This taxonomy is similar to the notion that developing country labour markets are characterized by the existence of a formal and an informal sector corresponding to the primary and secondary labour markets. In the previous section we did discover a positive correlation between some aspects of working conditions and wages. As we explore the usefulness of a segmented labour markets approach for explaining the size premium, it will be useful to also think about whether jobs in small firms can usefully be characterized as secondary jobs. Table 8.21: Earnings function with size-variant coefficients small medium large very 18rge Constant -.0850 1.546) -.3940 (1.552) -. '220 (.700) '.2670 (.769) Experience .0450 (3.465) .0570 (5.032) .0820 (7.636) .0790 (7.243) Experience 2 ~.0005 (2.302) -.0006 (5.301) -.0012 (5.940) -.0013 (5.527) Primary .1770 (1.900) .5300 (2.570) . , 810 (2.296) .7580/2.213) Secondary .4450 (3.827) 1.1240 f4.7051 .8610 (7.606) 1.3940 (4.081) University 1.5880 (27.266) 2.0160 (5.371) 2.652017.474) 2.9090 (6.800) Vocational .2470 (.7530) .2060 (.7630) .1750 (.534) .0640 (.246) Poly technical .9610 (4.5251 1.1340 (4.0981 Former Apprentice -.3920 (2.506) .8470 (4.317) .5520 (3.854) .6920 15.5401 Dependent Variable is Log Hourly Earnings N = 1609 Adjusted R2 = .8391 Absolute value of t-statistics in parentheses 125 The first step in the analysis is to estimate an earnings regression in which the coeffi- cients are allowed to vary by firm size. The results of estimating such a function using the four size categories used in the previous section are presented in Table 8.21. The specifica- tion uses dummies for primary, secondary and university education where the returns are measured relative to no schooling, as this was the specification in which at least one individual in each size class was in each group. This formulation produces some serious anomalies. The returns to primary school for large firms are very low in comparison with the medium and very large firms. This is also true, although to a lesser extent, for the returns to secondary school. This results in the indication of no significant difference in the returns to primary school between small and large firms, though these returns are significantly different at the 10% level between large and very large firms. These results are probably due to some features of the dataset. Only 19 individuals in the entire dataset belong to the reference group of having no schooling, 1 in a small firm, 8 in medium, 7 in large and 3 in very large firms. This generates a situation in which outliers potentially have a great deal of influence. The difference between the large firms and the others is that while the average earnings of individuals with primary education only in large firms falls between those of similar individuals in the medium and very large firms, the earnings of individuals with no schooling are substantially higher in large firms than in any of the other size classes, and the variance smaller. It is these high earnings for the uneducated in large firms that are likely to generate the relatively lower returns to schooling for large firms. 6 One way of dealing with this problem is to re-estimate the earnings function omitting the measure of primary schooling so that the reference group includes both the unschooled and the primary schooled. Such a formulation yields the results presented in Table 8.22, which does indicate the present of differential returns to human capital across firm sizes. These results are less subject to the problems posed by having such a small reference group in each size class, and thus provide a more reasonable means of examining differences in the returns to human capital. Table 8.22: Earnings function with size-variant coefficients small medium large very large Constant .0880 (.508) .107 1.685) .051 (.355J .468 (2.872) Experience .0450 (3.458) .061 (5.112) .082 {7.6241 .081 (7.2381 Experience 2 -.0006 (2.284) -.0007 (3.515) -.0011 (5.927) -.0014 (S.529) Secondary .273 (2.328) .603 (6.290) .688 (8.094) .643 16.097) University 1.414 (13.679) 1.508 (S.154) 2.478 (7.80SI 2.1S6 (7.186) Vocational .247 (.754) .213 (.788) .175 (.534) .067 (.258) Poly technical .961 (4.527) 1.134 (4.014) former Appr- -.391 (2.500) .853 (4.2731 .652 (3.858) .693 (5.5521 entice Dependent Variable is Log Hourly Earnings N =1609 Adjusted R2 == .8387 Absolute value of t-statistics in paremheses I) Among the individuals with no schooling. all but two an: employed as production workcrs or support sl.ll.ff. The other two are a supervisor and an equipment maintenance worker, both of whom work in large finns. ThcK two individuals, whose wages are quite high given their education. may be driving this result. 126 Table 8.23 presents the results of F-tests of the hypotheses that the coefficients of each of the human capital variables are the same across firm size classes. The first column reports a test of the hypothesis that the coefficient is the same for all sizes, while the subsequent columns report the results of pairwise tests. With the exception of the constant, which is zero for all but the very large firms, and the returns to vocational school, which are zero throughout, the results indicate that all the coefficients are significantly different from one another at at least the 10% level. These overall differences, however, are primarily due to differences between the small firms and all other firms. With the exception of the constant, which may be picking up some of the premium associated with working in very large firms, there is no difference in the returns to human capital in the two largest size classes, so it does not appear that segmenta- tion explains the size wage premium at this end of the spectrum. With respect to the small and medium classes, smaller firms clearly provide lower returns to human capital than do large and very large firms. This is particularly obvious for the returns to apprenticeship, which are significantly negative for the smaller firms. This probably reflects differences in the nature of the apprenticeships these individuals completed, with small firm apprenticeships relating to production work, rather than to the technical training provided in large firms. Table 8.23: F-tests of equality of coefficients across firm sizes small- small- small- small- medium medium large = medium medium large very large:: large = very very = large large large very large Constant 1.46 0.00 0.03 2.54 0.07 2.56 3.64 (.2224) (.9366) (.8708) (.1112) (.7948) (.1097) (.0565) Ex per. 2.12 0.85 4.73 4.40 1.60 1.44 0.00 (.0961) (.3565) (.0298) (.0361 ) 1.2064) (.2303) (.9673) Exp2 2.29 0.32 3.22 5.17 1.71 3.47 0.56 (.0765) (.5700) (.0731) (.0232) (.1914) (.0626) (.4529) Second. 2.97 4.78 8.23 5.54 0.43 0.08 0.11 (.0307) (.0290) (.0042) (.01871 (.5105) (.77821 1.7746) Univ. 4.77 0.09 10.16 5.46 5.05 2.39 0.55 (.00261 1.76271 (.095) (.0196) (.02471 (.1223) (,4603) Vocation 0.08 0.00 0.02 0.19 0.00 0.'5 0.07 (.9710) (.93651 (.8771) (.6659) (.9290) (.6957) (.7952) Apprent. 12.74 24.06 20.52 29.36 .059 .046 .004 ('0000) (.0000) 1.00001 (.0000) 1.44431 (,4985) (.8463) Significance Levels in Parentheses Columns represent separate hypotheses The difference between the small and medium firms is not so clear. Returns to experience are the same for both classes, while returns to schooling are lower in the small firms. A comparison between medium firms and the other two size classes also generates a puzzle. Returns to human capital are generally the same, though the return to university education is lower in medium firms. These patterns are probably an artifact of the size categories used. Were this a general test of labour market segmentation, it would be appropriate to engage in additional empirical work in order to correctly identify the appropriate segments. Future research with this data will include a more general test for the existence of segmented labour markets that does attempt to estimate the appropriate segmentation patterns. At this stage, however, we are engaged in the much more simple task of asking if there is a pattern of segmentation that 127 corresponds with the wage effects we found using a particular pattern of size classes. We must draw our conclusions about the relevance of differences in wage setting processes for the size wage premium from this evidence, leaving the broader question of general labour market segmentation for later research. It seems clear that the difference in wages between small firms and the other firms is driven at least in part by differences in the return to human capital. This is not obvious with regard to wage differences between medium and larger firms, and clearly not the case at the upper end of the size spectrum. Since the evidence is clear only for the difference between small firms and all other firms, it is useful to re-estimate the earnings function including only two size classes, small and big, where big includes the medium, large and very large firms. These results are presented in Table 8.24. The first column is a regression for the whole sample, constraining the coefficients to be equal and including a dummy variable for big firms. This coefficient measures the size premium. The second two columns present the coefficients for each size class separately. Table 8.24: Earnings functions with size-variant coefficients combined small big Constant - .491 14.667) .088 (0.51) .'15 (1.25) Experience .075 112.00) .045 (3.46) .078 (11.511 Experience 2 ·. 0011 (8.898) .. 0005 12.281 .. 0011 (6.761 Secondary .683 113.331 .272 12.33) .710 113.01 University 2.239111.2761 1.414113.68) 2.299 (10.47) Vocational .136 (0.884) .247 10.75) .103 (.61) Former Appren· .692 16.691) - .391 (2.50) .780 (7.39) tice Big .661 (7.663) Adjusted R2 .3327 Dependent Variable is Log Hourly Earnings Absolute value of t-Statistics in parentheses N = 1609 Adjusted R2 for size variant regression = .8254 The coefficient on the size dummy, .661, represents a wage premium of 94% associated with working in a firm with more than 10 people. To estimate how much of this is due to the difference in the returns to human capital, we calculate what the workers in small firms would have earned had their human capital earned the rate of return available in big firms. Comparing these predicted earnings with actual earnings yields a measure of the increase in earnings associated with the greater returns to human capital available in big firms. The mean value of this wage difference for workers in small firms is .65. Since the earnings measures and the predictions are in logs, this implies that these workers' earnings would be an average of 65% higher if they worked in larger firms. This is approximately two thirds of the wage premium estimated in column 1. It appears that labour market segmentation does have a role to play in eXl'laining the relationship between earnings and firm size at the small end of the spectrum. ". This result raises two imponant questions. First, why are the returns to human capital lower in small firms? Second, what barriers to mobility allow these differences to persist? If workers could move freely across firms, they would leave the smaller firms and this would cause the returns to human capital to be equalized. One possibility is that these differences in returns to schooling reflect differences in worker quality. If workers in larger firms are, for example, more intelligent or more diligent, then they would be more productive, even holding schooling constant, than are workers in smaller firms. If larger firms are able to select the best workers from any 128 educational class, then each of these workers has more human capital per unit schooling, and the returns to the part of human capital we can measure would be higher. If these workers are in fact better, they would be able to learn more from their work experience, so the returns to experience would also be higher. If quality differences are the source of the differences in the returns to human capital, then workers are sorted, rather than rationed, into smaller firms. They would probably not be employable in larger firms, or able to get as high a return to their human capital even if they were. In this case, segmentation would be efficient, and the large share of the wage effect associated with it would in fact be consistent with market clearing in the labour market. A second explanation for the different returns to human capital is that these returns are greater where there is more opportunity for career advancement within the firm, and this advancement requires the existence of a set of differentiated and hierarchical jobs within the firm. Since small firms generally do not have such structures, opportunities for advancement through promotion and other career changes are limited. In a similar vein, if human and physical capital are complements, we would expect the returns to human capital to be larger where more physical capital is used, so the greater capital intensity of larger firms generates the greater returns to human capital. Both of the foregoing arguments require that there be a barrier to mobility between the segments, so that jobs in the larger firms are rationed. One potential source of this rationing is the efficiency wage stories discussed in the previous section. One of the primary implica- tions of efficiency wage theories is the existence of unemployment in labour market eqUilibrium. If larger firms are indeed using efficiency wages to reduce turnover or to increase the quality of the applicant pool, while smaller firms are not, this effectively limits the number of workers they will hire and generates a pool of workers who are rationed out of this labour market. Such workers may well take jobs in the smaller firms. As was pointed out in the discussion of endogenous selection, the dataset does not permit a good analysis of the determinants of how workers are allocated into firm sizes, so it is not possible to describe the rationing mechanism. It does seem clear, however, that luck and connections probably have a role to play in obtaining jobs in the larger firms. It is not clear, however, that the small firms in this sample constitute a secondary labour market of the type discussed in the dual labour markets literature. While jobs in these firms do pay less and generally have worse working conditions, and are characterized by few opportunities for advancement, it is not the case that there are no returns to human capital in these firms, or that their earnings profiles are. This does not imply that a segment in which all of these characteristics are true does not exist, but rather that it does not encompass all firms with 10 or fewer workers. It would be interesting to look at an even smaller size class to see if such" labour market exists. Unfortunately, there are only 43 workers in the sample who are in f.,lS with 5 or fewer workers, and estimates based on such a small sample are unlikely to be reliable. Overall, these results indicate that at least-part of the difference in wages between the small firms and other firms in the sample is due to differences in the return to human capital, and that these differences are persistent. Segmentation does not appear to have much explanatory power for the continued increase in wages as firm size rises beyond 10 workers, although such explanations as efficiency wages are consistent with the existence of a low wage labour market in the secondary sector. The most plausible explanation of the pattern of earnings across firm sizes appears to be a combination of an efficiency wage story that raises earnings and decreases employment in large firms, and a segmented labour markets story in which the workers who are excluded from the primary labour market obtain employment in 129 the small firm secondary labour market, where the returns to human capital, and hence earnings, are low. 8.6 Internal wage structure The previous sections have focused on the manufacturing labour market as a whole, and have attempted to characterize the process of earnings determination in that market by estimating the returns to human capital and exploring a number of explanations of the relationship between firm size and earnings. One clear implication of the results is that firms' individual wage setting practices differ, and these differences result in the premia associated with employment in large firms. Whether the firms use wages as a means of reducing turnover, or improving their applicant pool or sharing their rents with their workers, and whatever their reasons for finding these practices desirable, there are obvious differences across firms in the design of their compensation mechanisms. This section will continue our exploration of the wage determination process by examin- ing the distribution of wages across individuals within a firm and the occupational and earnings histories of individual workers during their tenure with their current employers. This analysis not only adds an additional dimension to our portrait of employee compensati- on in Zimbabwean manufacturing, but also reveals patterns of behaviour which add depth to our understanding of the differences between the quality of jobs and career paths in small and large firms. The analysis examines two types of evidence. The first is firm level data about the wages paid to workers in different occupational categories, which is used to derive a summary measure of the degree of wage dispersion in a firm, and to explore a number of hypotheses about the sources of this dispersion. The second type of evidence comes from individual work histories, particularly the experience of job change or promotion and the growth of wages over time. The analysis focuses on the determinants of individual promotion possibilities and the slope of earnings profiles, including both individual and firm level characteristics. Distribution of wages within firms Each firm reported the average wages paid to each of the 10 occupational categories of workers they employed, as well as the average allowances paid to each group. This allowed the construction of average total monthly earnings measures for each occupational category. From these data, a summary measure of wage dispersion as the ratio of the highest of these averages to the lowest could be constructed. In most cases this is the ratio of average management earnings to the average earnings of unskilled production workers or support staff. 130 Table 8.25: Wage dispersion within firm by firm size: small medium large very large Mean 2.12 8.39 13.51 13.53 Standard Deviation 1.22 6.45 8.72 6.64 by sector: food wood metel textile 9.98 9.16 10.02 9.32 Mean Standard Deviation 7.66 6.23 7.45 8.45 Measured as ratio of highest to lowest earnings in firm The mean of this ratio for the 198 firms reporting average wages by occupation is 9.6 (standard deviation 7.8). Table 8.25 reports mean values of the dispersion measure by firm size and sector. The table indicates that while there are no significant sectoral differences in the degree of internal wage dispersion, there are significant differences across firm sizes. In particular, small firms have much more compressed wage structures, with the highest paid employees earning only twice as much as the lowest paid. This low level of wage dispersion in small firms does provide some insight into the low returns to human capital in these firms. Even if additional education would allow an individual to obtain a "better" job in a small firm, perhaps as a supervisor or in administrati- on, these jobs carry a smaller wage premium, relative to unskilled labour, in small firms than in large ones. The more compressed wage structure implies that even where education allows workers to advance within the firms hierarchy, such advancement brings relatively little change in compensation in small firms. The relationship between wage dispersion and firm size could also be generated by a number of other characteristics which, while correlated with firm size, also exert a direct influence on the structure of compensation. This analysis will use a simple regression framework to explore some additional hypotheses about the source of variation in the degree of wage dispersion within firms. Part of the effect of firm size may be higher rates of unionization of larger firms. While the analysis in the previous section showed that unionization is not a primary source of size related differences in wage levels, that does not imply that unions do not have an effect on wage distributions. In particular, unions often impose strict seniority based wage scales, which would have the effect of raising the wages of the most senior blue collar staff. In order to avoid overlapping compensation between these workers and junior members of the white collar staff, firms may choose to raise the wages of white collar workers, which would push up the entire top end of the wage scale. This would wide the gap between the top and the bottom of the wage scale, increasing measured dispersion. A dummy variable indicating that the firm is unionized is also included in the regression. Another potential source of differences in the degree of wage dispersion is the age of the firm. Younger firms, which are generally also smaller, may have less differentiated job structures, in which individuals do a number of tasks and occupational and wage hierarchies are Jess clearly defined, resulting in smaller earnings differentials across occupational categories. These firms may also be able to offer less compensation to their white collar staff on the understanding that, if the firm is successful, greater earnings, profit shares, and even ownership may be possible in the future. This would reduce wages at the top of the distribution, and decrease dispersion. Since the effects of age may disappear quite quickly, both firm age (measured in years) and its square are included in the regression. Wage dispersion may also be affected by the ownership structure of the firm. Certain types of firms, particularly cooperatives and ·parastatals, may have an egalitarian ethos, and 131 choose more compressed wage distributions to reflect this philosophy. Multinational corporations, on the other hand, may set managerial wages in accordance with international corporate guidelines, and transfer management personnel across countries. Such practices would tend to increase the wages of management relative to factory workers. Among privately owned Zimbabwean firms, one might also expect to see a difference in managerial compensation between owner managed firms and those managed by hired managers. Owner-managers may not have reported their own earnings as wages, since they are not clearly employees, or reported only a fraction of their total earnings as wages, which would reduce measured wage dispersion in these firms. Equally important, firms with hired managers may increase managerial compensation as a means of generating better managerial performance without engaging in extensive and expensive monitoring. In order to capture these features of ownership structure, the regression includes dummies for the following ownership types: cooperative; parastatal; subsidiary of multinational; firm with external owners, including stockholders; and owner managed firms which are part of a family structure group. The reference category is an owner-managed firm that is not part of a group, which is also the category to which most small firms belong. The results of estimating a regression including these characteristics are presented in the first column of Table 8.26. The dependent variable is the ratio of the highest to the lowest wages in the firm. The results generally conform with our predictions. Older firms and unionized firms have more dispersed wage structures. Cooperatives and parastatals have less dispersed structures than do owner managed private firms, though the effects are not significant. Multinational corporations, family groups, and firms managed by hired managers have more dispersed wage structures. Interestingly. the size coefficients are not significant. This COUld. however, be an artifact of functional form. The second column of the table presents results of a specification in which the dependent variable, size and firm age are measured in natural logarithms. The pattern of signs is the same as in the first column, but the significance of coefficients is generally reversed. The high degree of correlation among the right hand side variables may be the root of this sensitivity to functional form, and makes it difficult to disentangle these effects. In general, however. the results confirm that the differences in wage levels across firm sizes we saw in the previous section are paralleled by differences in wage distributions. The higher wages paid in larger firms are complemented by more dispersed wage distributions. This helps explain the difference in the size of the wage premium between unskilled production workers and the labour force as a whole, since the more dispersed wage structure of larger firms would generate larger wages differences at the top of the distribution than at the bottom. The evidence also indicates that jobs in smaller firms, in addition to paying lower wages, also offer less opportunity for wage growth along the career path. This question can best be explored by examining individual career histories, which we will do in the next two sections. 132 Table 8.26: oLsestimate of the determinants of wage dis- persion (1/ (2) dependent variable ratio of highest to log of ratio of lowest earnings highest to lowest earnings Size .002 (0.992) Size 2 -5.40 x e· 7 (1.188) LnlSize) .2726 (6.113) Firm age .1566 11.909) Firm age' - .0014 (1.336) LnlFirmage) .022 (0.696) Union 5.112 (4.257) .520 (3.618) Cooperative -1.183 (0.623) - .504 (2.630) Parastatal -5.509 (0.804) -1.019 (1.553) Subsidiary of Mul- 5.351 12.971) .294 (1.6271 tinational Firm in Family Group 3.454 12.326) .1131 (0.755) Firm with hired man- 2.465 (1.679) .012210.081) ager Adjusted R2 .3444 .5625 OLS Estimate of the Determinants of Wage DispersIOn Absolute value of t-statistics in parentheses N 198 Internal promotion The career path of any individual, and hence his ability to reap the returns to his human capital and experience wage growth, depends in pan on receiving promotion to more responsible and hence better paying jobs. This section will explore the career paths of the 1609 workers in the sample, and relate the experience of job change to characteristics of the workers and their employers. The first task is to define promotion. Within the dataset, we can use information about the individual's current job and the job he had when he staned in the firm to identify individuals who have changed broad job classifications at least once during their tenure at the firm. If we take this as a definition of promotion, then 515, or 32 percent of the workers in the sample have been promoted_ Defining promotion in this way will understate the number of promotions actually received. Some workers will have changed jobs several times in their careers, and this measure only picks up one job change. More imponant, many promotions, particularly within the white collar jobs, will occur within a single job classification. For example, an individual promoted from accounts clerk to the head of the accounts receivable depanment will have been classified in administration in both instances, even though he has clearly received a promotion. In that sense, then,' our measure captures some aspects of job mobility, but certainly not aiL 133 Table 8.27: Share of workers having changed jobs by sector and firm size firm size sector Small 11.4% Food 34.9% Medium 23.9% Wood 20.7% Large 39.2% Metal 31.3% Very 41.5% Textile 33.8% Large Table 8.27 presents the fraction of workers who have changed jobs broken down by firm size and sector. Job mobility, in the sense in which we have defined it, is greater in larger firms, though the difference between large and very large firms is insignificant. Job mobility is also significantly lower in the wood sector than it is in the other three sectors, perhaps because these firms are also smaller. This negative correlation between job mobility and firm size is another important difference between small and large firms, particularly in terms of the desirability of such jobs. Table 8.28: Share of workers having changed jobs by initial and current occupation in percent initial occupation current occupation Management 6.7 Management 40.0 Administration 25.6 Administration 40.1 Commercial/Sales 30.0 Commercial/Sales 50.9 Supervisor 40.5 Supervisor 85.5 Technician 36.8 Technician 61.3 EQuipment 20.0 EQuipment 42.8 Maintenance Maintenance Skilled Production 23.0 Skilled Production 42.1 Unskilled Preduc- 25.0 Unskilled 15.9 tion Production Apprentice 42.5 Apprentice 17.8 Support Staff 65.7 Support Staff 17.8 Table 8.28 breaks down the total number of promotions according to the initial and the current jobs of the worker. The number in the table give the fraction of individuals who initially held (or currently hold) each type of job who have changed jobs since they began in the firm. The table has several interesting features. First, only 6.7% of those who began as managers were no longer managers at the time of the survey, which is not surprising since a move from management would be, in general, a demotion. Second, individuals who began their careers as support staff are much more likely to have changed jobs during their careers than are other workers. This is not surprising, since these jobs are generally the poorest jobs in the firm. Of the 176 individuals who have changed jobs from support staff, 105, or 60 percent, were working as unskilled production workers at the time of the survey. Third, 85 percent of the supervisors in the sample began their careers in the firm in other jobs, primarily as unskilled production workers. Not surprisingly, current unskilled production workers, apprentices and support staff are least likely to ever have changed job classificatj· ons. This description of promotion patterns across occupations, however, cannot answer the important questions about which kinds of individuals, in which kinds of firms, are likely to ... have experienced job mobility of this type. Table 8.29 presents the results of a probit analysis of the determinants of job mobility, or promotion, where the dependent variable takes the value 1 if the individual's current job classification differs from his classification when he was hired. The results indicate the following. 134 Table 8.29: Probit analysis of the probability of having changed jobs Tenure .1127(8.498) Tenure 2 .. 0023(5.734) Size .. 0001 (2.538) Food - .0411 (0.469) Wood - .4354(3.611) Metal - .1781(1.792) Primary - .4836(1.615) Some Secondary - .2250(0.741) Complete Secondary - .2671 (0.755) University - .3919(1.161) Vocational .3121 (·1 . 142) Polytechnic .8837(1.837) Former Apprentice .1342(0.908) On the job training .2358(1.836) Outside training .2849(2.584) Male .2570(2.528) European - .1241(0.547) Asian .2721 (0.906) Married .2102(2.077) Age at hire - .0129(2.553) Constant - .9644(2.644) Dependent Variable 1 if current job is not the same as initial job Absolute value of t-statistics in parentheses N = 1609 Chi,SQuare (20) = 262.87 Pseudo R2 = .1303 First, workers with longer tenure are more likely to have experienced job change, though this effect diminishes as tenure increases, reaching a peak at about 50 years. Second, workers in larger firms are more likely to have changed jobs. This is probably due in part to the fact that larger firms recognize more different job classifications, and have more differentiated employment structures. Third, workers in the wood and metal sectors are less likely to have changed jobs than are workers in food and textile firms, even holding size constant. Fourth, for the most part, education has no effect on promotion, though workers with polytechnic training are more likely to have changed jobs. The results also indicate that workers with primary education only are less likely than workers with no education to have experienced job change. This seems implausible, except for the fact that there is a very small number of individuals with no education in the sample, and these individuals are more likely to have begun their careers as support staff, the group with the highest promotion probability. In general, education exerts an influence on the job one gets when he is first hired, but appears to have no subsequent influence on career paths. Fifth, individuals who have changed jobs are more likely to be doing~training both on and off the job. The training may be related to the job change, so it seems reasonable to assume that the causation runs from job change to on the job training. The outside training, however, may reflect worker characteristics such as diligence or ambition which may be correlated with being selected from promotion, particularly since much of this training appears to be voluntary. The results also indicate that men are more likely to change jobs within a firm than are women. This may, however, be an artifact of our definition. As we saw earlier, women are heavily concentrated in the administration classification. The job changes and promotions they receive would not necessarily be captured by our measure, although the fact that they are equally likely as men to be unskilled production workers, but generally less likely to 135 change job classifications, does indicate that there may be some broad difference in the advancement possibilities afforded to women. Married workers, who employers generally consider to be more stable long term prospects, are also more likely to have changed jobs, while workers ,who were older at the time they were hired are less likely to have done so. Older workers may be hired in at higher levels, and in effect be combining job change experience with changes in firm. None of these results are particularly surprising, although they do raise some interesting points about the difference between jobs in small and large firms. One of the common compensating differentials arguments (see Brown, Medoff and Hamilton, 1990), is that jobs in small firms offer more room for advancement. This evidence indicates, however, that workers in small firms are less likely to experience job changes. This evidence, combined with the discussion of wage dispersion within the firm, indicates that workers in larger firms have better career paths, both in terms of the probability of changing jobs and in the returns to such a job change in the form of higher wages. Jobs in small firms, it appears, offer lower wages and less room for advancement, so that the notion of detining them as a "secondary" labour market is not unreasonable. While the information about wage compres- sion and job mobility provides some information about a worker's potential earnings path, the dataset allows us to look at the path workers earnings have actually followed, and this is the subject of the next part of the analysis. Earnings profiles and the returns to tenure Each worker was asked to report the wages he received when he began working for his current employer. Using these data along with current wage information, we can construct a measure of the average annual rate of wage growth for each individual. While this approxi- mation drawn from two observations will not give us a picture of the exact shape of each individual's profile, it does provide some sense of the steepness of the profile, and hence allows some discussion and comparison of the determinants of the evolution of individual earnings over time. Rates of growth can be computed for both nominal and real wages. The real wage measure used a series of the consumer price index from 1970 to 1993 (1980= 100) to convert both initial and current wages to constant 1980 Zimbabwe dollars. Because the cpi series only went back as far as 1970, the 98 workers who began their current employment before then were omitted, as were an additional 43 individuals who did not report their starting wages. The 139 workers with tenure of less than one year were also omitted. Most of them had not yet received raises, so their nominal rate of wage growth was zero, and the real rate quite negative, although we really do not know what will happen when the receive their first increases. For purposes of comparability, our discussion of nominal growth rates will use the same sample. For this sample of 1329 workers, the mean of the individual annual growth rates was -1 percent per year in rea) terms, and 20 percent per year in nominal terms. This evidence of negative real wage growth on average masks a great deal of variation across cohorts in the sample. There has been dramatic inflation in Zimbabwe since the mid 1980's, and it is likely that almost all workers experienced declining real wages during that period. Workers who began their current jobs after the mid 1980's will exhibit wage profiles dominated by the effects of this inflation. Tables 8.30 and 8.31 present mean growth rates broken down by firm size and five year tenure cohorts for real and nominal wages respectively. 136 Table 8.30: Average annual rates of growth of real wages by firm size and year of hire all small medium large very large All 1329 102 443 402 382 -.001 -.037 -.014 -.013 -.00009 .118 .157 .143 .095 .093 Hired 1970 to 134 0 41 49 44 1974 .031 .025 .027 .042 .046 .046 .036 .055 Hired 1975 to 210 10 64 74 62 1979 .032 .027 .025 .024 .050 .055 .062 .049 .047 .064 Hired 1980 to 296 6 103 84 103 1984 .008 -.032 -.002 .009 .020 .074 .098 .086 .062 .066 Hired 1985 to 354 33 113 108 100 1989 -.033 -.044 ·.037 -.045 -.010 .087 .085 .102 .061 .090 Hired 1990 to 335 53 122 87 73 1992 -.051 -.045 -.037 -.047 -.082 .189 .202 102 .161 .112 Each cell contains: Number of observations Mean Standard Deviation Table 8.31: Average annual rates of growth of nominal wages by firm siZe and year of hire all small medium large very large All 1329 102 443 402 382 .201 .205 .293 .191 .208 .150 .190 .192 .129 .105 Hired 1970 to 134 0 41 49 44 1974 .176 .170 .171 .187 .053 .054 .041 .062 Hired 1975 to 210 10 64 74 62 1979 .197 .187 .189 .187 .218 .064 .073 .057 .054 .076 Hired 1980 to 296 6 103 84 103 1984 .186 .139 .174 .187 .200 .086 .114 .101 .073 .078 Hired 1985 to 354 33 113 108 100 1989 .184 .164 .180 .169 .213 .107 .134 .118 .071 ., '2 Hired 1990 to 335 53 122 87 73 1992 .243 .240 .265 .237 .214 .254 .233 .324 .222 .157 Each cell contains: Number of observations Mean Standard Deviation This evidence indicates, first, that the overall pattern of declining real wages did indeed begin in the mid 1980's, as it is the cohort hired in 1985 or later for whom mean rates of real wage growth are negative. This occurred despite the fact that individuals hire in 1990 or later experienced much more rapid increases in nominal wages. The evidence on real wages seems to imply that workers with longer tenure have higher average annual rates of wage growth. This would mean that the overall shape of the tenure 137 earnings profile is convex, getting steeper at higher levels of tenure, an unusual result if it were in fact true. The nominal growth rate evidence, however, exhibits the opposite pattern, with individuals with longer tenure having lower average rates of wage growth, correspon- ding to a more typical concave tenure profile. These results are probably driven by the variation in rates of inflation over time. Individuals with longer tenure have spent less of their working lives experiencing real wage decline, making their overall average growth rate higber. This implies that we are not going to be able to use the relationship between grol,l,'th rates and tenure to draw a reliable picture of the actual shape of a general wage tenure profile, although we will wish to control for tenure in econometric work. The evidence also indicates a weak relationship between wage growth and firm size, with workers in small firms experiencing lower real growth rates than those in larger firms, particularly those in very large firms. This relationship is not apparent in the nominal data, nor is it very strong. There are no sectoral differences in wage growth rates. Table 8.32: Determinants of individual wage growth dependent variable real nominal real nominal Constant -.073 (2.161 ) .2521 (5.791 ) -.0896 (3.249) .2353 (6.399) Tenure .0091 (3.627) -.0078 (2.229) .0093 (3.900) -.0076 (2.218) Tenure 2 -.0002 (1.887) .0002 (1.940) ·.0002 (2.087) .0002 (1.890) Primary -.0046 (0.388) -.0077 (0.526) -.0043 (0.384) -.0077 10.559) Some Secondary .0137 (1.015) .0132 (0.761 ) .0145 (1.192) .0138 (0.898) Comp Secondary .0550 (2.774) .0614 (2.322) .0521 (2.693) .0579 12.254) University -.0777 (4.152) -.0914 (3.601) -.0708 (3.765) -.0838 (3.338) Vocational -.0537 (3.0511 -.0703 (3.021) -.0517 (2.957) -.0682 12.963) Polytechnic -.0251 (1.577) -.0322 (1.559) -.0216 (1.620) -.0261 11.462) Former Apprent ·.0022 (0.173) -.0002 (0.013) -.0006 (0.051) -.0010 (0.064) Male -.0156 (1.138) -.0240 (1.297) -.0143 (1.142) -.0227 (1.333) European -.0527 (4.183) -.0660 (3.306) -.0566 (4.315) -.0708 (3.455) Asian -.0049 (0.278) -.0029 (0.9061 -.0101 (0.564) ·.0028 (0.114) Harare -.0095 (0.6241 -.0152 (0.747) ·.0111 10.717) -.0171 (0.812) Bulawayo .0163 (0.945) .0217 (0.9631 .0168 (0.979) .0222 to.9751 Union -.0005 (0.034) ·.0068 10.312) -.0030 (0.196) -.0103 (0.475) Minimum Wage -.0274 (3.240) ·.0371 (3.313) -.0201 (2.508) -.0290 12.656) Size .000006 (2.59) .000008 12.59) .000002(1.168) .000005 11.42) Cooperative .0468 (0.987) .0451 (0.801 ) Parastatal .0569 (6.424) .0470 (3.798) European Owner .0178 (2.108) .0193 (1.622) Percent Foreign .0003 13.016) .0003 12.8391 Adjusted R2 .0937 .0352 .1030 .0409 Dependent Variable is average annual rate of wage growth Absolute value of t-statistics in parentheses N = 1329 Table 8.32 reports the results of regressing each of the growth rates on a set of individual and firm characteristics which were relevant determinants of the level of wages as discussed in the section on the employer size wage premium. The first two columns include human capital controls, firm size. a union dummy, and a dummy indicating that the firm feels that the minimum wage is binding. The results for the real and nominal growth rate measures are strikingly similar. with the exception of the reversed pattern of signs on the tenure measure. The results indicate that individuaJs working for larger firms have experienced higher rates of wage growth in both real and nominal terms. This effect is not due to unionization. as the coefficient on the union dummy is not significantly different from zero. The effects of education indicate that completion of secondary school raises the rate of growth of wages relative to having no schooling, but university education decreases wage 138 growth. This is likely to be due to the fact that university educated persons enter the firm in high level jobs, and are less likely, therefore, to have as much room for wage-increasing career advancement than are secondary school graduates, who begin their careers at a generally lower level. This explanation probably also explains the slower wage gro\l/th of Europeans, who also start their careers at a higher level. A panicularly interesting result is the significant negative coefficient on the dummy variable indicating that the firm perceives the minimum wage as binding. This effect can be explained if such firms raise their wages only as the minimum wage rises. If the minimum wage rises more slowly than inflation, so its real value is falling, then firms for whom this value is a binding lower bound on wages will be paying declining real wages. The Econom- ist Intelligence Unit Country Repon for Zimbabwe, 1990, indicated that the real value of the minimum wage had indeed been falling. The results in columns three and four include ownership structure variables: dummy variables indicating that the firm is a cooperative, a parastatal, or a firm with Zimbabwean owners of European descent, and a continuous measure of the share of foreign ownership. These results indicate that parastatals, European owned firms and foreign firms all provide higher wage growth rates. The inclusion of these measures also reduces the effect of firm size, rendering the coefficient on size insignificant in both the real and nominal specificati- ons. The rapid inflation from the late 1980's, however, distorts our measurement of the slope of these earnings profiles, and hence our understanding of their determinants. The wage profile includes both the returns to experience in the firm and any adjustments to for cost of living, and the two cannot be separated. If, for example, wages are adjusted based on past levels of inflation, real wages may fall even if employers are attempting to generate positive returns to tenure and upward sloping earnings profiles. Although the two components of the slope of the wage profile cannot be distinguished, we can attempt to minimize the effect of the inflation by focusing our attention on the cohan of individuals hired prior to 1985, for whom the rapid inflation of the later period consti- tutes a smaller share of their total earnings histories. These individuals generally exhibit positive annual real earnings growth. This does not totally eliminate the problem, since these individuals' earnings histories include the period from 1985 to 1993, but more of their measured wage growth will actually be related to the returns to experience. Since it is possible that the coefficients reported in Table 8.32 were driven largely by the difference between individuals with positive and negative rates of real earnings growth, and therefore by the effects of inflation, the contrast of these results with those for a restricted sample will provide some insight into the slope of the returns to experience. Results for the restricted sample are reponed in Table 8.33. These results differ significantly from those for the full sample. First, the effects of tenure have the same sign pattern for both the real and nominal measures of wage growth, though these effects are smaller than for the full sample. This is consistent with the hypothesis that the relationship between tenure and wage growth in Table 8.32 is driven by differences in the experience of inflation, and the difference between individuals who experience positive and negative real rates of wage growth. Second, the effects of being male or resident in an urban area are larger in this sample, having a significant positive effect on the slope of earnings profiles. Third, the coefficients on the ownership structure variables are different, though the signs are the same. The results in Table 8.33 indicate that parastatals and cooperatives offer steeper earnings profiles, but European and foreign owners do not. This may imply that these firms offered their workers better protection against inflation than did other firms, but did not offer greater returns to 139 experience on the job. Table 8.33: Determinants of individual wage growth _. individuals hired before 1985 dependent variable real nominal real nominal Constant -.1057 (3.193) .0863 (2.674) -.1039 (3.706) .0885 (2.649) Tenure .0048 (1.942) .0031 (1.013) .0045 (1.7471 .0028 (0.870) Tenure 2 -.00007 (1.17) -.00013 (1.60) -.00007 (0.99) -.0001 (1.419) Primary .0069 (O.S77) .0087 (0.632) .0076 (0.643) .0095 (0.699) Some Secondary .0336 12.673) .0396 12.67S) .0336 12.668) .0397 (2.782) Camp Secondary .0480 (2.793) .0569 12.887) .0427 (2.646) .0511 12.748) University -.0585 (2.879) -.0697 (2.94S) -.0558 (2.636) -.0669 (2.716) Vocational -.0328 (1.626) -.0389 (1.652) -.0292 (1.510) -.0350 (1.538) Polytechnic -.0227 (1.368) -.0256 (1.356) -.0276 (1.994) -.0309 (1.963) Former Apprent -.0154 (1 .137) -.0159 (1.000) -.0170 (1.183) -.0179 (1.055) Male .0169(2.791) .0196(2.796) .0174(2.839) .0203 12.844) European -.0484 (3.801) -.0562 (3.848) -.0458 (3.644) -.0532 (3.6871 Asian .0764 (1.491) .0909 (1.530) .0761 (1.495) .0906 (1.5341 Harare .0191 (2.628) .0219 (2.604) .0181 (2.576) .0208 (2.556) Bulawayo .0135 (1.7171 .0151 (1.654) .0132 (1.724) .0148 (1.665) Union .0239 12.164) .0289 (2.256) .0242 (2.21 1) .0293 (2.302) Minimum Wage -.0049 (0.733) -.0057 (0.731) -.0054 (0.749) -.0063 (0.755) Size .00001 (3.53) .00001 (3.56) .000009 r2.742J .00001 (2.78) Cooperative .0459 (2.753) .OS34 (2.729) Parastatal .0487 (4.679) .0535 (4.404) European Owner -.0018 (0.304) -.0022 (0.322) Percent Foreign .00007 (1.09) .00007 (1.0S) Adjusted R2 .1207.1081 .1225 .1094 Dependent Variable is average annual rate of wage growth Absolute value of t-statistics in parentheses N= 640 Most interestingly, the effects of unionization are larger than in the results in Table 8.32, and are significantly positive. One interpretation is that unions impose strict seniority wage scales, generating a return to experience, but are not as effective in mandating cost of living increases and protecting workers from inflation. This would be particularly relevant if union contracts are negotiated with a less than annual frequency, so that cost of living adjustments lag behind the rate of inflation, but seniority based wage scales are binding. On the other hand, the effect of binding minimum wages is eliminated in the sample of workers hired before 1985. Minimum wages do not necessarily impose any structure on wage profiles, and therefore do not generate any growth of individual wages over time, though movement in the minimum wage will determine the growth of real wages for those individuals for whom it is binding. Finally, the size effects are larger in the sub-sample, and persist even when ownership structure variables are included. Larger firms do provide steeper wage profiles, though they may not have done better than other firms .in protecting workers from inflation. These steeper wage profiles would be consistent with many of the explanations offered for the higher levels of wages paid by large firms. If these firms provide more on the job training, postponing compensation and making earnings profiles steeper would help reduce worker turnover and protect the firm's investment. More on the job training would also imply faster rates of productivity growth for these workers, which would also make wage profiles steeper. Steep wage profiles, along with high wage levels, would also act to reduce turnover, and the results in Table 8.11 indicate that quit rates in large firms are indeed lower. Although the current data set does not allow for detailed exploration of on the job 140 training, so no definitive comparison of the training practices of large and small firms can be made, it seems reasonable to assume that the more sophisticated technologies and more hierarchical job structures used by large firms will cause them to provide more training. Certainly, these firms wage setting behaviour, as well as their workers' turnover behaviour, is consistent with the existence of training differences across firm sizes. The next round of the survey will include more detailed measures of on the job training, and provide a better opportunity to address the training issue. The results of this section conform with our discussion of wage dispersion and job change rates. If workers in smaller firms are less likely to change jobs, and less likely to be rewarded for such advancement due to more compressed wage distributions within the firm, we would expect these workers to experience lower rates of wage growth. The evidence in Table 8.33 indicates that they do. Overall, the results of this examination of wage patterns within firms and individual job and wage histories reinforce the idea that there are differen- ces in wage setting mechanisms across firm sizes. 8.7 Summary and conclusions This chapter has explored a number of issues concerning earnings determination in Zimbab- wean manufacturing. The results indicate that wage determination in this labour market does conform with the general pred ictions of human capital theory, and the returns to education are similar to those found in other studies. We also found evidence of discrimination by race and gender, which also conforms with results for other countries in Africa. The chapter includes a lengthy discussion of the relationship between earnings and firm size, the results of which indicate the presence of a premium associated with employment in larger firms that cannot be explained by differences in worker quality or job characteristics, nor by the influence of unions or minimum wages. While labour market segmentation explains some of these differences, it does not explain the wage differentials between large and very large firms. Efficiency wage arguments, though not subjected to explicit tests, do appear to have some explanatory power. The last section of the paper uses data on internal wage distributions and individual job and earnings histories to continue the exploration of the differences in firms' compensation practices across firm sizes. The analysis finds that larger firms have more dispersed internal wage structures, offer a higher probability of career advancement through job changes, and provide steeper earnings profiles. What do these results mean? On a broad level, they point to two main conclusions. First, the manufacturing labour market in Zimbabwe does appear to include a secondary segment associated with small firms. Workers in these firms receive low wages, work long hours in unpleasant conditions, receive minimal returns to their human capital, have little opportunity for advancement through promotion, and can expect slow growth of their real wages over time. In all these dimensions, these can be characterized as "bad" jobs. This would be particularly troublesome if, as is likely, workers are rationed rather than sorted into these jobs. The implications of this for individual welfare depend on individual mobility out of this sector, which cannot be examined using these data. It also depends on whether small firms grow, and how their compensation behaviour changes as growth occurs. The evidence here, however, ind icates that workers in small firms are worse off than their counterparts in large firms. Second, and more surprising, the evidence indicates that the functioning of the labour market in Zimbabwean manufacturing is not much different from the functioning of developed country labour markets. In particular, while the employee size wage premium is 141 significantly larger in Zimbabwe than, for example, in the U.S., it does not appear to be driven by the intervention of unions or minimum wages, or by other "non-competitive" forces. The efficiency wages arguments which appear to provide the best explanation for this premium are consistent with competitive labour markets. The differences between Zimbabwe and developed countries appear, in general, to be attributable to differences in the level of development of the manufacturing sector, rather than to fundamental differences in the way markets function. If the returns to schooling are higher, educated workers are also more scarce. Even the larger magnitude of the size premium in Zimbabwe can be explained if the variance in worker quality is larger, so qualified applicants are both more scarce and less easily identified, or if less experience with industrial production implies that larger firms must devote more resources to training and are more concerned about turnover. Although blind application of theories developed to explain features of developed country labour markets would be unwise, these "standard" theories do, in fact, appear to be relevant for this case, and the manufacturing labour market in Zimbabwe seems to share many features of similar labour markets elsewhere. The final conclusion which can be drawn from this research concerns the quality and usefulness of the data itself. The labour markets data drawn from the RPED survey provide an extremely rich source of information and create the opportunity for deep empirical exploration of a large number of issues, of which those addressed here are just a sample. There are a number of additional questions which can, and will, be explored in future work. 142 References Armitage, J. and R. Sabot, 1991, Discrimination in East Africa's Urban Labour Markets, in N. Birdsall and R. Sabot, editors, Unfair Advantage: Labour Market Discrimination in Developing Countries, Washington: The World Bank. Brown, C. and J. Medoff, 1989, "The Employer Size Wage Effect,· Journal of Polirical Economy 97(5): 1027-1059. Brown, C., J. Hamilton and J. Medoff, 1990, Employers Large and Small, Cambridge, MA: Harvard University Press. Dickens, W. and K. Lang, 1985, "A Test of Dual Labour Market Theory .. American Economic Revie.,v 75 (4): 792-805. Garen, J., 1984, "The Returns to Schooling: A Selectivity Bias Approach with a Continuous Choice Variable,· EcorlOmerrica 52(5): 1199-1218. Idson, T. and D. Feaster, 1990, "A Selectivity Model of Employer Size Wage Differentials," Journal of Labour Economics 8(1): 99-122. Knight, J. and R. Sabot, 1991, "Labour Market Discrimination in a Poor Urban Economy," in N. Birdsall and R. Sabot, editors, Unfair Advantage: Labour Market Discrimination ill Developillg Countries, Washington: The World Bank. Lang, K., 1991, ·Persistent Wage Dispersion and Involuntary Unemployment: Quanerly Journal of Economics: 181-202. Little, LM.D., D. Mazumdar and J. Page, 1987, Small Manufacturing Enterprises: A Comparative Study of India and Other Economies, New York: Oxford University Press. Montgomery, J., 1991, "Equilibrium Wage Dispersion and Interindustry Wage Differentials," Quarterly Journal of Economics: 163-179. Schaffner, J., 1994, "Larger Employer Wage Premiums in Peru," Mimeo, Stanford Upi versity. van der Gaag, J. and W. Vijveroerg, 1989, "Wage Determinants in Cote d'Ivoire: Experience, Credentials and Human Capital,· Economic Developmefl/ and Cultural Change 37: 371-82. Weiss, A., 1980, "lob Queues and Layoffs in Labor Markets with Flexible Wages," Journal of Political Economy 88(3): 526-38. Weiss, A. and H. Landau, 1984, "Wages, Hiring Standards and Fil'lJ! Size,· Journal of Labor Economics 2(4): 477-499. White, H., 1980, "A Heteroskedasticity-Consistent Covariance Matrix and a Direct Test for Heteroskedasticity," Econometrica 48: 721-46. 143 9 Labour demand Ronald Chifamba Wouter Zant In this chapter we examine the determinants of labour demand by estimating labour demand functions. We begin by deriving a standard labour demand function from theory and then estimate this function, separately for skilled and unskilled employment. In order to formulate a relationship that can be estimated we follow a standard procedure: a production function is postulated with capital and labour as arguments. The objective of the firm is to maximise profits, taking the capital stock as given. Demand for labour can then be derived from the first order condition as a function of the capital stock and the price of labour. In the estimations a loglinear transformation is applied so that the coefficients can be interpreted as elasticities. All data for the statistical work are taken from the introductory section of the survey, the general firm questionnaire and the labour market questionnaire. Total labour use, the dependent variable in the statistical analysis, is not observed directly. With the help of the number of workers by type and by sex and the information on the average number of hours worked by type of worker, a labour use variable is created in full time equivalents (fte). A full time working week is assumed to be 45 hours: workers reporting more than 45 hours per week are taken to be full" time workers (1 fte). Available data on the physical capital stock are based on the replacement value and on the sales value of the equipment. Both concepts of the physical capital stock have been experimented with in the estimations. The price of labour is calculated as the sum of the average base wage plus the value of allowances (both cash and in kind). The resulting earnings figure was transformed to 'per month' data with the help of the information on the pay period and aggregated over the various categories of employment distinguished in the questionnaire. In the aggregation all categories were weighted with employment per category. In the estimations we also experimented with a narrow concept of the price of labour, namely the base wage (and also weighted with the volume of labour by type of labour). Categories classified as unskilled workers are 'other production workers', 'apprentices and formal trainees' and 'support staff'. Skilled workers contain the remaining categories, namel y 'management', 'administrative/clerical', 'commercial/sales', 'supervisors/foremen', technicians', 'equipment maintenance' and 'skilled production workers'. The labour demand equations are estimated with OLS. The results are reported in Tables 9.1 and 9.2. Some preliminary remarks are in order. Estimations with the two concepts of the physical capital stock showed that the replacement value of the capital stock in all estimations generated a 20 to 30% better explanation (in terms of adjusted R::) of the demand fro :oour than the sales value of the physical capital stock. Therefore we do not report estimation results with the latter concept of the physical capital stock. Also the narrow concept of the price of labour did not perform well: the coefficient of this variable turned out to have the wrong sign in most specifications although often significant and whenever the sign was correct, it was not or only weakly significant. For this reason we do not report estimation results with this concept of the price of labour. In Table 9.1 the estimation results of demand for skilled labour are presented. All economic variables perform reasonably well: the capital stock is highly significant as is the own price elasticity; the cross price elasticity is only weakly significant; all these coefficients have the correct sign. The size of the own-price elasticity (in between -0.388 and -0.437) compares reasonably well with other studies. Controlling for regional and sectoral differ- ences does not seem to affect the estimation outcome substantial! y. An F-test on zero 144 restrIctIOns on the sectoral and regional dummies must be rejected in case of regional dummies and accepted in case of sectoral dummies at the conventional levels of signifi- cance. 1 In Table 9.2 the estimation results of demand for unskilled labour are presented. Again, all economic variables perform reasonably well: the capital stock is highly significant as is the cross price elasticity; the own price elasticity is only weakly significant; all these coefficients have the correct sign. The size of the own-price elasticity (in between -0.154 and -0.241) is also not out of range of results in other studies. Again, controlling for regional and sectoral differences does not seem to affect the estimation outcome substan- tially. An F-test on zero restrictions on the sectoral and regional dummies must be rejected in case of sectoral dummies and accepted in case of regional dummies at the conventional levels of significance. 2 Table 9.1: Demand for skilled labour dependent variable: labour of skilled employees (in 1te) estimation method: OLS number of observations: 168 mean of dependent variable: 3.07325 independent variables (1 ) (2) (3) (4) constant -3.709 (5.2) -3.854 (5.3) -3.371 (5.3) -3.548 (5.6) capital 0.593 (18.7) 0.600 (18.8) 0.600 (19.1) 0.609 (19.3) price of '-0.388 (3.8) -0.411 (4: 1) -0.415 (4.1) -0.437 (4.4) skilled labour price of 0.106 (1.0) 0.147 (1.3) 0.092 (0.9) 0.139 (1.3) unskilled labour food 0.166 (0.8) 0.209 (1.0) textile 0.002 (0.0) 0.009 (0.0) wood 0.354 (1.8) 0.356 (1.8) Harare 0.187 (0.9) 0.225 (1.1) Bulawayo -0.213 (0.9) -0.174 (0.8) ssr 0.0124 .0129 .0128 0.0133 adjusted R2 0.743 0.736 0.741 0.733 notes: all variables except dummies are transformed logarithmically figures in brackets next to the coefficients are the (absolute) values of the t-statistic ssr = sum of squared residuals adjusted R2 = coefficient of determination adjusted for degrees of freedom The results imply that demand for unskilled labour reacts more strongly to the price of skilled labour - apparently there is strong substitution - than to its own price. However, with skilled labour the conclusion is slightly different: demand for skilled labour reacts much more strongly to the own price, the price of skilled labour, and much less strongly on the price of unskilled labour. Apparently substitution in this direction is hardly important. Hence, in the Zimbabwean labour market the'influence of the price of skilled labour is felt both in the demand for skilled labour and in the demand for unskilled labour. Finally, from I The F-statistic is calculated as F=[(ssrr-ssru)fdlf[ssri(n-k») and has a value of 1.71 in case of zero restrictions on sectoral dummies. and 3.21 in case of zero restrictions on regional dummies. Critical values at 5% significance are respectively F(3, 159)=2.67 and F<2,159)=3.06. 2 The F-statistic has a value of 5.21 in case of zero restrictions on sectoral dummies, and 0.71 in case of zero restrictions on regional dummies. Critical values are as above. 145 the calculations it appears that there is a significant different regional impact in the case of skilled labour, and a significant different sectoral impact in case of unskilled labour. Two qualifications should be noted. First, while there is some justification in treating the capital stock as given in a cross-section estimate, one can object that firms will adjust both capital and labour in response to changes in wages so that the capital stock cannot be treated as exogenous. Secondly, the analysis in the previous chapter suggested that efficiency wage explanations may have some relevance in Zimbabwe. In that case differences between firms in labour demand cannot be explained in terms of firms facing different factor prices: employment and wages would both be determined endogenously. We will be in a better position to resolve these issues when a second round of data becomes available. Table 9.2: Demand tor unskilled labour dependent variable: labour of unskilled employees (in he) estimation method: OLS number of observations: 168 mean of dependent variable: 4.12447 independent variables I' ) (2) (3) (41 constant -5.485 (8,1) -5.492 (8.2) -4.348 (7.0) -4.297 17.0) capital 0.490 (16.2) 0.492 (16.5) 0.495 116.0) 0.498 (16.3) price of 0.378 (3.9) 0.380 (4.11 0.325 (3.3) 0.336 13.5) skilled labour price of -0.156 (1.5) -0.154 (1.5) -0,233 12.2) -0.241 12.4) unskilled labour food 0.340 '1.7) 0.345 11.7) textile 0.658 (2.7) 0.662 12.7) wood 0.697 13.7) 0.703 (3.8) Harare 0.076 (0.4) 0.163 (0.8) Bulawayo 0.030 (0.1) 0.136 (0.61 ssr 0.0112 0.0113 0.0123 0.0124 adjusted R2 0.769 0.772 0.752 0.754 notes: all variables except dummies are transformed logarithmically figures in brackets next to the coefficients are the (absolute) values of the t-statistic ssr = sum of squared residuals adjusted R2 = coefficient of determination adjusted for degrees of freedom 146 10 Transaction costs and institutional environment Jan Bade Ronald Chifamba 10.1 Introduction In this chapter we shall describe and analyze what nowadays is typically the field of industrial and new institutional economics. In a world where competition and markets are imperfect institutions matter. Transaction costs, such as the cost of trade credit, contract enforcement and access to finance or foreign exchange will be looked at. But also the effect of infrastructure, networks and business support services will be reviewed. There are a number of general remarks that can be made about the importance of the environment in which a business operates. A firm operates in an uncertain world where information is limited and costly. Good management will decide on production and investment while recognizing the trade off between collecting information and taking risk. But firms do not all have the same opportunities. There may be large differences between firms when we look at the way they interact with the outside world. Relationships with suppliers and clients are obviously of great importance. Not all firms are equally endowed to enforce timely delivery of raw materials by suppliers and timely payment by clients. Sharp competition may be countered by privileges or by forward or backward integration. As Tirole states I in a comment on the Coase theorem (Coase (1937»: "The efficiency of private contracts generally requires perfect information, the absence of transaction costs, and the absence of third party externalities" Problems, such as infrastructure problems, problems with getting foreign exchange, shortages of raw materials, equipment break downs, etc. may be solved by some firms because they have good networks, whereas other firms have to cope with these problems on their own. Size of the firm, race of the owner, access to credit, access to government institutions and the legal system are among the items that shall be dealt with. The chapter is structured in the following way. In the section 10.2 we look at the extent of competition and the way firms set their prices. In 10.3 contractual arrangements with suppliers and clients are analyses, concentrating of the form of payment. Section lOA deals with the typical phenomenon of in kind lending and borrowing that was found to be of non trivial importance in Zimbabwe. In 10.5 contract enforcement and conflict resolution are analyzed. Infrastructure problems are dealt with in section 10.6, while the use of institu w tionalized business support is assessed in 10.7. A few remarks on problems with getting foreign exchange are made in section 10.8. The chapter ends with a summary of con w elusions. 10.2 Competition and price policy Introduction The manufacturing sector exhibits a high degree of firm concentration. Market structures tend to be either monopolistic or oligopolistic. Half of the manufacturing sector's 7000 products are produced under monopolistic conditions while 80 per cent of the remaining products (40 per cent of total) are produced under oligopolistic conditions (UNIDO (1985». A Tirolc, Jean, The Theory of Industrial Organization, MIT Press, 1989, page 113. 147 recent repon by the Zimbabwe Monopolies Commission (1992) revealed that 69% of Zimbabwean manufacturing originates in industries where the four largest enterprises account for at least 80% of gross output (calculated on the basis of eso data on the 4-digit level). With the introduction of ESAP the level of monopoly is expected to go down. To achieve this objective appropriate policies are required and finns have to believe that the government will stick to these policies. The effect of such policies can only be estimated after assessing the existing market structure in detail by seeking answers to questions such as: What are the attributes of a company which uses a specific method of price setting? What is the probability that it will diversify into the production of new products, or innovate its products. What is the probability that it will sell directly to end users in the domestic market. By answering the questions it becomes possible to ascenain the existing level of competition and analyze the measures which firms have taken to reinforce the positive and cushion the negative effects of ESAP. Examples of such measures are reduction of costs through forward or backward integration. Measures adopted are expected to differ with finns because competition is expected to have different effects depending on sector, location and finn size. Thus we look at the source of competition and relate the price setting mechanism to the channel of output, levels of integration, sector, finn size and diversifica- tion of the product range. Source of competition When the relationship between sector and main source of competition (Table 10.1) is considered, we had to reject the hypothesis that there is no association between sector and main source of competition. It is therefore imponant to differentiate analysis by sector because different sectors will have different sources of competition. The textiles sector faces more competition in expon markets. This is largely due to the fact that the textiles sector is the one which exports the highest average proportion of output. The metal sector also has the highest likelihood of facing competition from imports. Table'O.1: Main source of competition by sector food wood textile metal total none 4 2 4 2 12 domestic firms 45 22 66 28 161 foreign competitors in export markets 0 1 11 0 12 imports 0 1 6 7 14 total 49 26 87 37 199 Analysis of variance of the proportion of exports per sector has been be used to verify these results. On the basis of the F-statistic we reject the null hypothesis that sectors export on average the same percentage of output. The mean percentages of output which are exponed are: · 4.9 for the food sector · 10.1 for the wood sector · 16.6 for the textiles sector and · 7.5 for the metal sector. The results indicate that the textiles sector export the highest proportion of output. The smallest proportion is found in the food sector. This is to be expected because of the nature 148 of the products. We used blow-up factors to correct for the sampling method and make sure that each firm had the same probability of being in the sample. Then we looked again at main source of competition by sector. For reasons of comparison we downscaled the total number of firms to 199 again and rounded the figures. The weighted figures, which should tell us about the population, can be compared with the unweighted figures, which only reflect the sample. The results are shown in Table 10.2. Table 10.2: Main source of competition by sector (weighted) food wood textile metal total none 0 3 5 0 B domestic firms foreign competitors in export markets 58 0 18 0 85 , 23 0 184 1 imports 0 0 5 1 6 total 58 21 96 24 199 Comparison makes clear that our sample is a bit biased. For the whole population of Zimbabwean manufacturing firms in the four sectors we distinguished the main source of competition comes from domestic tirms. Only 4% of the firms indicate that they do not face competition in the markets where they sell. Import are only important to the garments and textile sector2. The question about the main source of competition gives an idea of where the competition comes from. Now we shall try to look at the intensity of competition Measurement and importance of competition Some of the variables which have been used such as the measure of the intensity of competi- tion, backward and forward integration are difficult to measure. Below we specify how they have been measured and their importance. Intensity of competition Economic theory suggests that the level of competition is revealed by the price setting mechanism. The extent to which firms are free to charge their own prices reflects the intensity of competition. The methods of price setting adopted by firms has been used to determine the intensity of competition facing firms. The firms were asked to mention the two most important factors that determine how they set their prices. About two-third answered that they set their price as a markup over costs. The enumerators were urged to ask about constraining factors if that answer was given. So in more than 80 cases a second answer was given or constraints were indicated. Based on the two factors mentioned by the firms we constructed a measure of competition. This was done in two steps. First the two answers were combined in the most dominant and informative one. The results are presented in Table 10.3. :2 In the period that the interviews were held. the import of second hand clothing Willi allowed. Many gannent manu(acturens complained about that and claimed that this Willi a kind of dumping by western economics. 149 About 29% of firms set their price as a markup over costs. The same percentage take the market price as given. As many as 37 firms (20%) follow the lead of one or two competi- tors. Only 12 % indicate that they negotiate, which is surprisingly low. Especially in a situation where inflation and exchange rates fluctuate heavily and the outcome of the policies under ESAP are insecure, one would expect more flexible pricing. Only 2 % of firms have their prices set by the government. In these cases the effects of competition cannot be analyses fully. These firms may charge prices which do not cover costs and be subsidized, or they may have a protected monopoly. In both cases the relation between price setting and competition is disturbed. We grouped these answers into three categories. Markup without constraint was grouped together with keeping the price high to signal quality, following the lead of one or two competitors and an association price. The idea is that in those cases there is monopolistic or oligopolistic price setting. Firms that negotiate or charge different prices for the same product in different markets (onJy 2) were also put into a separate category. And the last category consists of firms that take the market price as given or adjust to import prices. Firms for which the government sets the price were left Ollt3. Table lOA shows the categories by sector. Firms facing little competition are expected to be in the oligopolistic group. Firms that face a lot of competition are expected to take the market price as given. And the remaining firms, the ones that negotiate, are expected to face an intermediate level of competition. Table 10.4: Grouped price setting by sector food wood textile metal total oligopolistic 32 10 39 27 108 negotiation 0 3 20 3 26 market price 13 13 30 7 63 total 45 26 89 37 , 97 As the table shows the wood and textile sector are the most competitive sectors. That is if our sample provides us a correct picture of the economy. Again we used blow-up factors to correct for the sampling method and make sure that each firm had the same probability of being in the sample. Then price setting and sector were related as shown table to.5. For reasons of comparison we again downscaled the total number of firms (to 197) and rounded the figures. 3 These were three butcheries and a milling company. The government had a monopoly on slaughtering and the price of maize WlIll controlled. 150 Table 10.5: Weighted grouped price setting by sector food wood textile metal total oligopolistic 12 9 35 15 71 negotiation 0 3 51 1 55 market price 41 9 13 8 61 total 53 21 99 24 197 Again the weighted figures, which tell us about the population of firms, can be compared with the unweighted figures, which reflect the sample. So we can say that for the whole of Zimbabwe the metal sector is the most oligopoJ istic sector, which is also indicated by the unweighted figures that only apply to the sample. In the food and textile sector the weighted picture is different. The food sector is probably more competitive than reflected by our sample and in the textile sector negotiation is more important and given market prices are less prevalent. Looking at the column totals we see that there are little shifts, indicating that weighing does not alter the relative importance of the sectors. The metal sector was given a bit less weight and the food sector a bit more. Row totals show that on the basis of our sample alone, we would have overestimated the instance of oligopolistic competition and underestimated negotiation. These differences are important and it is important to keep the in mind while reading the rest of this chapter. It would not make sense to do a similar kind of weighting for every detailed piece of information. Neither is it necessary to weight if we compare two variables. Forward integration Forward integration is expected to be measured by the extent to which firms have adopted subsequent stages in production. Examples are a food factory producing flour. It could either sell the flour to factories which producing confectionery or the firm could integrate and produce confectionery. In the textile industry a firm may do only spinning or integrate garment making activities. Such qualitative information is not explicit in the questionnaire. What we do know is whether firms bypass the wholesaling and retailing stages. This is the measure which has been used for analysis4 . It turns out that 95 large and very large firms in the sample sell on average 19.2% of their domestic sales directly to end users. Looking at use of marketing channels by sector, metal firms sell on average almost 60 % to end users. Food, wood and textile firms sell respectively 27, 42 and 37 percent. So the least competi- tive sector is the most forwardly integrated one. Table 10.6 shows how the firms in the sample set their prices in relation to marketing channels. There is no significant relation between price setting and marketing channel. So competition does not vary with the channel. Competition is about the same at the two levels of the marketing chain we distinguished. This means that e.g. oligopolistic firms do not sell more to retailers and wholesalers. The fact that metal firms sell more to end users than wood processors is not because they are more oligopolistic, but for other reasons. The question than is, why they bother to set up a marketing system to sell to end users. A possible answer is that retailers and wholesalers are oligopsonistic, whereas end users are not. That could be a sound reason for forward integration. What we register is only the result. We do not know what caused this market structure. " Note that a manufacturer owning a retail ehain will report that he sells to retailers even though the chain is his. So the estimate of forward integration is biucd downwardly. 151 Table 10.6: Average use of marketing channels by price setting group oligopolistic negotiate market price firms (%1 1%1 1%1 Ino., end user, private 34.46 38.71 31.52 58 end user, public 5.12 10.77 1.40 58 retailer or wholesaler, private 53.17 42.00 65.72 5 retailer or wholesaler, public 1.47 4.87 0.41 2 other 5.78 3.65 0.95 total 100.00 100.00 100.00 201 Backward integration As with forward integration backward integration has not been adequately captured in our questionnaire. For analysis it has been measured by the extent to which firms source raw materials from firms which they own. This is expected to reduce the probability of failure to meet delivery deadlines which could results in loss of clients. In the section on conflict resolution we shall see whether backward integrated firms complain less about late delivery and bad quality. Only 18 firms reported that they purchased raw materials from their own businesses. Eight took the price as given and 10 were oligopolistic. Half of the 18 were food processors. They all owned farms or wells (in the case of beverages). Six textile firms and three wood working firms made up for the rest. Diversification and innovation Oligopolistic firms are likely to suffer from X-efficiency and are less likely to innovate or diversify as incentives from competition are lacking. We looked at diversification and innovation in the last three years. Firms reported whether they used investment to improve the production process, add to capacity, introduce new products or a different variety of a product. It turned out that about 60% of the firms had done this and that there were no differences between the prise setting categories, which we took as indicators of competition. So we can not conclude that firms that face more competition are more likely to diversify or innovate5 . 10.3 Transactions In this section we shall look at the transactions of firms with their primary suppliers and with their clients-. Questions on relationships with suppliers and clients took about one fifth of the questionnaire and were pretty detailed. The major categorization of transactions was done on the basis of the form of payment. In this section we shall also concentrate on this. We will focus on the question why trade credit is given. The economic literature gives several explanation for interlinked product and.. credit markets. First, interlinkage may reflect imperfections in financial markets (Schwarz (1974), Emery (1984). Some clients cannot get bank credit because of lack of collateral. For the bank the cost of information and/or enforcement of a credit contract may be too high. The supplier may have an advantage: he may be better informed about the financial position of his client as he is in a better position to judge his performance as an entrepreneur. He is also in a better position to enforce S In fact this conclusion is a bit shaky. because we assume that the intensity of competition has not changed over the last three years and especially has not changed as a result of the diversification or innovation. 152 payment: he can always repurchase (part of) the goods at moderate costs. If in this case credit is a purely commercial activity one would expect this to be a portfolio decision of the supplier. Implicit interest rates have to be equal to the return on alternative means of investment then. We shall refer to this explanation as the financial motive. Secondly, interlinkage may reflect sales promotion. There are two elements in sales promotion, a liquidity motive and a promotion motive. Clients may have to purchase on credit simply because they are unable to pay cash. To sell to these clients a supplier has to sellon credit (Johnson and Kallberg (1986). The lenght of the production cycle is expected to determine the period that credit is needed. The second element is that giving credit adds to the ties between supplier and client. The assumption is that a client with an outstanding position is more likely to buy from the same supplier again. This applies to oligopolistic markets, where firms compete to get a share of the market. This binding element may be purely psychological. Nevertheless both supplier and client are likely to benefit from a well established relationship, because it reduces risk and transaction costs (looking for offers, looking for customers, stock keeping) for both. Trade credit reduces transaction costs if the date of delivery is unknown. It brings us to the next explanation. Thirdly, interJinkage may be related to risk. Favourable credit terms may be part of a complex and ongoing relationship between supplier and client which reduces risks for both. The client wants his favourable terms, ordering with an outstanding balance, good quality, timely delivery and often specific products. The supplier wants a share in the market, continuity of sales and (in time) payment. Finally, there are explanations in terms of transaction costs. We mentioned already the cost of information and enforcement with respect to the purely financial motive. Suppliers can stop deliveries or take their product back, which is easier than enforcing repayment. We may add the cost of idle resources. As has been pointed out by Ferris (1981), this may play a role if the exact day of transfer of the goods is unknown. This may happen e.g. because transport companies are unreliable. Both sides would than be forced to hold resources idle. The supplier would have to keep the goods in stock longer than he would like and the purchaser would have to hold money ready without knowing when it is needed. Cuevas et al. (1993) argue that this would lead to short credit terms and that overdrafts would allow payments at any time without requiring the firm to hold resources idle. There is a stronger point to be made if we combine transaction costs and the liquidity motive. Economies of scale in transport induce bulk sales. The. supplier can give an incentive to the buyer to purchase large quantities by giving a quantity discount, but large quantities could result in cash flow problems on the side of the buyer. Credit can solve this problem. So, quantity discounts and credit are both meant to give an incentive to buy large quantities and economize on transport and administrative costs. Transactions with suppliers Relationships between the firms and their' suppliers of raw materials, intermediates and other material inputs are very important in Zimbabwe. One of the entrepreneurs told us that in his opinion there is a big difference between manufacturing in the first and the third world. He said that in the first world the main problem seems to be how to sell, whereas in the third world the problem is how to buy. The idea is that western enterprises are more concerned with marketing and selling at competitive prices in competitive markets, whereas, at least for many years in Zimbabwe, firms had hardly any problems selling their products. In Zim- babwe getting material inputs seems (or seemed) to be part of the core of the business. Many entrepreneurs complained about supplies, although the situation seems to improve 153 under ESAP. To illustrate this we shall give a few examples. Sugar is an important input for the food industry. All sugar comes from the Zimbabwe Sugar Refineries (ZSR), a parastatal. Supply however is sometimes inadequate and long periods of rationing have occurred. The monopoly position of ZSR was guaranteed and protected by government policy. The main supplier of timber is the Forestry Commission. Large industries now complain that the Forestry Commission rather exports than sells on the local market. The export incentive of 8 % given by the government in addition to the premium on ERS makes it more attractive for the Forestry Commission to export than to sell on the local market. The industry claims that they have to pay more than world market prices, as they also have to compensate for the export incentive and the foreign exchange scarcity premium. Complaints in the metal sector are comparable with the ones in the food industry. ZISCO, the Zimbabwe Steel Company still has to a large extent a monopoly on the supply of steeL Various types of steel are on the list of items that cannot be imported. Only in the textile and garments sector is the situation slightly different. There are several large spinners and weavers, and although they probably have price arrangements, there is ample supply. This is also due to the fact that cotton is available within the country. Spinners and weavers do complain about the Cotton Marketing Board, who have a monopoly on cotton supply. The leather industry has tried to integrate backwards as far as possible. Shoe manufacturers have started their own tanneries and are now also buying hides form other sources than the Cold Storage Commission (the parastatal that used to have a monopoly on hides, as they had a monopoly on slaughtering). In general, especially before ESAP, firms were highly dependent on their suppliers. The data reflect this. The average duration of the relationships with primary suppliers was 14.8 years, when firms were asked about relationships with their three main primary suppliers. The meaning of this figure becomes even more significant when we consider that the average age of the firms in the sample was 23.4 years. The dependence is also reflected in the fact that although there are many complaints about late deliveries and bad quality of inputs, there are hardly any real disputes. In only 3 out of 120 cases a lawyer was hired. These three were all cases of deficient quality. Although only 48% of the disputes were settled, the outcome was marked sarisfacrory in 60% of the cases and in more than 90% of the cases the relationship was not broken off. In a situation where suppliers have such a strong position, it is less risky for them to give credit to their clients. Clients have a strong incentive to pay within the agreed term. The question however remains why they should give any credit at all. What is their motive? We shall analyze whether we can explain the means of payment to primary suppliers. The following table presen~ an overview of the form of payment required by primary suppliers. Multiple answers were allowed. The 201 firms in the sample reported about 554 suppliers. It is important to note that these are not necessarily different suppliers. The same producer of cloth may be mentioned by many garment manufacturers as a primary supplier, bUl, obviously, the form of payment can differ. In ten cases a second form of payment was indicated and only once a third one was mentioned. Details are given on the number of times a cettain kind of payment was mentioned per sector (565 in total). Table 10.7: Form of payment by sector sector credit or running account cash edvance payment total food 92 45 1 138 wood 54 21 2 77 textile 159 81 8 248 metal 73 28 1 102 total 378 175 12 565 154 The question we are interested in is credit or no credit. And as the table shows no credit means in 94% of the cases cash. To assess the importance of the financial motive and the liquidity motive, we look at access to bank loans and overdraft facilities. As Table 10.8 shows, there are no differences between finns that receive trade credit and others. Neither is there much difference between the groups that paid zero and positive (implicit) rates 6 . In fact finns getting trade credit had better access to the financial sector7 . Table 10.8: Trade credit and formal credit overdrafts loans yes no yes no no credit firms 30 39 29 41 credit at 0 % firms 65 17 37 44 credit at > 0% firms 41 5 28 18 total 136 61 94 103 But having access to the formal financial sector does not imply that one is not credit con- strained. Some of the firms that have overdraft facilities may be at the limit of this facility and therefore need credit. On the other hand, some of the firms that do not have overdrafts may in fact be so rich that they do not need one and therefore also do not need credit. They may negotiate better prices instead. Overdraft facilities are very important in Zimbabwe. About two third of all firms has overdraft facilities. Many finns do not report loans, but have a very large current balance on overdraft facilities. The average balance of those finns that have overdrafts and wished to reveal the balance was Z$ 2.6 million (132 obs.) Unfortunately we do not know whether they used their facilities up to the limit. If they do not and receive credit the liquidity motive does not apply. As is clear from Table 10.9 and Table 10.10, older and larger firms received more overdraft facilities than small finns or firms that have started recent Iy. Table 10.9: Overdraft facilities by firm size firm size by number of employees employment overdraft < =10 11-100 101-250 > =251 total yes 9 40 43 44 136 no 31 26 4 4 65 total 40 66 47 48 201 Table 10.10: Overdraft facilities by start-up year overdraft <'65 '65·'79 'SO·'89 >'89 total yes 63 44 20 9 136 no 8 12 29 16 65 total 71 56' 49 25 201 In contrast with the 68% of firms that have overdraft facilities, only 48% of the firms has ever received a loan from a formal institution. As could be expected a higher percentage of 6 Implicit I'tltes wen:: calculated using foregone cash discounts as implicit interest payments. 7 Note that there lire lome missing data. Information on both overdrafts and bank loans was collected for 197 firms. But the 197 reporting on overdrafts are not the same lIS the 197 reporting on bank loans. 155 large and old firms have received such loans. The difference between regions is however curious. There seems to be no a priori reason why firms in Harare are more likely to apply for or get a loan than firms in BUlawayo. Yet this is the case as Table 10.11 shows. Table 10.": Use of Formal and Semi-formal Institutions Harare Bulawayo Other total ever received a loan 61 21 12 94 never received a loan 46 34 23 103 total 107 55 35 197 The proportion of a certain raw material supplied by one supplier tells us something about the importance of that supplier. If risk is important, a high proportion may indicate dependence of the buyer and reduce the risk of default. On the other hand dependence may also reduce the necessity of credit from a promotion point of view. Table 10.12 shows that in 57% of the relationships described with primary suppliers the supplier provided over 85% of the raw material or input. In 45% of the cases the primary supplier even supplied 100% of the raw material, intermediate or other inputs mentioned. Table 10.12: Proportion supplied by the primary supplier 5-20% 25-40% 45-60% 65-80% 85-1 00% total count 22 47 75 91 308 543 percent 4.1 8.6 13.8 16.8 56.7 100.0 Of the 308 suppliers that supplied 85% or more, 244 actually provided 100%. The question whether this supplier was also the only supplier available was answered confirmative in 107 cases. That means that in 20% of all cases the primary supplier provided 100% for the simple reason that there was no alternative supplier available. The frequency of deliveries may be an important variable, if transaction cost motives are an important factor in explaining trade credit. The question What is the frequency of these purchases? was not very clear however. Respondents did not know whether orders, deliveries or payments were meant by purchases. In many cases payment was on a monthly basis while deliveries were made whenever raw materials and transport were available. The frequency of orders varies even more_ A bakery explained about the rationing of sugar by the Zimbabwe Sugar Refineries. Ordering was of little use in that period since the bakery simply wanted as much as possible. Table 10_13 presents the frequency of the answers given. Table 10.'3: Frequency of purchases daily weekly fort- monthly quarterly half- yearly occasional· depending total nightly yearly Iy on orders count 67 114 35 179 63 20 15 35 28 556 percent 12.1 20.5 6.3 32.2 11.3 3.6 2.7 6.3 5.0 100.0 We already mentioned that the transaction cost motive may induce trade discounts. The interviewers tried to distinguish trade discounts and cash discounts. Cash discounts are often given if the payment is made within 30 days from statement. This is not the cash discount as defined in the questionnaire. But, as it is merely an incentive to pay in time, it is also not a trade discount. Trade discounts were defined as quantity discounts (price discounts because 156 large quantities were ordered) and discounts given to specific clients apart from standard terms of payment. Table 10.14 shows that a trade discount was reported 116 times. In 80% of the cases the discount was 10% or less. Table 10.14: Trade discount given by the supplier 0% ,- 10% , 2.5-20% 22.5-30% lotal count 425 93 19 4 541 percent 78.6 17.2 3.5 0.7 100.0 To analyze the form of payment we composed a data basis where the unit of analysis was a transaction rather than a firm. Most of the firms gave information on the primary supplier of their three main inputs. The explanations for credit already mentioned were all taken into account. Rather than choosing one explanation for trade credit, we assume that there is some validity in all of them. We used the following variables: dependent variable: supplier credit equal to I if the form of payment is credit or running account independent variables: expected sign supplier share percentage of input supplied by primary supplier relationship length of relationship + bulk deliveries dummy for regular deliveries less than once a week + monopolist supplier dummy for monopolist suppl iar +7 trade discount percentage trade discount 7 overdraft dummy for having an overdraft facility ~7 profitability profits (as reported) divided by sales + stocks end of year stock of raw materials divided by sales + employment firm size + race dummy for black ownership age age of firm + foreign supplier dummy for foreign suppliers foreign dummy for foreign owned firms exporting dummy for exporting firms + An exporting firm has the option to import its raw materials, so the promotion motive seems more rel~vant for its suppliers, especially because imports usually have to be paid cash (Letter of Credit). Also, an exporting firm is likely to be more solvent (less risk). Foreign owned firms may be more solvent, but their reputation may be worse. They may not be part of a Zimbabwean network. Foreign suppliers often require a letter of credit, which is basically a kind of cash payment. So we expect a negative sign. The size of the firm is expected to have a positive effect if any, because of reputation. Race, firm age and the length of the relationship are also related to reputation and the advantages of mutual trust. Signs are expected to be negative for black entrepreneurs and positive for firm age and the length of the relationship. Infrequent supplies are expected to generate a positive effect, because of an economlzmg effect on transaction costs. Suppliers like to supply large quantities and are willing to supply on credit to compensate. For the same reason we expect a positive sign for stocks of raw materials and intermediate inputs. After taking all these factors into account, the percentage that the primary suppliers supplies is expected to have a negative effect. There is no need to 157 give credit if your client is not likely to switch. So the higher the percentage the lesser the promotion argument applies and the more important price competition will be. Monopolist suppliers will care linle about enforcement and are not concerned about price competition. But they do worry about entry. Binding may serve as a barrier to entry. The cost of credit can be retained by setting the price a bit higher. So if giving credit is a bener barrier to entry than keeping the price a bit lower we expect a positive sign if a supplier is the only supplier available. By contrast this does not apply for monopolies regulated by law, because entry is not possible. For them only the financial and liquidity motive are left. A supplier may ask cash payment from a firm with financial problems (screening to reduce risk). Profitability and access to financial markets (overdrafts) indicate whether a firm is solvent or not. So positive signs are then expected. Profitability does not tell much about cash flow however. Having an overdraft could tell us something, but only if the firm is not at its maximum. We assume that most firms are not at their maximum, especially because interest rates on overdrafts were very high at the time of the interviews. As already mentioned, the liquidity motive requires that firms do not have or can not use an overdraft. If the liquidity motive prevails we shall therefore have to find a negative effect. Sector dummies have been incorporated to see whether the length of the production process is important. In the food sector the processing is expected to be very short, leading to more cash payments. In wood and metal there are large state monopoly suppliers (regulated by law). A dummy for the food sector was only signiticant when we left out the variable indicating the percentage of the primary supplier. This indicates that concentration is more important than the duration of the production cycle. In other words, the negative sign of a food sector dummy reflects that primary suppliers have more market power in this sector rather than the effect of a short cycle and proof for the liquidity motive. None of the sector dummies turned out to be significant. Access to the financial sector may be limited outside the cities. On the other hand, contract enforcement may also be difficult. Dummies for location were tried but had no effect. We also tried other indicators of personal relations and reputation (e.g. the relation of the respondent with the person in the supplying firm and whether he belongs to the same ethnic group) and type of supplier dummies (public or private firms). None of them turned out to be significant. Table 1 O. , 5: logit estimation dependent variable: supplier credit 444 observations (305 Yes, 139 No) independent variable estimated coefficient standard error t-statistic constant 0.51646 0.60960 0.84172 supplier share -0.01262 0.00592 -2.13079 relationship 0.01024 0.01306 0.78377 bulk deliveries 0.82344 0.28414 2.89803 monopolist supplier -0.11657 0.34342 -0.33394 trade discount 0.09342 0.03657 2.55444 employment -0.00014 0.00019 -0.71674 firmage -0.00022 0.01030 -0.02159 stocks 2.41097 1.11026 2.17154 profitability 3.48572 0.88561 3.93595 race -2.30690 0.37566 -6.14097 foreign -0.58137 0.38695 -1.50245 exporting 0.25168 0.32556 0.77307 foreign supplier -0.69480 0.37241 -1.86069 overdraft 1.38207 0.31867 4.33696 percent correctly predicted: 82.0 158 The estimation results are quite interesting. Black owned firms are definitely less likely to get credit from their suppliers. This is important since we control for all kinds of other reasons why black owned firms should be constrained. So, it is not because black owned firms are also smaller, younger, bank credit constrained or more producing for the domestic market. This seems to be pure discrimination. We should add immediately that this does not reveal who is discriminating. Rasmussen (1993) in a study on micro enterprises notes that there is widespread mistrust among blacks and that contract enforcement is difficult. A second conclusion is that suppliers apparently are well informed about the financial position of their clients. The fact that profits of the client are significant indicates that suppliers screen and are risk averse. As already mentioned, the most interesting explanatory variable is access to overdraft facilities. The liquidity motive would imply that firms without overdraft facilities are more likely to get trade credit. This does not turn out to be the case. Having an overdraft facility makes it more likely to get trade credit. Again this might be the result of risk averse supplier, who do not have to worry about selling, but are concerned about getting their money. Personal relations and networks are not as important as we expected. If there is such a thing as an old boys network support is probably given up to the point where is starts to cost money. Neither the length of the relationship nor being part of a group of companies was important. Transaction cost motives do get some support. Both the relative size of the stock of raw materials and the dummy for regular but not too frequent purchases are significant. Finally, the size of a firm turns out to be totally insignificant. 8 So the sheer size of a firm, which could have a reputational effect does not matter. It is because of the other factors that large enterprises are less likely to pay cash. The data set provides detailed information on 169 purchases on credit or running account. This detailed information consists of the product purchased, the value, outstanding balance, agreed and actual term of repayment, interest payments, optional cash discount and enforce- ment. Of the 169 cases, only once a payment was made on order, three times there was a payment on delivery. In six cases there was no specific term to repay. In fourteen cases payment was made in instalments. Interest was paid in only ten cases. The standard in Zimbabwe is that suppliers start charging interest after 60 or 90 days from statement. A discount is normally given if the client pays within 30 days from statement. The question- naire asks about time elapsed from delivery, not from statement. In many cases however the delivery date was only important in the sense that it determined on which monthly statement the purchase was put. We tried to ask specifically about time elapsed from delivery. In those cases where supplies were delivered continuously in combination with payment within 30 days from statement we took the average to be 45 days. In fact we assumed that deliveries were made regularly during the month and that payment was made as late as possible. This assumption was checked in a few cases and respondents generally agreed. The number of 45 days also turned out to be the average payment period for the 169 cases. The average payment period if no interest was charged was found to be 39 days (again from delivery). For clients who paid interest the average payment period was as high as 159 days, which illustrates that explicit interest is only charged-over longer periods, and often as a penalty. I addition to the credit transactions a total of 115 cash purchases and (only) eight cases of advance payment were described. In 20 of the 115 cases of cash payment the client was aware of a discount. The discounts reponed were on average 7%, ranging from 1% to 20%. Because of instalment payments and instalment deliveries and missing data implicit interest rates could only be calculated in 141 of the 169 individual trade credit transactions. In only six of these cases interest was charged explicitly. But in 53 cases a cash discount 8 Recall, however. that micro enterprises (less than 5 employees) were not in the I&ITIple. 159 was available. This allowed us to calculate the implicit interest paid. which was added to the explicit interest paid. We looked at the resulting interest rates, to see whether giving credit is a commercial activity. It turned out that interest rates on a yearly basis exceeded 15% in 49 cases out of the 141. But this does not tell us much, because in 88 cases the rate was zero. We cannot conclude that the firms that did not report cash discounts are indifferent and not aware of the time value of money. To illustrate that, 92 % of them had agreed upon a fixed term to repay. It is therefore more likely that the cost of credit was included in the price. The 46 firms (53 credit transactions) that were charged a positive rate paid on average 4.9%, with a minimum of 1% and a maximum of 17.4%. However, on a yearly basis the average was 104% with a minimum of 3.2% and maximum of 1350%. The maximum was because a firm paid within a week while the agreed term was a month (the second highest rate was 607%). In fact these rates are not very high. The highest rate found in Ghana was 14299%, with a minimum of27%. The following table shows average rates (explicit and implicit added) on a yearly basis by sector. We also present the average size of the firms in terms of employment, the average term in days and the number of firms with an overdraft facility. This is done for all credit transactions first and then for those with a positive implicit rate. Table 10.16: Overview of credit transactions all credit transactions no. cases av. size (empll av. amount (Z$) av. term (days) overdrafts av. interest (%) food 38 457 570,920 40 31 19.0 wood 22 139 85,068 43 18 35.2 textile 60 496 300,503 37 53 65.4 metal 21 188 355,680 43 16 5.9 total 141 384 347,985 40 118 39.3 credit transactions with positive implicit interest rates no. cases av. size (empl) av. amount (2$) av. term (days) overdrafts 8V. interest (%) food 4 827 371,125 41 3 180.5 wood 14 144 93,257 48 13 55.3 textile 32 461 240,281 43 29 122.6 metal 3 444 1,671,500 106 3 41.2 total 53 404 292,332 48 48 104.6 Although the zero rates have to be interpreted with caution, it does not mean that our analysis has to stop. After all, firms accepting the credit terms with a positive implicit rate could have opted for the cash discount. So we are entitled to ask what kind of firms did not opt for advantageous cash discounts. We made separate subsets of trade credit transactions with a rate above and below 15%, because this was the prevailing interest rate on savings accounts9 . Then we tried to explain whether a firm had to pay a commercial rate (above 15%) or a concessional rate (below 15%) by it's characteristics. Neither size, nor. firm age, nor type of supplier, nor stocks of raw materials nor frequency of purchases was of any significance. Moreover, there were not more complaints about suppliers who charge a positive implicit rate with respect to late or non-delivery and these suppliers were not expected to use more or other means of contract enforcement. 9 This rate went up to 23$ during our stay. The IFS statistics give an increase of the average consumer price index between the first and the second quarter of 1993 (the time most of the reported transactions took place) of 16$. 160 So, on the basis of the table above we looked at sector and location dummies. We also tried variables that were transaction specific, such as the value of the sale and the term of credit in days, but they were not significant. Leaving out most of the insignificant explana- tory variables the results are presented in the next table. Table 10.17: Logit estimation dependent variable: commercial credit 121 observations (43 yes, 78 nol independent variable estimated coeHicient standard error t-statistic constant -3.56 0.89 -4.02 trade discount 0.10 0.05 2.02 profitability 3.14 1.40 2.24 race -0.51 0.98 -0.52 overdraft 1.22 0.73 1.67 Bulawayo -1.58 0.53 -3.01 wood 3.03 0.79 3.85 textile 2.66 0.62 4.27 percent correctly predicted: 76.9 The use of profitability as an explanatory variable reduced the number of observations from 141 to 121. Leaving out profitability and using 141 observations had a positive impact on the significance of all variables except race, which became even more insignificant. The coefficients did not change by more than 10%, except again for race, which became -.20. So we decided to take the reduction in number of observations for granted. The next step was to try a tobit estimation of the implicit interest rates, with the same explanatory variables and the same observations. Table 10.'8: Tobit estimation dependent variable: implicit yearly rate 121 observations (75 zero) independent variable estimated coeHicient standard error t-statistic constant -3.81 1.04 ·3.66 trade discount 0.14 0.06 2.41 profitability 2.79 1.66 1.68 race -1.36 1.22 -1. 11 overdraft 0.98 0.86 1.15 Bulawayo -1.81 0.66 -2.77 wood 2.27 0.91 2.50 textile 2.77 0.72 3.84 sigma::! 5.93 1.29 4.59 The results suggest that in sectors where competition is severe, such as the wood working and garments sector, rates tend to be higher. Why rates are lower in Bulawayo is not clear. Remember that firms in Bulawayo also made less use of formal loans. A possible explana- tion is the effect of the drought, which has affected the South of the country more than the rest. As a result firms in Bulawayo may have stretched the term more than other firms and may have been allowed to do that, leading to lower implicit rates. Perhaps more likely is that quite a lot of raw materials have to come from Harare. They are transported by rail or by road transport. Delays up to several weeks are not unusual then. The difference between cash on delivery and credit from the point of view of the suppl ier then becomes smaller. 161 That makes it more likely for the respondents to answer that they would have had to pay the same amounts if they had purchased on a cash basis. The financial position of the firm is again important. Firms that are doing well in terms of profitability are charged a higher rate and also firms that have overdrafts are more likely to accept unfavourable credit rates. Again this could mean that firms screen effectively, but it does not explain why the unfavourable terms are accepted 10. A negative sign for black owned firms indicates that these firms are less likely to pay high interest rates. This is not obvious and difficult to explain. It is possible however, that because there is discrimination, black owned firms that do get credit have to perform better than the average. That would mean that the liquidity motive does not apply even if they do not have an overdraft facility (discrimination by banks). Note however that race is not significant. The fact that the trade discount is significant is revealing that there is no such thing as a free lunch. Firms either get a trade discount or a cash discount. What can be concluded with respect to the motives for giving credit? The fact that profits have a significant positive effect on getting credit indicates that firms that sell on credit screen in general. We also saw that profits even have a positive effect on the level of the rate. This could reflect that credit constrained firms that make enough profit have their suppliers as a source of finance of last resort II (the financial motive). For the liquidity motive implicit rates do not have to be at a commercial level. But a positive effect of overdrafts is difficult to reconcile with this motive. One would also have expected a negative sign for the food sector, because of the short production cycle. So, we can conclude that the liquidity motive is not a very likely explanation. Pure sales promotion to get or maintain a share in the market could not be verified with the data we have. Most of the results indicate that suppliers are willing to give credit if there is little risk attached. Relationships between firms and their suppliers are long lasting. Probably both parties benefit from this in terms of risk reduction, but this is hard to measure. In the section on conflict resolution we shall see whether this has any effect on late delivery and bad quality of inputs. The fact that the size of stocks and the frequency of purchases was significant indicates that the transaction cost motive may be important. However, these variables should then appear in the tobit estimation with a negative sign, which in fact they do, but not significant! y. Transactions with clients In the previous section we looked at transactions with suppliers. We have tried to explain whether trade credit was given or not, mainly on the basis of the characteristics of the recipient and information on relations. Now we consider transactions with clients and investigate whether we can explain trade credit by the characteristics of the supplier of credit in combination with information on the relation between the buyer and the seller. Ideally one would like to look at characteristics of suppliers and recipients of credit simultaneously, i.e. with respect to the same transaction. Unfortunately data does not aHow this to a larger extent than done in the previous section. We can only look at different transactions. An advantage is that the data do allow us to assess whether credit is passed on, as is suggested by Weston 10 Possible reasons could be that these firms are at the maximum of their overdraft facility or that complaints are taken more seriously in case of an outstanding balance, for which they arc willing to pay. 11 Thill would mean that firms with an overdraft that pay rates above the overdraft rate (26) are all at the maximum of their overdraft facility. 162 and Brigham (1981). The transactions we reviewed in the previous section were between manufacturing firms and their suppliers. Typically these were supplies of raw materials and intermediate inputs. The transactions we analyze in this section are between manufacturing firms and their clients. Goods traded between them are typically final products and intermediate inputs. This may be quite an important difference. In general final products will be a lot more specific than raw materials. So, repurchase of goods in case of non-payment may be less attractive. Especially when clients have asked for specific designs, sizes or quality, the producer may end up with goods that can not be sold without considerable loss. We would expect therefore that firm supplying final goods will be more cautious with respect to giving trade credit than firms supplying raw materials. On the other hand, there may be more competition , so the promotion and the liquidity motive may be stronger in the case of final products. Clients may also have less access to financial markets, e.g. because they are smaller. Table 10.19 shows the primary form of sales by type of client and whether written agree- ments were made. Multiple answers were allowed and therefore, the number of firms is also reponed. For all types of clients, except private end users, credit is the primary form of sales. Private end users pay mostly in cash. Consignment and advance payment were rarely mentioned. Table 10.19: Primary form of sales by type of client type of client no. of firms credit consignment cash advance payment end user, private 122 68 1 82 15 end user, public 41 31 2 8 3 retailer or wholesaler, private 143 119 3 21 1 retailer or wholesaler, public 12 10 0 2 0 foreign 66 56 0 11 6 other 8 7 0 0 We defined a firm as a creditor if it indicated that the primary form of sales to one of its clients was credit or consignment. Of our sample of 201 firms, 170 are creditors. In contrast with the previous section we can not use data -of transactions to explain whether a firm is a creditor or not: we have to use the firm as the unit of analysis. The following table shows that 92 % of the firms that receive credit from at least one of their primary suppliers, also sell on credit to at least one type of client. Only 58% of the firms that did not receive credit gave credit to their clients. So we can conclude that credit is passed on. Table 10.20: PaSSing on of trade credit credit given credit received yes no total yes 144 12 158 no 28 19 45 total 170 3T 201 To explain whether a firm .gives credit or not we used the following variables, which were panly based on the previous section. dependent variable: client credit equal to one if firm sells on credit or consignment 163 independent variables: expected sign ') monopolist dummy for monopolist supplier negotiator dummy for negotiating the price with clients + race dummy for black owned firms overdraft dummy for having an overdraft facil ity + employment firm size + advenisement cost of promotion and advertisement divided by sales trade debtor dummy for receiving credit from suppliers + food dummy for food sector wholesalelretail client dummy for selling to private wholesalers/retailers + cooperative/external owner dummy for cooperatives and firms with individual external owners Clearly we do not know anything about the financial position of the recipients. But we do know whether the supplier has access to formal sources of finance, i.e. has an overdraft. We have to be cautious however, because if a firm is credit constrained and at the ceiling of its overdraft facility this may have a negative effect on giving credit. We expect that large firms are more likely to be creditors, although there is no sound economic reason for this. Monopolists do not have to give credit to promote sales, unless their customers are credit constrained (the liquidity theory). On the other hand monopolists run less risk and may want to prevent entry. Their customers can hardly breach the contract as they depend on the monopolist. Firms that indicate that they negotiate the price of the product with their clients rather than use a preset or otherwise given price are more likely to give credit, because they can raise the price if they expect late payment or non payment, to compensate for the risk. Also personal relations with clients may be at stake here. Negotiations lead to a more personal involvement, which reduces the risk of default and thus makes the granting of credit more likely. We assume that adverse selection is not a serious problem. We expect less credit in the food sector because of short turnover periods. There are two reasons to include this variable even though it was not significant in the previous section. First, it then was insignificant if we introduced the variable indicating the relative importance of the supplier. Here we do know the relative importance. Secondly, the liquidity motive is more likely to apply to clients of manufacturing firms than with manufacturing firms themselves. A negative sign for race is expected if there is such a thing as a white financialllegal system and access to formal finance and enforcement depends on race. Note that now race refers to the firm that gives credit. We introduced a dummy (trade debtor), which is equal to 1 if the finn gets credit itself. Advertisement is the amount spent on advertising divided by turnover. The idea here is that firms which invest in strong relationships with clients do not have to spend mu'~ on advertising. A positive signs is expected for being a trade debtor and a negative ... ne for advertisement. Finally, we incorporated a dummy for clients who are private retailers or wholesalers, which we expect to be positive because of the transaction cost motive, and a dummy for cooperatives and firms with external owners, which is expected to be negative, because these firms are not so much part of a network. We tried a number of other variables. Regional dummies had no effect. Neither had firm age as (albeit weak) proxy for the length of the relationship. We also found domestic competition to be insignificant. Competition was expected to increase the probability of default, because it is easier for a client to go to a competitor. Besides, competition reduces the possibilities of asking a higher price to recover the cost of giving credit. But competition may also enlarge the need for sales promotion and binding. These two effects seem to neutralize each other. Dummies for the other types of clients, a dummy for foreign ownership (more solvent) and 164 even capacity utilization (idle capacity being a reason for sales promotion) were also tried in vain. And finally a dummy for exporting firms, tried because they do not only depend on the local market, was tried and found to be not significant. The results of the logit estima- tion were as follows: Table 10.21: logit estimation dependent variable: client credit 200 observations (169 yes. 31 no) independent variable estimated coefficient standard error t -statistic constant 0.05 0.77 0.06 monopolist 0.15 1.17 0.13 negotiator 1.97 0.91 2.16 race -0.72 0.72 -1.01 overdraft 0.24 0.61 0.39 employment 0.01 0.01 1.90 advertisement -8.44 4.57 -1.85 trade debtor 0.92 0.64 1.44 food -, .62 0.63 -2.59 wholesale/retail client 1.12 0.58 1.93 cooperative/external owner -1.37 0.80 -, .72 percent correctly predicted: 89 When we look at these results we can conclude that all coefficients have the expected sign. Firms that advertise could be firms that sell search goods, products of which the quality can be ascertained before purchase (Nelson (1970) and Tirole (1989». They try to maintain their share in the market not by binding but by advertising. If the goods are 'experience goods'. goods of which the quality is learned after some time the clients may value a close relation- ship as well (if in turn their clients complain, they want to able to go back to their supplier). This is more likely to apply to wholesalers and retailers than to end users, which is reflected by the variable wholesale/retail clients. Being part of a network means that one has better access to information on credit worthiness of clients e_g .. Cooperatives and firms with external owners are less likely to be part of such a network. By contrast firms that set the price by negotiation probably do have good connections and access to information. Further- more, negotiation allows for tailor-made terms of credit. It seems that the size of a firm in terms of employment does influence the terms of sale. Trade debtors are more likely to give credit. This indicates that a firm's liquidity position is important. Access to banking may indeed not be a good indicator for liquidity. It is not even remotely significant. In the food sector we expected less credit because of a shon turnover period (liquidity motive) and again because most of the goods traded in this sector will be search goods (clients do not benefit enough from long relationship. so they do not push for credit). The racial factor became totally insignificant when we introduced the trade debtor variable. That is interesting. It means that black owned enterprises give less credit to their clients largely because they are discriminated by their suppliers. A modest conclusion could be that giving credit is part of a binding strategy, chosen by firms who have a comparative advantage in maintaining good relationships with clients and who are in a financial and competitive position to do so, e.g. because they also buy on credit. Whether the firms screen and whether transaction costs playa role, such as was the case in the previous section on transactions with suppliers, could not be checked. We now turn to specific sales transactions on credit, by the firms that are creditors. Just as in the case of transactions with suppliers, we calculated implicit interest rates on the basis of information on individual transactions. The data set provides information on 177 sales on credit or running account. Some firms reported more than one sale. Again we summarize the results. The 177 cases can be subdivided by type of client as shown in Table 10.22. As was 165 already clear from Table 10.19, the majority of the clients are private and domestic. Table 10.22: Completed transactions by form of sales and type of client type of client credit consignment cash advance payment end user, private 46 1 62 19 end user, public 12 0 2 1 retailer or wholesaler, private 85 4 37 0 retailer or wholesaler, public 6 0 3 0 foreign 24 0 6 4 other 4 1 0 total 177 6 111 24 In only six of these 177 cases there was a payment on order. Five of them were made to medium sized firms, the other one to a small firm. Seven times there was a payment on delivery, 5 of which to a small firms. In 95% of the 177 cases there was a specific term to pay. In 15% of the cases payment in instalments was accepted. Interest was paid in only 3 cases. The average time elapsed between delivery and full payment was 50 days. The remarks made with respect to purchases on credit about the terms from the date of statement also apply to sales on credit. In 64% of the cases there was no guarantee to pay and in 20% of the cases there was a signed invoice. Details of a completed sale on consignment were given in only 6 cases. A total of III completed cash sales were described. Table 10.22 lists them by type of client. Again private domestic clients are most important, in line with Table 10.19. In 35 of these III cases a cash discount was given. On average the discount was 6%, ranging from 1% to 40%. 24 completed sales with advance payment were specified in detail. The number of firms that gave credit to their clients was 170 (31 did not). Because we did not ask the time between prepayment and delivery or the amounts of and periods between instalments, we could not calculate all implicit interest rates. If the firm had received a prepayment it was impossible (8 cases). In the case of instalments other than on the day of delivery, we could only certify whether the implicit rate was zero or positive, by looking at the cash discount. The exact implicit rate could not be calculated in these cases, because the number of days between instalments and the amount of the instalments was not asked. Fortunately all but one did not give an optional cash discount, so the implicit rates were zero. In total, there were 80 cases of zero interest rates and 81 positive implicit interest rates that could be calculated. Only two times interest payments were actually received. The following table presents details by sector. Table 10.23: Overview of credit transactions aU credit transactions no. of firms av. size (emp!) av. amount 12$1 av. term (days) overdrafts av. interest % food 34 403 344,918 46 27 21.3 wood 19 139 82,414 39 13 54.0 textile 81 482 404,604 42 64 121.3 metal 27 258 399:545 67 21 38.7 total 161 375 353,129 47 125 78.4 credit transactions with positive implicit interest fetas no. of firms ey. size (empl) ay. amount (2$) av. term (days) overdrafts ev. interest % food 10 446 94,843 60 8 72.4 wood 10 110 61,622 43 8 102.6 textile 49 513 185,839 41 45 200.5 metal 12 444 636,405 43 10 87.3 total 81 417 226,020 44 71 155.8 166 All rates were above 8% on an annual basis and only four were below 15%. We get the same picture as before in the case of transactions with suppliers, either the rate is zero, or above the rate of return on bank savings. There are however also differences. Firms charge on average higher implicit interest rates to their clients than they pay themselves to their suppliers. We analyzed which firms ask zero and which positive implicit rates. Of course we started again with the explanatory variables suggested by the questionnaire, such as the question whether the client was a friend, relative or business relation, whether the person dealt with was from the same ethnic group and the length of the relationship. None of these turned out to have any effect. Also the value of the sale and the term of payment were insignificant. Therefore we tried the same variables as used for the transactions with suppliers, in combination with wholesale/retail client and negotiator. The size of the firm turned out to be totally insignificant. We left it out and then used a probit estimation rather than a logit, as there was no reason any more to assume a logistic relationship. Table 10.24: Probit estimation dependent variable: commercial client credit 161 observations (77 yes, 84 no) independent variable estimated coefficient standard error t-statistic constant -1.14 0.31 -3.66 monopolist -0.35 0.48 -0.73 negotiator -0.01 0.31 -0.04 race -0.17 0.32 -0.54 overdraft 0.55 0.28 1.96 wood 0.50 0.35 1.42 textile 0.61 0.23 2.65 wholesale/retail client 0.53 0.21 2.50 Bulawayo 0.20 0.24 0.83 percent correctly predicted: ~~.4~ Table 10.25: Tobit estimation dependent variable: client credit rate 161 observations (80 zero) independent variable estimated coefficient standard error t-statistic constant -5.26 1.63 -3.22 monopolist -2.22 2.47 -0.90 negotiator -0.14 1.50 -0.01 race -0.68 1.57 -0.53 overdraft 2.32 1.45 1.61 wood 1.57 1.73 0.91 textile 2.81 , .16 2.43 wholesale/retail client -0.01 1.05 -0.01 Bulawayo 0.45 1.13 0.40 sigma 2 30.73 4.99 6.16 The estimation results are not impressive. As before. the more competitive sectors charge higher rates and also having an overdraft has a significant positive effect on the probability of charging a positive implicit rate. We would have expected that firms that negotiate the price are more likely to apply a positive implicit rate. After all we asked If you had made the same sale on a cash basis, how much would you have charged? From negotiating firms we would expect that they charge a lower price then. However, this is not the case. Being a 167 negotiating firm turns out to be insignificant (the sign i"s even negative). Firms in Bulawayo were more likely to get credit from their suppliers at a zero rate. but that does not mean that treat their clients in the same way. That supports the assumption we made that the distance between creditor and debtor is important. Private wholesalers and retailers are more likely to pay a commercial rate, but th.is does not effect the level of the rate in a significant way. Overdrafts do effect the level of the rate, indicating that firms paying overdrafts charge higher implicit rates to stimulate their clients to pay cash. The probit and tobit analysis does not alter the conclusion that the financial position of the creditor is important and that clients use their market power to demand credit to reduce the risk of bad qUality. We shall concentrate on this last item in the section on conflict resolution. When we finally compare the analysis of getting credit from suppliers and giving credit to clients, there are some remarkable differences. The effect of access to formal credit on getting trade credit turned out to differ significantly from zero. A striking difference is that black entrepreneurs receive less trade credit, whereas race does not influence giving credie A fourth point is that being part of a network does not significantly affect receiving credit but does affect giving credit. This is a bit awkward. Ont! would expect that networks matter in both cases. The assumption is that personal and institutional relationships reduce risk. We do not have specific information on the relationship between creditor and recipient except in the case of the primary suppliers, where the length of the relationship was not significant at a 90% level. In the case of transactions with clients, the length of the relationship did not influence the implicit rate. We looked at the net position of debtors and creditors, i.e. (accounts receivable - accounts payable) + (prepayments made - prepayments received). As firm size, age or location were not rea II y important factors, we present a Table by sector. Most firms (72 %) have a positive net trade credit position. Again an indication that credit is passed on. The difference between the food sector and the other sectors is remarkable, because we found that there were no significant differences between sectors when form of sales was looked at. A possible explanation could be that manufacturers of food items (bakeries, butcheries, etc.) sell more to end users in cash. Apparently the use of forms of sales does not differ much, but the quantities per form of sales do. - Table 10.26: Net trade cradit positions by sector food wood textile metal total net recipient 18 4 12 2 36 zero position 3 2 9 2 16 net grantor 25 16 62 32 135 total 46 22 83 36 187 A topic for further research would be to look at product specificity. We already mentioned the theory developed by Williamson a.o., that the specificy of the product influences the type of contract and the possibilities of contract enforcement. We also mentioned the difference between search goods and experience goods. For most of the transactions described in this section we also asked which product was purchased or bought. It should therefore be possible to classify these product and verify whether specificy or the difference between search and experience goods is important. 168 10.4 In-kind lending and borrowing The RPEO survey revealed that in Zimbabwe in-kind lending and borrowing are important. One of detailed questions on specific informal loans was: What was the amount lent in-kind? This question was originally meant as a check to see whether there was anything lent in addition to cash loans. However, it turned out that a number of firms lent out raw materials and equipment on a regular basis. This happened among competitors as well as between subsidiaries. Having noticed this phenomenon, the enumerators were urged to ask specifi- cally about pure in-kind lending and borrowing. This probably was not done systematically since there was no separate item in the questionnaire (enumerators may have opted to report on a cash loan rather than an in-kind loan). Nevertheless, 45 firms indicated that they had lent or borrowed material of substantial value to or from other enterprises (we left out 5 firms that reported lending or borrowing material worth less than Z$ 100 (US$ 16». This helping each other may have started or may have been stimulated during the UOI period, when there was a feeling among Zimbabwean entrepreneurs that they had to collude against the rest of the world. In those days temporary shortages of raw materials and spare parts of equipment were, of course, quite common. But our data suggest that in-kind loan transac- tions are still quite common. Even though we asked for completed transactions, 65% of the reported transactions had taken place within the last half year. Another indication of commonness is that the borrowing firms had on average tried .84 other source. The following table presents an overview of in kind lending and in kind borrowing by type of firm. The typology is as follows: · cooperative: firm owned by workers · working owner firm: firm with internal working owner(s) ('entrepreneurial firm') · external owner firm: firm with external owner(s) · part of group firm: firm with internal working owner(s), but part of a (family-structured) group of firms, not being a stock exchange fund or holding · subsidiary: subsidiary of a Zimbabwean stock exchange fund · foreign subsidiary: subsidiary of a foreign firm · corporation: stock exchange firm parastatal: government owned firm This classification differs from the legal status and ownership structure variables in the questionnaire. It was constructed on the basis of these questions and the firm history. Table 10.27: In kind lending by type of firm yes % no % row total % cooperative 0 0 14 100.0 14 7.0 working owner 17 15.9 90 84.1 107 53.2 external owner 0 0 5 100.0 5 2.5 part of group 9 31.0 20 69.0 29 14.4 subsidiary of fund 7 30.4 16 69.6 23 11.4 foreign subsidiary 5 27.8 13 72.2 18 9.0 corporation parastatal 1 0 25.0 0 3 1 75.0 100.0 , 4 2.0 0.5 column total 39 19.4 162 80.6 201 100.0 169 Table 10.28: In kind borrowing by type of firm yes % no % row total % cooperative 0 0 14 100 14 7 working owner 8 7.5 99 92.5 107 53.2 external owner 20 4 80 5 2.5 part of group 7 24.1 22 75.9 29 14.4 subsidiary of fund 6 26.1 17 73.9 23 11.4 foreign owned 4 22.2 14 77.8 18 9 corporation 0 0 4 100 4 2 parastatal 0 0 1 100 1 0.5 column total 26 12.9 175 87.1 201 100 The rough picture one gets when looking at the percentages is that networks are important. Firms that are part of a group, or a subsidiary are more engaged in this kind of transactions. We do not know wether the transactions were really within the group (or with the parent company or other subsidiaries). It could be that these firms are in general better connected. Between the four sectors differences are small. In the wood and textile sector lending occurs relatively more. This may reflect the influence of the trade associations, i.e. the fact that these sectors are better organized. Of the 26 firms that are borrowing in kind, 20 also lend in kind. We do not know whether lending and borrowing was between the same firms. So we can only speculate about mutuality. There are more firms that report lending than firms reporting that they borrowed. There are several possible explanations. First borrowing may have been underreported. Second, there may be a flow to very small enterprises. In this way we would miss the firms with less than 5 employees who only borrow, but do not lend out. Third, borrowing may in fact be more concentrated that lending. That very small enterprises borrow a lot seems unlikely if we look at the size of the firms that are involved in borrowing. Table 10.29 shows that firms with less than 10 workers do not borrow at all and hardly lend in kind. This is in line with the conclusion of Mead and Kunjeku (1992) that Zimbabwe has a dualistic economy with weak linkages between small and large firms. Table 10.29: In-kind lending and borrowing by size of employment employment in-kind lending in-kind borrowing yes % no % yes % no % total % < =10 1 2.5 39 97.5 0 0 40 100.0 40 , 9.9 11 -.100 7 10.6 59 89.4 7 10.6 59 89.4 66 32.8 101-250 16 34.0 31 66.0 10 21.3 37 78.7 47 23.4 > = 251 15 31.3 33 68.8 9 18.8 39 81.3 48 23.9 total 39 '9.4 162 80.6 26 12.9 175 87.1 201 100.0 To assess the importance of relationships going back to the UOI period we looked at the age of the lending and borrowing firms and the length of the relationship between the two parties. A problem however with the question: How long have you known the /enderlrecipient? is that it either tells you something about a personal relationship or a relationship between two companies 1:2. This explains why the relationship sometimes exceeds firm age. In most cases the question was probably assumed to refer to the personal relationship. The following table shows lending and borrowing by firms that were estab- lished during or before UOI and firms that were established after independence. It also presents details on whether the relationship existed already during UDI. 12 ICC the foot note on the next page . 170 Table 10.30: In-kind lending and borrowing relationships age of firm and length of relationship in-kind lending in-kind borrowing pre-independence firm 29 21 after independence firm 10 5 pre-independence relation 13 12 after independence relation 22 11 As the table shows not all firms answered the question with respect to the length of the relationship. In the sample there were 74 firms that were established after independence and 124 firms that were established before independence. In-kind lending and borrowing is definitely more common among pre-independence firms (two-third of all cases), but this may simply reflect that they have better networks because they are older and does not have to do anything with the VOl period. Only 45% of the relationships go back to the VOl period, but if these are personal relationships instead of relationships between enterprises that does not tell us much. Only in the case of owners personal relationships will be at least as long as business relationships. We looked closer at firms with working owners, because their race could be identified. The following table presents lending and borrowing by those firms. Table 10.31: In-kind lending and borrowing by race of owners race in '"land lending 'nOOkind borrowing yes % no % yes % no % total % African 3 7.3 38 92.7 1 2.4 40 97.6 41 38.3 Asian 3 16.7 15 83.3 2 ".1 16 88.9 18 16.8 European 12 26.7 33 73.3 5 1 1. 1 40 88.9 45 42.1 Other 0 0.0 3 100.0 a 0.0 3 100.0 3 2.8 total 18 16.8 89 83.2 8 7.5 99 92.5 107 100.0 Looking at all firms of which we know the race of the owners (working within the firm or not) the tables look essentially the same. Table 10.32: In-kind lending and borrowing by race of owners of working owner firms race in-kind lending in-kind borrowing yes % no % yes % no % total % African 3 5.4 53 94.6 2 3.6 54 96.4 56 32.4 Asian 5 21.7 1S 78.3 2 8.7 21 91.3 23 13.3 European 18 22.2 63 77.S 12 14.8 69 85.2 81 46.8 Other 1 7.7 12 92.3 1 7.7 12 92.3 13 7.5 total 27 15.6 146 84.4 17 9.8 156 90.2 173 100.0 Clearly white-owned and asian firms are overpresent. This is not very surprising though. The average duration of the relationship between lending and borrowing panies turns out to be slightly more than 14 years. The average black-owned firm exists about 10 years in contrast with white-owned 27 years and asian-owned firms 25 years. So, if the length of the relationship is imponant or existence during UOI, then one would expect fewer black owned firms. We checked whether firms lending and borrowing were older than 14 years, i.e. existed during VOl and whether these were black-owned. Only three black-owned firms were involved in lending (one older than 14 years). The youngest asian firm lending was 9 years old; the other 4 were over 20 years old. Recipients, however, were not known for such long periods. Only one recipient was already known to the lender during VOL There were 13 white-owned lending firms older than 14 years and 5 were younger. Relationships with 171 recipients went back as long as to UOI in only 9 cases, one of which was among the 5 younger firms. In the cases of borrowing by firms with owners of whom we know the race, the two cases of black-owned firms were also lending. So mutuality may be important. One was 15 years old the other part of a group of companies. Two asian firms borrowed. One post UOI firm from a lender he already knew during UOI and one UOI firm who borrowed from a lender known only 5 years. Only 2 out of 12 white-owned borrowing firms were post-UOI firms. But relationships with the lender exceeded 14 years in only 5 cases. So among firms with identifiable owners the picture is exactly the same. Two-third of the firms existed already during UOI, but only 45% of relationships went back to that period. Another question asked was whether the recipient was from the same ethnic group.13 The question was: Was the recipient a member oj the same ethnic group or tribe?14 The answers were as follows: two of the three black firms were lending to non-black firms, with a missing data for the third one which is part of a group. One of the five Asian firms lent to an Asian firm, three to non-Asian firms. Finally, the white-owned firms reported 4 times the same ethnic group, 4 times another ethnic group and 11 Don't Knows or Not Applica- bles. Firms that are part of a group a subsidiary are more likely to lend or borrow because they are affiliated. But again relatively more Asian and white-owned firms do this. Table 10.33 shows type and race of owners for lending and borrowing firms. Obviously, we do not know the race of the owners of parastatals and firms that are listed on the stock exchange. We also do not know the race of the owners of most of the subsidiaries, although one can expect that most of them are white owned and, what is more important, managed by Europeans. What the table shows in addition to table 10.32 is therefore that also lending and borrowing firms that are part of a group are predominantly white and Asian owned. Table '0.33: In-kind lending and borrowing by race of all owners 'n""ind lending in""ind borrowing type of firm African Asian Euro- Other Total African Asian Euro- Other Total pean pean working owner 2 3 12 0 17 1 2 5 0 8 part of group 1 2 4 0 7 1 0 4 1 6 subsidiary of fund foreign subsidiary 0 0 0 0 1 ,0 1 2 1 0 0 0 0 0 0 1 total 3 5 18 27 2 2 11 16 It is hard to draw conclusions on the basis of this information. White and Asian owned firms lend and borrow more, but the limited information on the race of lenders and recipients suggests that they do not discriminate. That is of cause only half of what we want to know. The fact that black owned firms that managed to become part of a network are not discrimi- nated is not very surprising. The question remains whether it is more difficult for black owned firms to become part of a network than it would be for a white or Asian owned firm of the same size and age. Data does not allow us to answer that question. 13 Neither the enumerators nor the interviewees liked this question. 14 Often, the respondent (or this part of the questionnai~ was the financial manager. Now he may happen to be an Indian, worlc.ing for a white-owncd firm. approving a request by eo black production manager working for another white-owned firm. The fact that they do this on behalf of their employers. who play golf together will not be revealed. 172 The last table presents an overview of details with respect to in-kind lending and borrowing. Table 10.34: Details on in-kind lending and borrowing in-kind lending in-kind borrowing unit # obs. unit # obs. average amount (Z$) 15303.4 39 76361.5 26 average maturity (days) 41 34 34 22 average amount repaid (Z$) 14916.6 38 90695.9 15 25 collateral required (yes) 1 39 2 26 first loan transaction with this firm? 8 38 average duration of relationship (years) 13 3Ei 16 23 average no. of other sources tried .84 25 Altogether it is difficult to draw conclusions about the influence of race or the UDI period. Because white-owned and Asian firms tend to be larger and older, they are expected to have larger and better established networks. A fact is however that they are part of these networks and black owned firms generally not or not yet. This gives these firms an advantage and provides a good example of colonial hysteresis. 10.5 Conflict resolution In this section we shall look at late payment and non-payment by the clients of the firms we visited and at late/non delivery and deficient quality of inputs and services delivered by their suppliers. We shall relate this to the items that we already discussed in the previous sections. So we shall assess the importance of the extent of competition, the type of client, the length of the business relationship and L1e effect of written agreements. Late payment and non- payment shall only be analyzed for the 170 firms that indicated to sell on credit to their clients. Late and non-payment We know that 170 firms of our sample of 201 sell on credit to at least one type of client. Of these 170 firms 59% make written agreements with at least one type of client. Table 10.35 shows details. Table 10.35: Primary form of sales and written agreements by type of client type of client no. of firms credit written agreements end user, private 122 68 46 end user, public 41 31 24 retailer or wholesaler, private 143 119 60 retailer or wholesaler, public 12 10 3 foreign 66 56 42 other 8 7 2 Written agreements with foreign clients are of course not very surprising. It also was to be expected that, when we compare dealing with end users and dealing with retailers and wholesalers, relatively more firms make written agreements with end users. Now we would IS Lent 11lW materials were purehased and rctumc:d after the price had gone up substantially. That explains the difference between amount lent and amount repaid. 173 like to know whether these agreements sort any effect. So we would like to know whether firms that make written agreements have Jess problems with late and non-payment. We know whether firms had such problems during the last year and how often. It is however difficult to assess the number of cases of late and non-payment. One would like to know the total number of credit transactions in the last year as well, or rather even the total number of clients that bought on credit. We considered alternatives. From a theoretical point of view relating the number of late or non-payment cases to sales or size does not make much sense. To a factory building aircrafts, one non-paying client can mean the end of business, although one divided by sales or size in terms of employment will be very small. But also in practise these alternatives did not work. Late and non-payment problems were not related to size either measured by sales or employment. So what was left was to Jook at the characteristics of the firms that reported that they had had problems and those who did not and see whether their ways of contract enforcement were different or whether their monopoly power was bigger. It turned out that 88% of the firms that sell on credit had problems with late payment in the last year. Of these 88% two-third made written agreements. This should be compared with the remaining 12 %. Only 38 % of them made written agreements. So apparent! y firms start making agreements if late payment increases. And obviously written agreements do not solve the problem. We continued to look at firms that sell on credit only. Firms that set their price by negotiation had slightly less problems with late payment. But overall the percentages were very high (76% for negotiators, 90 % for price takers and 88% for oJigopolists). Negoti- ators managed to settle their disputed a lot better than the other categories. All of the reported cases of late payment conflicts between firms that set prices by negotiation and their clients were settled and business between the two parties was not broken off. Oligopolist settle only half of their disputes and price takers 75 %. In most of the cases where the dispute is settled the parties continue to do business. If it was the first transaction with a client, the conflict is less likely to be settled. Simple logit analysis showed that otherwise the length of the relationship between the two parties is not relevant for the likelihood of the dispute to be settled. Whether the client is an individual or a firms is not important. Conflicts with the government or foreign trading parties are less likely to be solved. We can conclude that conflicts are related to market power. In buyers and sellers markets there are more conflicts and they are less likely to be solved than in markets where negotiation based on equal power takes place. For non-payment the picture is roughly the same. Again negotiating firms are less likely to complain about non-payment and more likely to settle the dispute and continue the business relationship. And again the length of the relationship, which was on average 7 years, does not matter, but whether it.was the first transaction does. We summa:--:ze the details on late and nonpayment disputes in the following table. Table 10.36: Details on late payment and non-payment disputes late payment cases non-payment cases % no of obs % no of obs direct bargaining 54 63 51 72 private arbitration o 63 6 70 threatened to go to police 2 63 9 70 lawyer hired 29 63 71 72 threatened to go to court 24 63 66 71 dispute settled 68 62 43 72 satisfied with outcome 78 50 37 63 still doing business 60 60 18 71 174 The most striking fact from Table 10.36 is that so many disputes are settled, that, in most of the late-payment cases, the respondents were satisfied with the outcome and are still doing business with the client. This may be the result of the size of the business community in Zimbabwe and the imponance of reputation. Late/non delivery and deficient quality of inputs/services Whether a firm had had problems with late/non delivery or deficient quality of inputs was not related to its size, location or sector. Firms outside the cities were less affected, but they are expected to collect a larger pan of their inputs themselves. As in the case of late and non-payment, the number of times a supplier let a firm down can not be analyzed properly. We looked at firms that buy on credit from at least one of their primary suppliers. Of the 138 firms that buy on credit 61 % had had at least one problem with late/non delivery during the last year and 63% had had at least one problem with deficient quality of inputs. Surprisingly these percentages were much lower among the firms that did not buy on credit (33% and 37% respectively). The reason may be that these firms also collect themselves and thus immediately solve or avoid these problems. As we already noticed in section 10.3, that is only possible with search goods. Detailed descriptions were given about 44 cases of late or non-delivery and 75 cases of deficient qUality. The average relationship between supplier and client was again very long (16 years), but not of significant influence on the likelihood of the dispute to be settled. Table 10.37: Details on latelnon delivery and deficient quality of inputs late/non delivery deficient quality % no of obs % no of obs direct bargaining 45 42 64 75 private arbitration o 42 3 74 threatened to go to police o 43 o 73 lawyer hired o 42 4 72. threatened to go to court o 42 6 72. dispute senled 63 38 39 72. satisfied with outcome 62 39 58 71 still doing business 90 41 90 72. Again it is striking that most of the disputes were solved and that business between the parties concerned continued. The fact that business continued in 90% of the disputes and even if the client was not satisfied with the outcome, reflects the market power of the suppliers. Quite a number of times, when we asked about threatening and lawyers, the entrepreneur indicated that he was not in the position to have a real dispute as he depended too much on the supplier. This illustrates and underlines what was suggested already in section 10.3, that getting raw materials and other inputs is core business in Zimbabwe. It will be interesting to see whether and to what extent ESAP is going to change this. 10.6 Infrastructure The entrepreneurs were asked about obstacles in their business environment on a number of occasions during the interview. In all these instances they were asked to indicate on a one to 175 five scale the severity of problems presented to them; one indicating there was no problem, five that there was a serious problem. They were also asked to mention their biggest problem concerning the provision of public goods. It turned out that the provision of telecommunication, electricity and transportation for workers caused most of the hardship. Table 10.38: frequency of biggest problems mentioned no. of firms percentage electricity 29 15.34 water 12 6.3 freight transport 14 7.41 workers transport 27 14.29 roads 6 3.17 telephones 88 46.5 air/sea ports 2 1.06 waste disposal 1 0.53 security 9 4.76 other 1 0.53 When comparing the mean scores for the different categories the largest firms seemed to face the most serious obstacles in the provision of infrastructure. Between sectors the wood- working and metal firms complained the most. Telephones are by far the most serious problem. Irrespective of the size or the sector of the firms, the bad telecommunication system is identified as the most serious infrastructure problem. The large and the very large firms have more difficulty with this problem than the smaller firms, possibly because larger firms need more telephone lines. The telecommunication problems not only depend on the size of the firms, they also depend on their location. In Harare, the telephone problem is the most severe; the large and very large firms in Harare marked the telephone problem on average 4.6 (close to the maximum of 5), against 3.8 for firms in Bulawayo and 4.0 for the others. In Harare and Bulawayo, telephones are a problem because they do not work (i.e. lines are busy), in the other regions it is the unavailability of telephones lines, followed by the fact that they do not work. The second important infrastructure problem is electricity. The main reasons are the unstable power supply and the high prices. The unstable supply is partly related to the d- 'ght of 1992 (Zimbabwe's electricity is produced by hydro-electric plants) so that this mIght be a temporary phenomenon. Those complaining about the high prices referred either to the high surcharges that had to be paid when they used more than their electricity quota during the drought, or to the recent price increases. Very large firms complain the most about electricity. The third important problem is transport for workers. At the root of this problem is the monopoly on urban transportation of the ZANU/pF. Large firms consider this problem more important than the other firms and the problem is confined to greater Harare. With respect to this problem one has to take note of the fact that the transport problem is primarily a problem for the workers. The question was addressed to the manager rather than the workers, so the severity of the problem may be understated. Stories about workers that leave home between four and five in the morning to get to work at eight were often heard. The low frequency of the public buses, their unreliability or the fact that they are not available at all have been given as the main reasons for these problems. Table 10.39 shows the number of firms that have adapted by providing some infrastruc- ture themselves. 176 Table 10.39: Firms providing their own infrastructure no. of firms percentage generators 18 9.2 wells/boreholes 64 33.0 CB/walkie talkie/Radio 29 15.0 roads 8 4.4 workers transportation 40 20.5 waste disposal 76 39.0 labour at ports 11 6.3 freight transport 97 49.5 security 79 40.7 other 7 11.3 One might expect that those firms that have adapted themselves to the present state of the infrastructure, have fewer problems than those that rely completely on the public sector: firms that have serious problems with e.g. their water supply stan to provide their own, thus solving their water problem. However, this does not appear to be the case. For the three most serious problems (telephones, electricity, transponation for workers) plus the provision of water (included because 32% of all firms own a borehole), the firms that do not provide their own infrastructure rank the problem lower than those that do provide their own. Apparently, the entrepreneurs who provide their own infrastructure responded as if they were asked about the general provision instead of whether it was a severe problem to them. That may have caused them to ignore the fact that they had solved the problem. Alterna- tively, one might assume that own provision does not solve the problems adequately. Table 10.40 presents the distribution of firms with their own provision of infrastructure by geographical area. Table 10.40: Percentage of firms that provide their own infrastructure Harare Bulawayo Other telephones 17% 17% 7% electricity 8% 11% 10% workers Transport 24% 19% 16% water 21% 57% 32% waste disposal 47% 30% 26% security 51% 19% 45% When we consider the percentage of firms that adopted a measure for the problems that they encountered with the provision of infrastructure services, the Table shows that telephones, electricity and workers transportation are country wide provided for. Water is mainly a problem in Bulawayo. Problems with waste disposal are encountered allover the country, though especially in Harare. Security seems mainly a Harare a problem, though it can not be neglected in the country side. Compared to that Bulawayo is an oasis of safety. It is not easy to assess to what extent these problems affect the cost of doing business. We heard stories about managers who had had to drive into town to make telephone calls, for more than half a year because their phone was out of order. One can only speculate on the effect of that on a medium sized enterprise. Also the cost of workers coming late or being tired due to inadequate transport is hard to measure. The only thing that can be said is that to the extent that these externalities are caused by market imperfections such as 177 monopolies pricing and X-efficiency in the case of transport, electricity or telecommunica- tion, these costs can already be substantially reduced without large scale investments. 10.7 Business support Business support systems can be broadly placed in three categories: those provided by the government, by the private sector, and by donors or NGOS. Government activities include ZIMTRADE, the organisation for export promotion; the establishment of the Zimbabwe Development Bank (ZDS) and the Venture Capital Company of Zimbabwe (vccz); the recycling of blocked and surplus funds to these institutions; and the founding of the Credit Guarantee Company (cae). The functions of these institutions are not limited to the traditional provision of finance; in addition they provide extension and training. The government is in the process of establishing the Scientific and Industrial Research and Development Centre tSIRDC) under the auspices of the Research Council of Zimbabwe (RCZ). The main objective of SIRDC is to promote industrialisation of the country by developing indigenous industrial technologies. This will cover fields such as energy, microelectronics, mechanical engineering, building and construction and biotechnology. Through SIRDC the government promotes industrial research to harness and develop local technological capability and potential. Private sector institutions for business support include the Indigenous Business Develop- ment Centre (lBDC), the Confederation of Zimbabwean Industries (CZI), the Employers' Confederation of Zimbabwe (EMCOZ), the Zimbabwe Institute of Management and the Zimbabwe National Chamber of Commerce (ZNCC). Except for Zimbabwe Institute of Management, which provides courses, the main function of these institutions is to lobby the government for improvements in the business environment. The other category of business support systems are financial institutions funded by donor organizations such as the Africa Project Development Facility, the Africa Enterprise Fund and Commodity Import Programs. There is also a range of small private companies which assist budding entrepreneurs to prepare projects for funding and project approval by government authorities and donor organizations. The large banks also have their small business support programs, combining loans with courses in bookkeeping and management. Support to small-scale enterprises For this purpose the Small Enterprise Development Company (SEDCO) was set up to support small-scale enterprises by providing finance and extension services. SEDCO emphasises training since it believes that this is of crucial importance in complementing finance to determine the success of small enterprises. SEDCO claims that it advances short, medium and long term loans to small enterprises. Collateral is not a requirement and emphasis is on project viability. SEDCO'S activities are limited to the formal sector. Indeed, its attitude to the informal sector is quite hostile. Donor agencies and NGOS have support programmes particularly for small enterprises, e.g. CIDA (through SEDCO and cae), USAID (through CZI), ODA (through IBDc),the World Bank (through SED CO) and finally the ILO has its "Improve Your Business" program through EMCOZ. Most NGOS that deal with small scale enterprises offer a package of finance and training. As can be seen from the following table, the Zimbabwe Investment Centre, Foreign NGOS, the Confederation of Zimbabwe Industries, the National Chamber of Commerce, 178 Zimtrade, donor-financed Commodity Impon Programmes, and the Zimbabwe Institute of Management have been giving assistance to a substantial number of firms. Table 10.41: Use of business suppon services assistance no assistance never heard of Small Enterprise Development Company 11 174 16 Zimbabwe Investment Centre 41 138 22 Indigenous Business Development Centre 8 178 15 Foreign Non-Governmental Organizations 37 136 28 Confederation of Zimbabwe Industries 84 96 21 Zimbabwe National Chamber of Commerce 46 139 16 ZIMTRAOE 69 112 20 Employers' Confederation of Zimbabwe 26 131 44 Commodity Import Programs 55 87 59 Zimbabwe Institute of Management 35 143 23 Africa Project Development Facility 2 61 138 Africa Enterprise Fund 1 44 156 Other financial institutions 31 144 26 Larger firms and older firms make more use of these services. We found that even some large firms received assistance from sedco, the Small Enterprise Development Corporation. The question asked was whether they ever received assistance, so it could be that they have grown very fast, but we checked that and the result was negative. Only two of the smail firms (10 or fewer employees) reported NGO assistance from foreign NGOS. In defence of the NGOS we have to presume that quite a number of firm may know only the Zimbab- wean counter part. Table 10.42: Use of business suppon services by firm size small medium large very large total Small Enterprise Development Company 2 7 1 1 11 Zimbabwe Investment Centre 0 2 14 25 41 Indigenous Business Development Centre 3 3 1 1 8 Foreign Norr-Govt. Organizations 2 8 10 17 37 Confederation of Zimbabwe Industries 2 23 28 31 84 Zimbabwe National Chamber of Commerce :2 14 16 14 46 ZIMTRAOE a 19 23 27 69 Employers' Confederation of Zimbabwe 0 6 7 13 26 Commodity Import Programs ,1 16 27 55 Zimbabwe Institute of Management 1 7 11 16 35 Africa Project Development facility 1 a a 1 2 Africa Enterprise Fund 0 1 0 a Other financial institutions 3 6 7 15 31 There are no substantial differences in the use ·of these business suppon services by sector or by location. The various types of assistance that were given are specified in Table 10.43, for those institutions that were mentioned more than 30 times. 179 Table 10.43: Types of assistance Zimbabwe Investment Centre type of assistance no. of firms application for project approval 32 other assistance (or nor specified) 9 Foreign Non-Governmental Organizations type of assistance no. of firms management assistance 11 financial assistance 8 product promotion 3 training 4 other assistance lor nor specified) 11 Confederation of Zimbabwe Industries type of assistance no. of firms information 59 advice 9 assistance in getting import licenses 6 general representation 8 other assistance lor nor specified) 2 Zimbabwe National Chamber of Commerce type of assistance no. of firms information 34 assistance in getting import licenses 4 other assistance (or nor specified) 8 ZIMTRAOE type of assistance no. of firms export assistance 9 export promotion 31 information 24 other assistance 5 Commodity Import Programmes type of assistance no. of times mentioned Austrian aid 2 Canadian aid 16 Chinese aid 6 Dutch aid 2 German aid 7 Indian aid 2 Italian aid 3 Japanese aid 15 Norwegian aid 6 . Swedish aid 4 Swiss aid 5 UK aid 2 US aid 5 other (or not specified) 13 Zimbabwe Institute of Management type of assistance no. of firms courses 27 information 8 Again the effect of business support on the cost of doing business is hard to assess. The use of some of the suppo"rt is also the result of certain regulations. Commodity import programs were e.g. "ery popular because foreign exchange used to be very difficult to get. Foreign loans werl.. ;lOpular for the same reason. ZIMrRADE was mentioned several times as only a poor substitute for going abroad yourself, which was difficult because foreign exchange for business travel was rationed. So we expect that the use of a number of these institutions will 180 decrease with the progress of ESAP. 10.8 Foreign exchange regulations Until recently Zimbabwe's trade regime was highly regulated especially with respect to imports. The corner stone of these regulations was the administrative allocation of foreign exchange. A modified version of this system still existed at the time of the interviews despite the trade liberalisation program introduced in 1991. At six monthly intervals, the Ministry of Finance and the Reserve Bank used to forecast foreign exchange receipts and earnings. The amount available for imports was determined as a residual after deductions for known commitments such as debt servicing. The amount was then allocated through the Direct Local Market Allocation (DLMA) mechanism by the Ministry of Industry and Commerce to more than 30 competing item heads which included government ministries, parastatals, specific sectors and products, and various commercial and industrial import categories. Allocations to the private sector were determined as follows. An established applicant's allocation was a fixed share of the global amount, based on past use of foreign exchange. New applicants needed to satisfy certain criteria such as the ability to source imports more cheaply than established users or to produce for exports. The decision was usually made on an ad hoc basis. New entrants only became entitled to a regular foreign exchange allocation after several successive ad hoc allocations and satisfactory performance. In order to utilise the foreign exchange allocated, an import licence from the Ministry of Industry and Commerce was necessary. These import licences were issued almost automati- cally to holders of foreign currency allocations: their purpose was to monitor rather than allocate the use of foreign exchange. The trade regulations and exchange rate pol icy described above had several negative implications for the economy. The foreign exchange allocation system favoured established firms, thus creating considerable barriers to new entrants. Restrictions on imports of consumer goods protected domestic production. Private investment, both new and replace- ment, was seriously constrained by the rationing of foreign exchange. Foreign travel to identify new marke,ts or take knowledge of new technology was very strictly rationed. Partly as a result of this, capital goods were old and outdated, leading to uncompetitive production by international quality standards. The rationing of foreign exchange lead to rent seeking. The bureaucratic procedures for allocation of foreign exchange, import licensing and operating trade through authorised institutions were inefficient and time consuming. The Economic Structural Adjustment Programme includes: abolition of foreign exchange rationing, reforms of tariff policy, reform of exchange rate policy and improved export incentives. The administrative system of foreign exchange allocation has to a large extent been dismantled and replaced by a more market based system of allocation. An Open General Import License (OGIL) system has been put in place. Since the trade Iiberalisation program began, in August 1991, the Zimbabwean dollar has depreciated by over 100 per cent. It is the government's intention to have an exchange rate that is consistent with complete removal of the foreign exchange rationing mechanism by the end of 1995. Problems with foreign exchange Foreign exchange controls tend to become a more serious problem with the size of the size of the firm. The sector is not of any real influence, except for the wood work.ing sector 181 which is less affected by foreign exchange controls. The nature of the foreign exchange problem is not very well specified. All possible reasons (delays, availability of foreign exchange or the paperwork required), rank about equally high. It is interesting to note that with a mean score of about 2.5 foreign exchange regulations now present only a moderate problem, whereas in private sector studies during the eighties the difficulties of getting foreign exchange were consequently mentioned as the most burdensome problem to Zimbabwean entrepreneurs. The seriousness of the foreign exchange problems corresponds to the proportion of raw materials imported. This proportion is the smallest for the smal! firms (3 %) and the largest for the very big firms (28%). Also the low ranking of foreign exchange problems by the wood working firms is not surprising given their low import ratio. The higher ranking of foreign exchange problems in the food sector is more difficult to explain and probably due to difficulties with import of capital equipment. Due to the changes under the ESAP program it is nowadays possible to purchase foreign exchange in the form of Export Retention Schemes (ERS). With the depreciation of the Zimbabwe dollar the scarcity premium of ERS is going down. Part of the severity of the problems with forex may therefore already reflect the cost rather than the availability. This is not a trivial point and should get some attention. As one very bright manager pointed out, a large part of the industry runs on very old equipment, simply because there were no opportunities to replace it. ESAP is supposed to open the way to new investment in new machinery. But according to this spokesman the technology jump is too big for most of the existing firms. He therefore advocated to prolong the exchange rate policy, but in combina- tion with reestablishment of 100% deductibility of investment in one year, abolition of import tariffs on capital equipment and at least five year protection by means of high tariffs for finished goods. 10.9 Conclusions After all these separate sections on the business environment and the way business is done by the firms in the sample, it is time to summarize the main conclusions. We started with the source and intensity of competition. Only the textile sector was found to mention imports as a source of competition. The other sectors only mentioned domestic firms. So the effect of ESAP on competition from abroad is limited. Oligopolistic competition is prevalent among the firms in our sample in all sectors except the wood working sector. After correcting for sample bias, oligopolistic competition is only prevalent among firms in the metal sector. Setting a price by negotiation is a typical feature of the garment making and textile sector. This may reflect that in the sectors that are most affected by the ESAP policies competition increases. Firms are concerned about the relationships with their primary suppliers. These suppliers sell only on credit to solvent firms. Screening is done effectively and can be done, because in most cases the market is a sellers markel. The motive for selling on credit seems primarily to provide an incentive to buy in bulk and thus roouce transaction costs. Black ownoo firms are less likely to be allowed to buy on credit, which can only be explained as a form of discrimination. It is appropriate to talk about allowance of credit, because only in a third of the cases a positive implicit interest rate was charged. Again this was most dominant in the garment and textile, and the wood working sector. Firms accepting high implicit rates are likely to be profitable but credit constrained. The firms in our sample are found to sell on credit because their clients demand credit and firms do not want to loose them as customers. Clients probably demand credit to find out about the quality of the delivery before payment, in order to have a better bargaining position in case of deficient quality. 182 Less demand is given by firms that sell search goods and care less about binding clients. Also less credit is given by firms that do not have proper networks and therefore have less access to information. That makes it hard to screen properly and follow the correct binding strategy. Firms that can manipulate the price of their product in order to compensate the cost of credit are more likely to sell on credit. Typically this leads to different reactions depending on market power. Monopolists can raise the price but have less reason to do so. They do not have to sell on credit, because their clients are not in the position to demand credit. Price takers do not have the market power to refuse credit, but neither the do they have the option to compensate for the costs. Negotiating firms are most likely to sell on credit, because they can compensate for the costs, face an intermediate level of market power and are likely to have closer relationships to clients, which makes it easier to screen. In half of the cases there was a positive implicit interest rate. That is consistent when we compare it with the one third we found in the case of transactions with suppliers. Suppliers can more easily recover the cost of credit by charging a higher price. Again positive rates were found relatively more in the garment and textile and the wood working sectors. An important finding is that credit is passed on. Firms that receive credit are more likely to give credit to their clients. Only because of this and their own financial position do black owned firms give less credit to their clients. Not because of cultural reasons. Overall, the majority of firms in our sample are net grantors of credit. Almost a quarter of the firms are involved in in-kind lending or in-kind borrowing of raw materials and equipment of substantial value. Borrowing firms tend to lend also, the opposite is less salient. Borrowing and lending is mainly done by firms that are part of a network. Either because they are a subsidiary or because they have working owners with a long established network of contacts. Black owned firms are generally not (yet) part of these networks, which gives them a comparative disadvantage. Firms that set their prices by negotiation are not only more likely to sell on credit, but also manage to settle their disputes in a more satisfactory way than price takers and oligopolists. The fact that two third of the late payment disputes and half of the non-payment disputes with clients is settled, and that in almost all of these cases business between the two parties continues, illustrates that contract enforcement is not a very big problem and that relationships are tied. Conflicts with suppliers were also settled in two third of the cases of late and non-cielivery, but here business even continued in 90% of the cases. Telephones, electricity and transpon for workers cause a lot of hardship. Telephones because they do not operate and therefore hamper doing business. Electricity because of instability and high prices. And transpon for workers because long waiting and travel hours prevent them from getting enough sleep and cause them to be late every now and then. The costs connected with infrastructure problems are the highest in Harare, but infrastructure typical to the capital, such as being close to the government and the legislation probably compensates for that. Large firms tend to benefit more from business support services than small firms. Some of the reasons to make use of business support services will stop to exist due to the abolition of regulations under the ESAP policy reform. A good example are business support services that were used only to get access to foreign exchange. The extent to which foreign exchange regulations cause problems has decreased. It is much easier now to get foreign exchange. In fact there is a market for foreign exchange (ERS). There is a shift from problems with access to foreign exchange to problems with the cost of foreign exchange, or imports in general. Obviously firms that import raw materials complain more than firms that occasionally import equipment. The general conclusion that can be drawn on the basis of the findings in this chapter is 183 that in fact there are substantial differences between enterprises with respect to market power, contractual arrangements, contract enforcement and getting formal and informal support. These factors may be of importance for the probabil ity to survive. Following rounds of the survey have to be held to assess that. . 184 References Coase, R., 1937, The nature of the firm, Econometrica 4: 386-405 Emery, G.W., 1984, A pure fmancial explanation for trade credit, Journal of Financial and Quamatative Analysis 19 (3): 271-285 Ferris, S.J., 1981, A transaction theory of trade credit use, lournal of Political economy 96 (2): 243- 270 Government of Zimbabwe, 1992, Study of Monopolies and Competition Policy in Zimbabwe, Report submitted to the Inter-Ministerial Committee on the Monopolies Commission International Monetary Fund, 1994, InIernarional Financial Statistics, Washington DC Johnson, R. W, and 1.G. Kalberg, Management of accounts receivable and payable. In: E.!. Altman (ed), 1986, Handbook of Corporate Finance, Wiley & Sons, New York Mead, D.C. and P. Kunjeku, 1992, Business Linkages and Enterprise Development in Zimbabwe, unpublished paper, University of Zimbabwe, Harare Nelson, 1970, Information and Consumer Behaviour, Journal of Political Economy 78: 311-329 Rasmussen, J., 1992, The Small Enterprise Environment in Zimbabwe: Growing in the Shadow of Large Enterprises, IDS Bulletin 23 (3), IDS, University of Sussex, Brighton Schwarz, R.A., september 1974, An economic model of trade credit, loumal of Financial and QuanIitative AnaLysis 9 (3) Tirole, J., 1989. The Theory of Industrial Organization. Mrr Press UNIDO Regional and Country Study Branch, 1987 ·. Zimbabwe, Industrial Developmem Review Series, UNIDO, Vienna. Weston, J.F .· and E. F. Brigham, 1981, Maltagerial Finance. Hinsdale, fL, Dryden, 1981 Williamson, O.E., 1985 The Economic ltlStitutiollS of Capitalism. Firms Markets. Relatioltal Contract- ing, Collier Macmillan Publishers, London World Bank, 1993, Regioll.al Program on Enlerprise Development. Case Studies of Enterprise Finance in Ghana. prepared by Cuevas el a1. 185 · 11 Conclusion Who fonns firms in Zimbabwe? Typically entrepreneurs who are remarkably well educated (more than a quarter reached university), whose parents already had businesses, and who first acquired experience by working in other firms in the same sector. Few have been apprentices. Entrepreneurs acquire their skills typically on the job, working in other firms; this applies to two-thirds of the entrepreneurs. Even entrepreneurs in very small firms are well educated. They are diversified in the sense that close to half of the owners also own other firms. An important result is that the fate of the firm is not tied very closely to that of the owner: firms do survive their owners. Indeed, in about 40 per cent of the cases the firm was not founded by the present owner. It appears that the market for firms is well devel- oped: in almost a quarter of the cases the firms was bought by the present owner as a going concern. However, the tradability of capital equipment is quite limited in the sense that even new machines can be sold only at a substantial loss. The racial origin of entrepreneurs in Zimbabwe is changing rapidly. While only about 40 per cent of the entrepreneurs in the sample are African, two-thirds of the firms founded in the last few years are black-owned. The econometric analysis of chapter 3 showed that the fact that few blacks and few women are entrepreneurs is not simply due to, for example, differences in education or experience. Even if we control for such variables women and blacks are significantly less likely to become entrepreneurs. Education is important both in the decision to become an entrepreneur and in surviving the first few years. Survey evidence shows that the industrial sector in Zimbabwe is dynamic (unlike, for example, in Ghana): there has been substantial firm formation in the last few years, a period of structural adjustment (ESAP). That firm formation continued is remarkable, if only because in Africa adjustment programs tend to discourage investment in the short run through the policy uncertainty they generate. While the Zimbabwean government has been remarkably determined in its approach to adjustment policies. policy uncertainty is an issue. An important part of the adjustment effort, the trade liberalisation program, is not perceived as fully credible. Nevertheless, our results show new firms being formed. Growth in terms of employment has continued quite rapidly, except in the last few years, when it stagnated, probably mainly because of the drought. The analysis of firm dynamics of chapter 4 showed that while there are large differences between sectors in growth rates, with the textile and clothing industry growing far more rapidly than other sectors, these differences largely disappear once we control for firm characteristics such as age and size of the firm. In a sense this is disturbing: structural adjustment should be reflected in intersectoral differences in growth rates. For example, one would expect the textile and clothing sector (typically a tradable sector and heavily engaged in exporting) to grow faster than e.g. the food sector, a non-tradable. However, we did find evidence of exporting firms growing faster. We also found that race is irrelevant: black- owned firms are smaller than others but grow at the same rate. There is also evidence that membership of (formal or informal) conglomerates favours growth: stand-alone or entrepre- neurial firms grow more slowly. Employment in these firms has contracted sharply in the last few years (1991-93), as did employment in black-owned firms. The survey collected information on obstacfes to growth, as perceived by entrepreneurs. The responses revealed that regulations are now perceived as relatively unimportant, scoring far lower than market conditions and financial constraints. Chapter 5 compared indigenous (black-owned) with other firms. Black entrepreneurs are · not as well educated and have less experience. They also differ in the extent of their links 186 with other firms: black-owned firms are, in that sense, more isolated. How is investment financed? Recent empirical work, both for developed and for developing countries, stresses the importance of internal finance, company retained earnings in the case of existing firms and the entrepreneur's own savings in the case of start-up investment. Our results confirm onJy half of this story. External finance is indeed, and unsurprisingly, unimportant as a source for starting firms. However, there is substantial use of bank finance by expanding firms. Retained earnings financed less than a third of the value of investment in equipment, bank finance over a third. Much of this is in the form of bank loans. But a remarkable result is the heavy reliance of Zimbabwean firms on overdrafts for the financing of investment. The use of bank finance changes dramatically with firm size. Average balances on overdrafts are large, approximately Z$ 3 million per firm, about the same amount as outstanding on bank loans. Bank loans are trivial as a source of finance for land (financing less than 10 %) and buildings (less than 5 %). Chapter 6 established that Zimbabwean firms rely more on external finance for investment than is common in Africa and that there is relatively little reliance on informal borrowing. Nevertheless, we found evidence of informal loans, cash and in kind, even for the very largest firms. Whether a firm applies for bank loans is largely a question of size and race (black-owned firms are significantly more likely to apply). Whether an application is approved is largely a function of size and current profitabil ity. Our results are encouraging in the sense that the hypothesis that banks favour established clients and hamper adjustment by keeping loss- making enterprises afloat is not supported by the analysis of loan decisions. However, this must be treated with caution since this mechanism might work through overdrafts rather than loans. In chapter 7 we presented evidence that in the period of structural adjustment, which coincided with a very serious drought, investment rates were maintained at high levels. For 1992 alone 76 (out of 201) firms reported major investments. Output appears not to be constrained by investment: in all sectors there is enormous scope for increased capacity utilisation. Many firms reported that they were producing far below capacity as a result of the effect of the drought on the demand for their products. We found that differences between firms in investment rates are partly due to differences in profitability. There is evidence of rationing in the market for bank loans, including some indication of discrimina- tion of black entrepreneurs. An important result is the finding that exporting firms are less likely to be rationed by banks. A large part of the survey was devoted to questions about characteristics of individual workers: their wages and earnings, education, training and experience. These data were analyzed in Chapter 8. We find that payments in kind (food, clothing) are large relative to cash wages, particularly in the smaller firms and in the food sector where for production workers in kind payments amount to more than 20% of cash payments. The estimated earnings functions support the human capital theory: wage differentials to a large extent reflect differences between workers in schooling and experience. Returns to schooling are high compared to other countries, reflecting the scarcity of educated workers (relative to the unusually high level of industrial development in Zimbabwe). We also found evidence of discrimination, both by race and by gender. Workers employed in large firms enjoy a wage premium which cannot be explained in terms of differences between small and large firms in ., characteristics of the job or in the quality (to the extent we can measure it) of workers. We also established that standard segmentation arguments cannot be used to explain this size effect. Finally, we tried to test whether non-competitive forces (e.g. the effects of · unionisation or of minimum wages) are important in Zimbabwe and concluded that they are 187 not. In chapter 9 we considered the determinants of labour demand. The results indicate that employment is sensitive to wage levels (with own-price elasticities of 0.2 for unskilled and 0.4 for skilled labour). The reforms are causing a fall in real wages in Zimbabwe, partly · through the effects of exchange rate adjustment. The econometric results suggest that the wage changes may induce a substantial increase in employment. Chapter 10 presented evidence on the ties between firms and their suppliers and their clients. Links with suppliers are very strong and of long standing. Whether a firm receives credit from a supplier is to an important extent determined by its profitability. In that sense suppliers screen well. Again, there is some evidence of discrimination: black firms are less likely to receive credit, even when we control for other characteristics of firms, including profitability. A surprising result (and an indication of market imperfections) is that there is a substantial amount of in kind borrowing, even by and from large firms. The extent to which industrial firms in Zimbabwe are exposed to competition is still quite limited. What problems do firms face? Regulations now are relatively unimportant. This is a remarkable success of ESAP: even a few years ago price controls and foreign exchange regulations, to name just two, would have been identified as major problems. Foreign exchange regulations are still listed by many firms as problematic but they ate not described (as they would undoubtedly have been a few years ago) as major problems. Security of supplies is a problem. It is reported as such (with many instances of late delivery) but it is also reflected in the data on stocks of raw materials. For example, in the metal sector the value of these stocks amounts to more than a quarter of the value of sales. Some firms reported holding more than a year's worth of stocks. Another indication of market imperfec- tions (in the sense that firms cannot count on getting the supplies they need when they need them and in the desired quantity) is the (surprising) importance of lending in kind between firms, sometimes in the form of equipment. Finally, there is an issue of market structure. The data collected on the suppliers of inputs indicate that quite often the primary supplier is a monopolist. This, of course, reflects the extent of diversification of Zimbabwe's industrial sector, a legacy of the period of sanctions. This is likely to change if trade liberalisation continues, an issue to be pursued in subsequent rounds of the survey. With many regulations removed, infrastructure now forms a major obstacle. This applies particularly to telephones (identified by almost half of the firms as their greatest problem) and (to a much lesser extent) electricity supply and transport for workers. (This last problem does reflect regulations: competition from out-of-town bus companies is not allowed in Harare. The result is that workers spend an inordinate amount of time getting to and from work.) Restrictions on firing are important and have resulted in increased use of "tempor- ary" contracts. The number of conflicts is large but conflict resolution a'ppears to work well, particularly in cases of late payment: disputes are settled and usually the business relation- ship survives the dispute. The major constraint on firms at the time of the survey (June-July 1993) appears to be lack of demand and this is likely to be temporary, reflecting the drought and adjustment under ESAP. 188 ·