,VP s X1097 POLICY RESEARCH WORKING PAPER 2907 The Case for International Coordination of Electricity Regulation Evidence from the Measurement of Efficiency in South America Antonio Estache Martin A. Rossi Christian A. Ruzzier The World Bank 3 World Bank Institute E Governance, Regulation, and Finance Division and Latin America and the Caribbean Region Finance, Private Sector, and Infrastructure Unit October 2002 POLICy RESIEARCH WORKINGL PAPER 2907 Abstract A decade long experience shows that monitoring the approach that relies on performance rankings based on performance of public and private monopolies in South comparative efficiency measures. The authors show that America is proving to be the hard part of the reform with the rather modest data currently available publicly, process. The operators who control most of the such an approach could yield useful results. They provide information needed for regulatory purposes have little estimates of efficiency levels in South America's main interest in volunteering their dissemination unless they distribution companies between 1994 and 2000. have an incentive to do so. Estache, Rossi, and Ruzzier Moreover, the authors show how relatively simple tests argue that, in spite of, and maybe because of, a much can be used by regulators to check the robustness of their weaker information base and governance structure, results and strengthen their position at regulatory South America's electricity sector could pursue an hearings. This paper-a joint product of the Governance, Regulation, and Finance Division, World Bank Institute, and the Finance, Private Sector, and Infrastructure Unit, Latin America and the Caribbean Region-is part of a larger effort in the institute to increase understanding of infrastructure regulation. Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Gabriela Chenet-Smith, room J3-304, telephone 202-473-6370, fax 202-676-9874, email address gchenet@worldbank.org. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. Theauthorsmaybecontactedataestache@worldbank.orgormartin.rossi@economics.ox.ac.uk. October 2002. (34 pages) The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Produced by the Research Advisory Staff The Case for International Coordination of Electricity Regulation: Evidence from the Measurement of Efficiency in South America* Antonio ESTACHE World Bank and ECARES, Universite Libre de Bruxelles Martin A. ROSSI** University of Oxford Christian A. RUZZIER Centro de Estudios Econ6micos de la Regulaci6n, UADE Departamento de Economfa y Finanzas, UADE ' The paper was prepared as a background note to support ongoing policy dialogue between the World Bank and several Latin American Electricity Regulators. We are grateful to Antonio Alvarez, Phil Burns, Simon Cowan, Rafael Cuesta Alberto Devoto, Luis Guasch, Dany Leipziger, Martin Rodriguez-Pardina, Luis Orea, Sergio Perelman, Lourdes Trujillo and two anonymous referees for helpful discussions and suggestions during the preparation of this paper. Corresponding author: Linacre College, University of Oxford. St. Cross Road, Oxford - OXI 3JA. UK. E-mail: martin.rossi @economics.ox.ac.uk 2 INTRODUCTION Following the process initiated by Chile about 20 years ago, many South American countries have transformed their electricity sector. The changes started with a restructuring to increase competition in and for the markets. They entailed an unbundling of electricity generation, transmission and distribution and resulted in generally competitive generation markets but maintained monopolies for transmission and distribution which were generally auctioned to private operators. Whenever possible, reformers also broke up horizontally the former national distribution companies into several regional monopolies to reduce the strength of the residual monopolies. In most countries, these changes were associated with the creation of new regulatory agencies responsible for the monitoring of the performance of the residual public and private monopolies. A decade long experience shows that this monitoring is proving to be the hard part of the reform. The private operators control most of the specific information needed for regulatory purposes and have little interest in volunteering their dissemination unless they have an incentive to do so. Most of the regulators have tried to mandate the publication of information. Many have also relied on public audiences to promote public debates of relevant information. The results of these approaches to reducing the information asymmetry between regulators and operators have been mixed at best.' This paper argues that in spite of, and maybe because of, a much weaker information base and governance structure, Latin America's electricity sector could, thanks to a much more effective cross-country coordination, reduce the information asymmnetry by relying on 3 performance rankings based on comparative efficiency measures, as achieved with some success by various regulators in England and recently by the Dutch electricity regulator. While never spelled out quite in the specific terms adopted here, what the approach essentially achieves is a shift of the burden of proof for justification of bad performance from the regulator to the operators by relying on competition between markets more systematically.2 The authorized levels of recoverable costs or the performance levels recognized by the regulators to assess the share of efficiency gains to be passed on to consumers can be estimated from best practice benchmarks obtained by comparing performance across markets. Unless the operators can prove with the appropriate information that their performance is sub-par for specific reasons they will have to comply with the regulatory assessment of their performance based on the approaches suggested here. Coordination is needed because this benchmarking approach to regulation, which further promotes competition between markets, requires the best possible assessments of cost or production frontiers across countries and this in turn requires a minimum of coordination in terms of the definition and measurement of the indicators to be used in the process. As large as possible a number of operators must be monitored over 3-4 years at least to maximize the quality of the data available. The paper shows that with the rather modest data currently available publicly, such an approach could already yield useful results.' It provides estimates of efficiency levels in South A theoretical approach to this regulatory problem, in terms of principal-agent games, can be found in Bogetoft (1997), where the selection of a efficiency measurement procedure appears as the Nash equilibrium of a regulatory game. 2 This approach has also been advocated for the Mexican Port sector by Estache, Gonzalez and Trujillo (2002), for instance, and more generally in Coelli, Estache, Perelman and Trujillo (2002). 4 America's main distribution companies between 1994 and 2000. Moreover, it illustrates how relatively simple tests can be used by regulators to check the robustness of their results and strengthen their position at regulatory hearings. This is important since efficiency estimates used by regulators to shift the burden of proof on the operators are likely to be contested routinely by unhappy operators. The quality of the regulatory assessments should be such that improvements in efficiency measures would only come from additional information provided by the operators trying to make their case rather than from improvements in the use of the existing information. The paper is organized as follows. Section 1 specifies the model which could be used by coordinated regulators and argues for a production function rather than a cost function. Section 2 discusses the data currently available to test the chosen model and presents the main characteristics of the 39 distribution companies covered by the data sample. Section 3 covers the various estimation procedures among which to pick. Section 4 explains the test used to check the robustness of the results and discusses the various levels of confidence with which the regulators can argue their case. In particular, this section makes the case for at least a mild form of international yardstick competition between electricity distribution companies in South America. Section 5 concludes. 1. THE SPECIFICATION OF THE MODEL The main challenge for any regulator is to make the most of the information available. This basic, quite obvious, observation has already been internalized by most applied economists working on efficiency measures for electricity companies. This means that pragmatism will often rule over strict theory. While the theory would argue for a detailed structural model accounting for all possible factors, pragmatism implies that the best one can hope to achieve in practice is to estimate a single equation production function. 5 The estimation of a cost function (a valid alternative3) involves an assumption about firms' behavior, namely profit maxiniization. However, whenever there is public ownership, the firms, in general, will not seek profit maximization as their main goal. As Pestieau and Tulkens (1990) argue, public enterprises do not share the same objectives and constraints as their private counterparts, so their relative performance should only be compared on the basis of a production relationship which serves as a common ground. Moreover, the estimation of cost frontiers involves the utilization of variables measured in monetary units, which could be a serious problem if one wishes to make international comparisons. Production functions, instead, only require variables measured in physical units (i.e. homogeneous among countries -or at least much more homogeneous). Given that we are estimating an international frontier and that the sample includes private and public firms as well, we choose to estimate a production function. Having decided upon the relationship to be estimated, we still have to make a decision over the variables that should be included in the analysis. What are the outputs of the industry? What are the inputs? Are there variables beyond the firms' control? The first issue is to decide which output to focus on. According to Neuberg (1977), both number of customers served and total energy sold qualify as potential outputs in this sector. In order to decide between them, some regulatory insights must be taken into account. In particular, it is important to note that energy delivered to final customers is not really exogenous, especially in non-regulated public utilities. That is, the utility is not always compelled to provide its customers with whatever quantities they desire at given prices. Number of customers, on the other hand, cannot be controlled by utilities since in general everybody has the right to be connected to the local distributor. Therefore, energy delivered is a better output measure for the production function specification. 3 Just to name to two most common relationships that are estimated. 6 The next challenge is identifying the inputs. The number of employees is the standard labor input and is easily obtained. As for the capital inputs, the options are more complex. Transformer capacity is widely accepted as a required variable. However, kilometers of distribution lines, which measures the amount of capital in the form of network, can be misleading since it can reflect geographical dispersion of consumers rather than differences in productive efficiency (Kumbhakar and Hjalmarsson, 1998). Therefore, in a study of relative efficiency differences, network capital can either be treated as an output or as input but only after controlling for geographical dispersion. In this paper we adopt the second position and hence correct appropriately by accounting for consumer density. Regarding the environmental variables (variables beyond the firms' control) to be included in the model4, service area is unambiguously an exogenous operating characteristic of the firm's environment. As we argue above, the number of customers served and their distribution is also exogenous, so we include not only service area as a control variable, but also customer density. The idea is that customer density should capture the effect of demographic features, in the sense that higher values of this variable can be expected to enable a firm to deliver more output per unit of input. For similar reasons, we need to measure the effect of delivering energy at different voltages required by different customers, and therefore we include the proportion of total energy delivered that is distributed to residential customers as an additional operating characteristic. Finally, the variable GNP per capita is included to control for differences in the socio-economic environment in which firms operate in each country. 4 Introducing environmental variables in the production function specification assumes that these variables affect technology rather than computed efficiency scores, and generates net efficiency measures. See the discussion in Section 3. 7 The particular choice of variables made here follows the general consensus found in the current literature. We review this literature in the Appendix. Although comparison of some alternative modeling could yield additional insights, we believe that the model chosen is reasonably general in terms of the current literature and that the motivation for the choice of variables is rather convincing. In many cases there are good reasons why some firms do not follow an efficient pattern, but once the regulators have done this initial sorting out, the burden of proof should be on the regulated companies. That is to say, the initial model used as a yardstick is not so determinant, since the firms can impugn the proposed model until every part (fimns and regulators) agree about the final model -involving themselves in a "learning by doing" iterative process in which both firms and regulators learn while playing the game (see Burns and Estache (1998), Rossi and Ruzzier, (2000), Coelli, Estache, Perelman and Trijillo (2002)). Following the discussion above and the availability of data, the initial model for the production function will be: Initial Model Output: Inputs: Environmental variables: 1. Total sales 1. Number of employees 1. Service area 2. Distribution network 2. Customer density 3. Transformer capacity 3. Demand structure 4. GNP per capita The final model will be obtained after testing the statistical significance of the environmental variables. The idea is that a frontier model has two parts: the "core" of the model and the environmental variables (Rossi and Ruzzier, 2000). In a production function approach the 8 (theoretically determined) core is formed by the inputs, whereas the set of environmental variables includes those factors that might influence the firms' perfornance and are not directly controllable by them. The initial specification for the core of the model is subject to theoretical considerations. Environmental variables, on the other hand, are not theoretically determiined and will only be included in the final model if they are statistically and economically significant. 2. THE DATABASE The sample accounts for 39 electricity distribution companies (23 private, 16 public) spread over 10 countries. It is representative of the sector in the region and covers: Argentina (8 firms, including the two largest firms in terms of number of customers), Bolivia (2), Brazil (2), Chile (2), Colombia (2), Ecuador (4), Paraguay (1), Peru (12), Uruguay (1) and Venezuela (5), for the period 1994-2000. The only missing countries are the Guyana, French Guyana and Suriname. The Brazilian sector is probably underrepresented since we only have data on two firms, including the second largest one. Some details are provided in Table 1. Table 1: Firms, Countries and Ownership Country Number of firms covered by the sample Argentina 5 private, 3 public Bolivia 1 private, I public Brazil 2 public Chile 2 private Colombia 2 public Ecuador 3 private, I public Paraguay 'I public Peru 8 private, 4 public Uruguay I public Venezuela 4 private, I public 9 Firim data was collected from several sources. Data for the period 1994-1999 was mostly compiled from CEER (Comisi6n de Integraci6n Electrica Regional - Regional Electric Integration Commission) reports, "Datos Estadisticos. Empresas Electricas. Afno 1994", "Datos Estadfsticos. Empresas Electricas. Afios 1995-1996-1997", "Informaci6n Econ6mica y Tecnica de Empresas El6ctricas. Datos 1998-1999". Data for Peru was partly compiled from CTE (commission in charge of energy tariffs), and data for Argentina in the year 2000 was partly provided by ADEERA (an association of distribution companies). For the most recent data, we relied directly on firms. When possible, the data was cross-checked and completed using firms' balance sheets (or firms' web pages), and information provided by regulators and governmental agencies. When a particular piece of information was missing, in order not to lose the entire observation, some algorithm was used to fill the gap. After eliminating utilities for which data quality was insufficient, we obtained an unbalanced panel with 194 observations from the 39 firms in the period 1994-2000. We only included in our panel firms for which we had at least three consecutive observations. The following variables are going to be used in the estimations: sales (in GWh, calculated as total sales minus sales to other electric companies, in order to isolate the distribution activity in the case of integrated firms), number of employees (in vertical integrated fimns we use only employees in the distribution activity, as informed by the firms), total distribution lines (in kilometers), total transformer capacity (in kVA), service area (in square kilometers), residential sales' share (a proxy for demand structure), customer density in the service area (in customers per square kilometer), and GNP per capita (in purchasing power parity units, PPP). The PPP estimates of GNP per capita for the period 1994-1998 were obtained from the World Development Reports 1996-2000. We used PPP figures in order to correct for international differences in relative prices (for details, see World Development Reports technical 10 notes). The figures for the years 1999 and 2000 were calculated using the World Development Indicators database from the World Bank. The summary statistics are presented in Table 2. In all cases the sample size is equal to 194 observations. Table 2: Su Statistics Variable Sample Sample Minimum Maximum Mean Standard . Deviation Sales (in GWh) 3566 6944 31 37777 Distribution Lines (in km) 21103 55404 443 316997 Number of Employees 1518 2541 26 12239 Transformer Capacity (in kVA) 1440 2207 16 9986 Service Area (in kin) 77878 159682 59 823700 Customer Density 117 203 0.31 677 (in customers per kmn) Residential Sales / Sales (in %) 42 9 17 63 GNP per capita (in PPP units) 6568 2590 2400 13091 3. THE ESTIMATION PROCEDURES To provide a full assessment of the potential value of the information available, we cover as wide a spectrum of approaches regulators could adopt with the data available as possible. We present both econometric and Data Envelopment Analysis (DEA) estimates to assess the efficiency performance of South America's electricity distribution companies. More specifically, we test two parametric models, a stochastic frontier estimated by Maximum Likelihood (ML) and a random effects model estimated by Feasible Generalized Least Squares (FGLS), and two non- parametric DEA (one with variable returns to scale and another with constant returns to scale). 3.1. The econometic models We define the general stochastic frontier production function model by In Y, = f (Xi,, t;,l3) + ei, , where Y,, denotes output, Xi, is a matrix of inputs, t represents time, ,B are technological parameters to be estimated, and f is some appropriate functional form. The error term is it -i= vi, - Ui,, where vi, are assumed independent and identically distributed random errors which have normal distribution with mean zero and unknown variance, i,,, and ui, are non-negative random variables which represent technical inefficiency. The Battese and Coelli (1992) representation (ui, = exp [-q (t - T)] u;) is used for the technical inefficiency term. The time term is included to account for technical change. Representing technical change by including a time term in the production frontier may seem relatively innocuous but it is in fact a very strong assumption and is not always realistic. Many innovations and developments that one would like to subsume under the rubric of technical change are not consistent with this formulation, which assumes that technical change does not require new inputs and further that the production frontier maintains the same basic form as time elapses. However, as many authors point out, including a time term in production frontiers may not be perfect, but it is a workable alternative with some definitive advantages (i.e., analytical and econometric tractability) over some other approaches.5 The translogarithmic (or translog) and the Cobb-Douglas production functions are the two most common functional forms which have been used in empirical studies on production, including frontier analyses. The translog is a flexible function, since it is a second-order Taylor approximation (in logarithms) to any smooth, continuous function. The Cobb-Douglas production frontier is a special case of the translog in which the coefficients of the second order terms are zero. 5 Different null hypothesis associated with technical change are analyzed in Rossi (2002). The results show that non neutral technical change models or models with quadratic time trend do not differ significantly from the more parsimonious model present here. Therefore, in our preferred model we include only a linear time trend. 12 In this paper the most general functional form for the stochastic frontier for electricity distribution in South America is a translog production function: In Y1, = Ao + Xii,A + X2i,f2 + X + + lA + X32J33 + X,i,X2i,A2 +Xl,ix3itAf3 + X2itX3iJt?23 + tA? + Vj - Ui where Y indicates sales, XI is the natural logarithm of the number of permanent employees, X2 is the natural logarithm of distribution network, and X3 is the natural logarithm of transformer capacity. The production function above does not include enviromnental variables. Coelli, Perelman and Romano (1999) suggest that the literature offers two alternative approaches to their inclusion. One assumes that the environmental factors influence the shape of the technology and hence that these factors should be included directly into the production functions as regressors, while the other assumes that they directly influence the degree of technical inefficiency. In this study we adopt the position of including them as regressors in order to get efficiency measures that are net of environmental influences. As pointed out by Coelli, Perelman and Romano (1999), measuring net efficiency is an important issue as it allows one to predict how companies would be ranked if they were able to operate in equivalent environments. Therefore, the most general function to be estimated is as in equation (1) but including four additional environmental variables: ln Y1, = flo + Xii,Af + X20i2 + XA3j,3 + Xl,/31 + X2,,l +3i3 + XutX2J,t2 1i,X3j,itA3 + X21,X3Nf623 + tA + AIflAI + A2 8A2 + A3IA3 + A4f8A4 + Vit -ui where A, is the natural logarithm of demand structure, A2 is the natural logarithm of customer density, A3 is the natural logarithm of service area, A4 and is the natural logarithm of GNP per capita. 13 As it is now usual in this literature, we use the parameterization proposed by Battese and Corra (1977)', which uses y = 2/(a)2+cr2) The program FRONTIER 4.1, developed by T. Coelli (1996), is used for the estimations. In this paper we take advantage of the great flexibility of this model and we test the half- normal distribution hypothesis vis a vis the more general truncated normal distribution (Ho : u = 0), and we also contrast the hypothesis that the efficiency is time invariant (Ho: 0 = 0) . Finally, we test the null hypothesis that there are no technical inefficiency effects in the model; H0:y=0. As suggested by Coelli (1996), these alternative models are estimated and the preferred models are selected using a Likelihood Ratio (LR) test. This test is based on the Log Likelihood functions as follows: LR = -2[LR- Lu], where LR is the Log Likelihood of the restricted model and Lu is the Log Likelihood of the unrestricted model. Asymptotically, the LR statistic has a chi-square distribution with degrees of freedom equal to the number of restrictions involved.6 The ML estimates of the parameters in the unrestricted translog stochastic frontier production function (called Model 1) are shown in Table 4. Formal tests of hypothesis associated to Model 1 are given in Table 3. The first null hypothesis, Ho :,811 = fl22 = 1633 = A2 = /13 = 423 = 0, that the Cobb-Douglas is an adequate representation of the technology is rejected by the data. The second hypothesis, Ho: y = 0, which 6 It must be noted that in the case where the null includes the restriction that y = 0 (a point on the boundary of the parameter space), the likelihood ratio statistics will have asymptotic distribution equal to a mixture of chi-square distributions - + - (Coelli 1993, Lee 1993). 2 zo2 14 specifies that firms are fully efficient is strongly rejected. The null that the inefficiency has a half- normal distribution, H. , =0, cannot be rejected by the data, and therefore in our preferred model we work assuming a half-normal distribution for the inefficiency terms. The null hypothesis that the technical inefficiency is time invariant, Ho: q = 0, cannot be rejected. Finally we test the significance of the environmental variables. The null hypothesis Ho: 1A1 = PA2 = flA3 = I6A4 = 0 is strongly rejected by the data, suggesting that environmental variables cannot be omitted in the estimation of production frontiers in this kind of sector. A fact which would probably argued for by most operators. Table 3: Likelihood Ratio Tests Null Hypothesis Log Likelihood X2 value Test statistic* Given Model 1 167.00 _ Ho: __= _ _2 = _33_=_2 = _=23 = ° 155.79 12.59 22.41* Ho:rY = ° 17.40 6.25 299.21* Ho pU = 0 166.66 3.84 0.67 Ho: = 0 166.98 3.84 0.03 Ho: flAl = PA2 = PA3 = flA4 = 0 117.27 9.49 99.46* *An asterisk on the value of the test statistic indicates that it exceeds the 99' percentile for the corresponding X2 distribution and so the null hypothesis is rejected. The above tests suggest that the preferred model (we call it Model IP) is a translog stochastic production function with neutral technical change and time-invariant inefficiency, which is assumed distributed as a half-normal. The production function includes demand structure, customer density, service area and GNP per capita as environmental variables. Since we cannot reject the hypothesis of constant technical efficiency, we can run Model IP as a random effects model (we call it Model IG). The ML estimates of the unrestricted model (Model 1) and the preferred model (Model IP), and FGLS estimates of the preferred model (Model IG) are shown in Table 4. 15 Table 4: Econometric Results Variable Model 1 Standard Model 1P Standard Model iG Standard Errors Errors Errors Constant -4.861 1.393 -5.914 1.205 -5.223 1.550 Ln Employee -0.386 0.219 -0.388 0.211 -0.346 0.240 Ln Net 0.328 0.288 0.171 0.269 0.210 0.323 Ln Capacity 0.162 0.230 0.357 0.220 0.179 0.265 (In Employee) 0.029 0.025 0.043 0.023 0.014 0.026 (In Net) -0.012 0.022 0.001 0.023 -0.011 0.026 (In Capacity)' 0.156 0.030 0.156 0.032 0.145 0.032 Ln Employee x In Net 0.095 0.034 0.091 0.031 0.102 0.039 Ln Employee x In Capacity -0.129 0.047 -0.146 0.047 -0.118 0.054 Ln Net x In Capacity -0.108 0.036 -0.120 0.035 -0.102 0.039 Ln Demand Structure -0.517 0.061 -0.511 0.060 -0.536 0.064 Ln Customer Density 0.725 0.091 0.763 0.065 0.781 0.082 Ln Service Area 0.695 0.086 0.726 0.059 0.744 0.086 Ln GNP per capita 0.105 0.091 0.180 0.072 0.057 0.092 Time 0.016 0.009 0.013 0.005 0.014 0.005 . 0.982 0.013 0.987 0.005 P 0.495 0.157 77 -0.003 0.013 _ Average Efficiency 0.578 0.657 0.564 Since the coefficients of the translog production functions do not have any direct interpretation, we calculate the elasticities of output with respect to each of the inputs corresponding to models above 3ya EL e = =,lk + 2,8kXu, + E p8k X jj,, k = 1, 2, 3;j = 1, 2,3 . In general, returns to scale is calculated from the sum of the input elasticities as RTS = E E4. k However, it is sometimes noted that when the model includes environmental variables related to scale (such as service area), the scale elasticity is given by the proportionate effect on production of changes in the input variables and these environmental variables. The main point is that changing the scale of a firm would involve changing not only the inputs but also all of these characteristics (Burns and Weyman-Jones, 1994). Given customer density, demand structure and 16 the socio-economic conditions, returns to scale should be defined as relating the change in output to a change in all inputs and service area. That is, RTS = E ELk + 1A3- k The following table shows input elasticities, service area elasticity and returns to scale for both preferred models, Model IP and Model 1G. Input elasticities are calculated at the sample means values (the Taylor series expansion points). Table 5: Elasticities and Returns to Scale Elasticity with respect to Model Employees KM of Transformer Service Area Returns to network Capacity scale Model 1P 0.08 -0.01 0.36 0.73 1.15* Model IG 0.02 -0.01 0.40 0.74 1.16* *Reject the null of constant returns to scale at a 5% level. Elasticities with respect to service area and transformer capacity are positive and quite comparable across models. However, we cannot reject the null that labor and network elasticity are equal to zero in both models. As expected, in both models returns to scale are significantly greater than one. The estimated coefficients of the environmental variables have the expected signs. The negative influence of demand structure implies that firms with a lower proportion of residential customers benefit from a more favorable environment and hence perform better when no attempt is made to take into account this advantage. Customer density has a positive effect on output, which means that as the number of customers per square kilometer rises (ceteris paribus), energy delivered will consequently go up. Service area has also a positive sign, since given customer density it is playing an input role. Finally, the positive coefficient of GNP per capita suggests that firms operating in countries with high GNP per capita benefit from a more favorable socio- economic environment. 17 The annual rate of technical change is 1.3 percent in Model IP and 1.4 percent in Model 1G. Finally, average efficiency is around 66 percent in Model IP, and around 56 percent in Model 1G. These results suggest that there is scope for efficiency improving for the average firm in the sample. 3.2. The DEA estimates In order to allow for the comparison of the results, we used the same model as in the last section to perform the nonparametric estimation, i.e. we have a model with only one output (total -sales), three inputs (labor, km of distribution lines and transformer capacity), and four environmental variables (service area, customer density in the service area, a proxy for demand structure and GNP per capita). The orientation chosen is to the proportional augmentation in output achievable by a firm while maintaining the level of inputs, for this is consistent with the interpretation of the econometric results. There exist basically two alternative assumptions about the returns to scale: constant returns to scale (DEA-C) and variable returns to scale (DEA-V). The theoretical specification of the DEA-C model consists in an optimization problem subject to constraints, like the following: max A s.to Au i to avoid redundancy. Table 9 presents the average correlations by the number of years apart. In general, the n-year apart figures are averages of the 7-n correlations between efficiencies that are n years away from each other. Table 9: Correlations Between DEA Efficiency Measures Approach 1 year apart 2 year apart 3 year apart 4 year apart 5 year apart 6 year apart DEA-V 0.836 0.647 0.564 0.536 0.639 0.629 DEA-C 0.750 0.607 0.574 0.677 0.865 0.702 The correlations are high and statistically significant over all the available lags, suggesting that the efficiency scores of the DEA-V and DEA-C models are stable over time and giving additional support to the result of no efficiency change obtained with the parametric techniques used here. 5. CONCLUSIONS The most important result of this paper has been to show that yardstick or benchmark competition organized around measures of technical efficiency is possible, at least in a mild form. This is not to say that the operators will not complain and question not only the results but also the methodologies. But this is normal. Regulation amounts to a game played between regulators and operators, most of the time, with the purpose of allocating the rent generated by the regulated 25 monopolistic business between operators, users and the government. Too often in the past the game has been biased in favor of firms since they control much of the information. This implies that too often the efficiency gains actually achieved through restructuring and competition for the market have not been shared with the final users. This approach levels the playing field by providing the regulator in each country with an instrument that reduces the information asymmetry. By allowing the regulator to propose its own estimate of the rent to be distributed based on the best practice defined by the performance of the top 5 or 10 firms, the approach proposed here forces the regulated firms unhappy with the regulator's assessment to reveal more information than it otherwise would. A necessary condition for this form of competition to work is for regulators to coordinate with the other regulators in the region in a much more focused way than they have done in the past. For this sector and for most countries, the performance comparison can only be international. The more comparable across countries the information is, the more effective is this form of competition and the easier it is for each individual regulator to rely on useful results in its own regulatory settings. 26 APPENDIX The applied literature is a good starting point in the identification of the variables to be included in the model. In the following table we summarize previous works found in the applied literature, highlighting the specification used (cost vs. production), the estimation technique (econometrics vs. mathematical programming), the outputs, the inputs and the environmental variables chosen. Table A.1 Summary of Previous Studies Author/s Specification/ Output/s Inputs?3 Environmental Variables ____ ___ ___ Estimation Neuberg, Cost function, Customers Capital, labor MWh sold, KM of 1977 Econometrics distribution line, service area Huettner and Cost function, Total capacity, Labor Line transformers per Landon, 1977 Econometrics average demand customer, residential, as a ratio of commercial and industrial maximum sales per customer, and a set capacity of dummy variables Roberts, Cost function, High and low KWh input, capital 1986 Econometrics voltage (transmission and deliveries, distribution), labor serviced area, customers Nelson and Cost function, Number of Lines, Labor City size, a dummy variable Primeaux, Econometrics customers for the nature of the 1988 competitive environment New Zealand Cost function, Electricity Labor, capital, electricity Ministry of Econometrics distributed purchased and "other" Energy, 1989 Weyman- Production Residential, Labor, mains Jones, 1991 approach, commercial and distribution DEA industrial sales Weyman- Production Residential, Labor, network size, Jones, 1992 approach, commercial and transformer capacity DEA industrial sales, maximum demand Weyman- Production Customers Labor Network size, transformer Jones, 1992 approach, capacity, total sales, DEA maximum demand, population density, industrial share in sales Hjalmarsson Production High and low Labor, high and low and approach, voltage output voltage lines, Veiderpass, DEA (MWh), high and transformer capacity 13 In cost approaches, inputs prices are used in the models instead of input quantities. 27 Author/s Specificationl Output/s Inputsl3 Environmental Variables Estimation 1992a,b low voltage customers Hougaard, DEA Length of power Labor, operating 1994 lines, total power expenses, operating deliveries, capital, transmission number of losses customers Salvanes and Cost function, GWh produced, Labor, purchased Load factor, topography, Tj0tta, 1994 econometrics number of electricity climate, dummy rural area customers Kittelsen, DEA Length of power Labor, transmission 1994 lines, total power losses, extemal services deliveries, bought number of customers Bums and Production Customers, Labor, distribution Consumer density, market Weyman- approach, domestic, network, transformer structure Jones, 1994 DEA commercial and capacity industrial sales, maximum demand Pollitt, 1995 Cost function, Sales per Labor % of residential sales, Econometrics customer, ratio overground and underground maximum to distribution circuits, average demand, transformer capacity, service Customers area, and a set of dummy variables Pollitt, 1995 Production Customers, Number of employees, approach, residential sales, transformer capacity, DEA non-residential circuit kilometers sales, service area, maximum demand Bagdadioglu, DEA Customers, Labor, transformer Waddams electricity capacity, network size, Price and supplied, network losses, general Weyman- maximum expenses Jones, 1996 demand, service area Burns and Cost function, Customers Labor, capital Maximum demand, service Weyman- Econometrics area, consumer density, kWh Jones, 1996 sold, market structure,'4 kilometers of mains line, transformer capacity Thompson, Cost function High and low Labor (transmission and Service area, number of 1997 voltage sales distribution), power, customers capital (transmission and distribution plants) Zhang and DEA Total number of Transformer capacity, Bartels, 1998 customers labor, total km of I___________ distribution lines 14 Market structure is defined as the share of industrial energy delivered in total energy delivered. 28 Authorls Specificationl Output/s Inputs'3 Environmental Variables Estimation Forsund and DEA Distance index, Labor, energy loss, Kittelsen, customers, total materials, capital 1998 energy delivered Filippini, Cost function, KWh delivered, Labor, capital, Load factor, service area 1998 econometrics number of purchased power ______________ customers Kumbhakar Production High and low Labor, transformer and approach, voltage capacity, kilometers of Hjalmarsson, DEA and customers, high low and high voltage *1998 Econometrics and low voltage lines energy sold Scarsi, 1999 Production Energy delivered Labor, kilometers of approach, to final distribution lines DEA customers, number of customers Scarsi, 1999 Production Energy delivered Labor, kilometers of Customer density and a set of function, to final distribution lines dummy variables econometrics customers Scarsi, 1999 Cost function, GWh sold, Capital, labor, materials Customer density, demand Econometrics customers structure, % of third-party services, % of overhead low- voltage lines, % of primary substations, and a set of dummy variables Kittelsen, Cost Energy delivered, Labor, energy loss, 1999 approach, customers, line transformers, lines, DEA length 1-24 kV goods and services. DTe, 2000 Cost Units distributed, Operating expenditures efficiency, small customer DEA numbers, large customer numbers, network length, transformer numbers, network density Grifell-Tatje DEA Low, medium Low, medium and high and Knox and high voltage voltage lines, substation Lovell, 2000 customers, area, transformer capacity low, medium and high voltage sales, service reliability Langset, DEA Energy Supplied Labor, energy losses, 2000 (high and low capital, goods and voltage), number services. of customers (high and low voltage), length of lines (by kV) Jamasb and Econometrics Energy delivered, Controllable operating Distribution losses, number 29 Author/s Specificationl Output/s Inputs3 Environmental Variables Pollitt, 2001 and DEA, cost number of expenditures, capital of transformers function customers expenditures (residential and non-residential), length of network (overhead and underground cables) Filippini and Cost function, KWh transported Labor, capital Customer structure, load Wild, 2001 econometrics on the medium- factor, customer density, voltage grid average consumption, share of agricultural, forest and unproductive land, other revenues, dummy high- voltage 30 REFERENCES Battese, G. and G. Corra (1977), "Estimation of a Production Frontier Model: With Application to the Pastoral Zone of Eastern Australia." Australian Journal of Agricultural Economics 21, 39-56. Battese, G. and T. Coelli (1992), "Frontier Production Functions, Technical Efficiency and Panel Data: With Application to Paddy Farmers in India." Journal of Productivity Analysis 3, 153-169. Bagdadiouglu, N., C. Waddams Price and T. Weyman-Jones (1996), "Efficiency and Ownership in Electricity Distribution: A Non-Parametric Model of the Turkish Experience." Energy Economics 18, 1-23. Bauer, P., A. Berger, G. Ferrier and D. Humphrey (1998), "Consistency Conditions for Regulatory Analysis of Financial Institutions: A Comparison of Frontier Efficiency Methods." Journal of Economics and Business 50, 85-114. Bogetoft, P. (1997), "DEA-Based Yardstick Competition: The Optimality of Best Practice Regulation." Annals of Operations Research 73, 277-298. Burns, P. and T. Weyman-Jones (1994), "Regulatory Incentives, Privatization, and Productivity Growth in UK Electricity Distribution." CRI Technical Paper N°l. Burns, P. and T. Weyman-Jones (1996), "Cost Functions and Cost Efficiency in Electricity Distribution: A Stochastic Frontier Approach." Bulletin of Economic Research 48. Burns, P. and A. Estache. (1998), "Information, Accounting and the Regulation of Concessioned Infrastructure Monopolies", The World Bank, Policy Research Working Paper 2034 31 Charnes, A., W.W. Cooper, A.Y. Lewin, R.C. Morey and J. Rousseau (1985), "Sensitivity and Stability Analysis in DEA." Annals of Operations Research 2, 139-156. Charnes, A., W.W. Cooper, A.Y. Lewin and L.M. Seiford (1994), Data Envelopment Analysis. Theory, Methodology and Applications. Kluwer Academic Publishers, Boston/ Dordretch/London. Coelli, T. (1993), "Finite Sample Properties of Stochastic Frontier estimators and Associated test Statistics." Working Paper in Econometrics and Applied Statistics No. 70, Department of Econometrics, University of New England, Armindale. Coelli, T. (1996), "A Guide to FRONTIER Version 4.1: A Computer Program for Stochastic Frontier Production and Cost Function Estimation." CEPA Working Paper 96/07. Coelli, T., D. Prasada Rao and G. Battese (1998), An Introduction to Efficiency and Productivity Analysis. Kluwer Academic Publishers. Coelli, T., S. Perelman and E. Romano (1999), "Accounting for Environmental Influences in Stochastic Frontier Models: With Application to International Airlines." Journal of Productivity Analysis 11, 251-273. Coelli, T., A. Estache, S. Perelman and L. Trujillo (2002), A Primer on Efficiency Measurement for Utilities and Transport Regulators, The World Bank Institute, Studies in Economic Development, forthcoming. DTe (2000), "Choice of Model and Availability of Data for the Efficiency Analysis of Dutch Network and Supply Businesses in the Electricity Sector." Background Report, Netherlands Electricity Regulatory Service, February 2000. Estache, A., M. Gonzalez and L. Trujillo (2002), "Efficiency Gains from Port Reform and the Potential for Yardstick Competition: Lessons from Mexico." World Development, April, Vol. 30, No. 4. 32 Filippini, M. (1998), "Are Municipal Electricity Distribution Utilities Natural Monopolies?" Annals of Public and Cooperative Economics 69 (2), 157-174. Filippini, M. and J. Wild (2001), "Regional Differences in Electricity Distribution Costs and their Consequences for Yardstick Regulation of Access Prices." Energy Economics 23, 477- 488. F0rsund, F. and S. Kittelsen (1998), "Productivity Development of Norwegian Electricity Distribution Utilities." Resource and Energy Economics 20 (3), 207-224. Grifell-Tatje, E. and C. Knox Lovell (2000), "The Managers versus the Consultants." Mimeo. Hjalmarsson, L. and A. Veiderpass (1992a), "Efficiency and Ownership in Swedish Electricity Retail Distribution." Journal of Productivity Analysis 3, 7-23. Hjalmarsson, L. and A. Veiderpass (1992b), "Productivity in Swedish Electricity Retail Distribution." Scandinavian Journal of Economics 94 (S), 193-205. Hougaard, J. (1994), "Produktivitetsanlyse af Dansk Elproduktion." AKF-rapport, AKF Forlag, Copenhagen. Huettner, D. and J. Landon (1977), "Electric Utilities: Scale Economies and Diseconomies." Southern Economic Journal 44, 883-912. Jamasb, T. and M. Pollit (2001), "International Benchmarking & Yardstick Regulation: An Application to European Electricity Distribution Utilities." Paper presented in the Seventh European Workshop on Efficiency and Productivity, Spain, September 2001. Kittelsen, S. (1994), "Effektivitet og Regulering i Norsk Elektrisitetsdistribusjon." SNF- rapport 3/94, SNF, Oslo. 33 Kittelsen, S. (1999), "Using DEA to Regulate Norwegian Electricity Distribution Utilities." Presentation at the 6h European Workshop on Efficiency and Productivity Analysis, Copenhagen. Kumbhakar, S. and L. Hjalmarsson (1998), "Relative Performance of Public and Private Ownership Under Yardstick Competition: Electricity Retail Distribution." European Economic Review 42, 97-122. Langset, T. (2000), "Benchmarking Distribution Systems." Paper presented in the Workshop on International Benchmarking of Electric Utilities: Experiences and Information Requirements, University of Cambridge. Lee, L. (1993), "Asymptotic Distribution of the Maximum Likelihood Estimator for a Stochastic Frontier Function Model with a Singular Information Matrix." Econometric Theory 9, 413-430. Nelson, R. and W. Primeaux Jr. (1988), "The Effects of Competition on Transmission and Distribution Costs in the Municipal Electric Industry." Land Economics 64, 338-346. Neuberg, L. (1977), "Two Issues in the Municipal Ownership of Electric Power Distribution Systems." Bell Journal of Economics 8, 303-323. New Zealand Ministry of Energy (1989), "Performance Measures and Economies of Scale in the New Zealand Electricity Distribution System." Wellington, Ministry of Energy. Pestieu, P. and Tulkens, H. (1990), "Assessing the Performance of Public Sector Activities: Some Recent Evidence from the Productive Efficiency Viewpoint." Discussion Paper N°9060, CORE, Universit6 Catholique de Louvain, Belgium. Pollitt, M. (1995), Ownership and Performance in Electric Utilities: the International Evidence on Privatization and Efficiency, Oxford University Press. 34 Roberts, M. (1986), "Economies of Density and Size in the Transmission and Distribution of Electric Power." Land Economics 62 (4), 337-346. Rossi, M. (2002), "Production Frontier Analysis: Electricity Distribution Sector in South America." Unpublished Thesis, University of Oxford. Rossi, M. and C. Ruzzier (2000), "On the Regulatory Application of Efficiency Measures." Utilities Policy, vol. 9, 81-92, June. Salvanes, K. and S. Tj0tta (1994), "Productivity Differences in Multiple Output Industries: An Empirical Application to Electricity Distribution." Journal of Productivity Analysis 5 (1), 23- 43. Scarsi, G. (1999), "Local Electricity Distribution in Italy: Comparative Efficiency Analysis and Methodological Cross-Checking." London Economics, December. Seiford, L. and R. Thrall (1990), "Recent Developments in DEA: The Mathematical Programming Approach to Frontier Analysis." Journal of Econometrics 46, 7-38. Thompson, H. (1997), "Cost Efficiency in Power Procurement and Delivery Service in the Electric Utility Industry." Land Economics 73 (3), August, 287-296. Weyman-Jones, T. (1991), "Productive Efficiency in a Regulated Industry. The Area Electricity Boards of England and Wales." Energy Economics, April, 116-122. Weyman-Jones, T. (1992), "Problems of Yardstick Regulation in Electricity Distribution". In Bishop, Kay and Mayer, The Regulatory Challenge, Oxford University Press. Zhang, Y. and R. Bartels (1998), "The Effect of Sample Size on the mean Efficiency in DEA with Application to Electricity Distribution in Australia." Journal of Productivity Analysis 9, 187-204. Policy Research Working Paper Series Contact Title Author Date for paper WPS2881 Returns to Investment in Education: George Psacharopoulos September 2002 N. Vergara A Further Update Harry Anthony Patrinos 30432 WPS2882 Politically Optimal Tariffs: Dorsati Madani September 2002 P. Flewitt An Application to Egypt Marcelo Olarreaga 32724 WPS2883 Assessing the Distributional Impact B. Essama-Nssah September 2002 0 Kootzemew of Public Policy 35075 WPS2884 Privatization and Labor Force Alberto Chong September 2002 H. Sladovich Restructuring around the World Florencio Lopez-de-Silanes 37698 WPS2885 Poverty, AIDS, and Children's Martha Ainsworth September 2002 H Sladovich Schooling: A Targeting Dilemma Deon Filmer 37698 WPS2886 Examining the Feasibility of Jerry R. Skees September 2002 E. Laguidao Livestock Insurance in Mongolia Ayurzana Enkh-Amgalan 82450 WPS2887 The Demand for Commodity Alexander Sarris September 2002 M. Fernandez Insurance by Developing Country 33766 Agricultural Producers: Theory and an Application to Cocoa in Ghana WPS2888 A Poverty Analysis Macroeconomic Luiz A. Pereira da Silva September 2002 R. Yazigi Simulator (PAMS) Linking Household B. Essama-Nssah 37176 Surveys with Macro-Models Issouf Samake WPS2889 Environmental Performance Rating Hua Wang September 2002 Y. D'Souza and Disclosure: China's Green- Jun Bi 31449 Watch Program David Wheeler Jinnan Wang Dong Cao Genfa Lu Yuan Wang WPS2890 Sector Organization, Governance, Antonio Estache September 2002 G. Chenet-Smith and the Inefficiency of African Water Eugene Kouassi 36370 Utilities WPS2890 Sector Organization, Governance, Antonio Estache September 2002 G. Chenet-Smith and the Inefficiency of African Water Eugene Kouassi 36370 Utilities WPS2891 Trends in the Education Sector from Nga Nguyet Nguyen September 2002 E. Khine 1993-98 37471 WPS2892 Productivity or Endowments? Hiau Looi Kee September 2002 P. Flewitt Sectoral Evidence for Hong Kong's 32724 Aggregate Growth WPS2893 Banking on Foreigners: The Behavior Maria Soledad Martinez September 2002 A. Yaptenco of International Bank Lending to Peria 31823 Latin America, 1985-2000 Andrew Powell Ivanna Vladkova Hollar WPS2894 Telecommunications Sector Jean-Paul Azam September 2002 P. Sintim-Aboagye Reforms in Senegal Magueye Dia 38526 Tchetche N'Guessan Policy Research Working Paper Series Contact Title Author Date for paper WPS2895 Telecommunications Reform in Jean-Jacques Laffont September 2002 P. Sintim-Aboagye C6te d'lvoire Tchetche N'Guessan 38526 WPS2896 The Wage Labor Market and John Luke Gallup September 2002 E. Khine Inequality in Vietnam in the 1990s 37471 WPS2897 Gender Dimensions of Child Labor Emily Gustafsson-Wright October 2002 M. Correia and Street Children in Brazil Hnin Hnin Pyne 39394 WPS2898 Relative Returns to Policy Reform: Alexandre Samy de Castro October 2002 R. Yazigi Evidence from Controlled Cross- Ian Goldin 37176 Country Regressions Luiz A. Pereira da Silva WPS2899 The Political Economy of Fiscal Benn Eifert October 2002 J. Schwartz Policy and Economic Management Alan Gelb 32250 in Oil-Exporting Countries Nils Borje Tallroth WPS2900 Economic Structure, Productivity, Uwe Deichmann October 2002 Y. D'Souza and Infrastructure Quality in Marianne Fay 31449 Southern Mexico Jun Koo Somik V. Lall WPS2901 Decentralized Creditor-Led Marinela E. Dado October 2002 R. Vo Corporate Restructuring: Cross- Daniela Klingebiel 33722 Country Experience WPS2902 Aid, Policy, and Growth in Paul Collier October 2002 A. Kitson-Walters Post-Conflict Societies Anke Hoeffler 33712 WPS2903 Financial Globalization: Unequal Augusto de la Torre October 2002 P. Soto Blessings Eduardo Levy Yeyati 37892 Sergio L. Schmukler WPS2904 Law and Finance: Why Does Legal Thorsten Beck October 2002 K. Labrie Origin Matter? Asl1 Demirgu,-Kunt 31001 Ross Levine WPS2905 Financing Patterns Around the World: Thorsten Beck October 2002 K. Labrie The Role of Institutions Ashl Demirgu,-Kunt 31001 Vojislav Maksimovic WPS2906 Macroeconomic Effects of Private Lourdes Trujillo October 2002 G. Chenet-Smith Sector Participation in Latin Noelia Martin 36370 America's Infrastructure Antonio Estache Javier Campos