POLICY RESEARCH WORKING PAPER   1276
Is There Persistence                                                   Asymmetry in the income
elasticity, of demand, and t--
in the Growth                                                          observed persistence of
of Manufactured Exports?                                               exports, suggest that long-
term buyer-supplier
relationships lead to the
creation of 'insiders' and
Evidence  from   Newly                                                'outsiders' in the world
Industrializing  Countries                                             market for manufactured
goods - a condition that
Ashoka Mody                                                            tends to perpetuate itself.
Kamil Yilmaz
Tlhe World Bank
Polity Research Department
Trade Policy Division
March 1994



I POLICY RESEARCH WORKING PAPER 1276
Summary findings
Price and income elasticities estimated from a country's       More importa;.t, when world income rises, exports
export demand function are used both to predict and to       rise relatively uniformly for different country groups. As
prescribe effective export strategies. But the focus on      world income contracts, the decline in exports is greater
elasticities has led to the neglect of an important          and is especially sharp for certain countries.
empirical regularity: a strong persistence in the growth       Mody and Yilmaz infer from this asymmetry in income
rate of a country's exports.                                 elasticity of demand, and from the observed persistence
Mody and Yilmaz shift the spotlight to this                of exports, that long-term buyer-supplier relationships
phenomenon and describe the degree and pattern of            lead to the creation of "insiders" and "outsiders" in the
persistence.                                                 world market for manufactured goods, a condition that
They find that a country's exports are influenced not      tends to perpetuate itself.
only by the elasticities, but also by the quality of its
transactional infrastructure (proxied by the penetration
of telecommunications).
This paper - a product of the Trade Policy Division, Policy Research Department - is part of a larger effort in the
department to study the factors which directly or indirectly affect the export performance of less developed countries
Copies of this paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact
Minerva Patefia, room N10-013, extension 37947 (43 pages). March 1994.
The Policy Research Working Paper Se,*s disseninates the findings of work in progress to encourage the exdhange of ideas about
development issues. An obfective oftheseries is to get the findings out quickly, even if the presentationsare kss thanfullypolished. The
papers ca4ry the names of the authors and should be used and cted accordingly. The findings, interpretations, and concusions are the
authos' oum and should not be attributed to the WorldBank, its EecutiveBoard of Diretors, or any of its member countries.
Produced by the Policy Research Dissemination Center



IS THERE PERSISTENCE IN THE GROWTH OF
MANUFACTURED EXPORTS ?
Evidence from Newly Industrializing Countries
Ashoka Mody and Kamil Yilmaz
The World Bank
This paper has benefitted from comments by Bela Balassa, Nancy Barry, Ken
Chowmitz, Mary Lou Egan, Ann Harrison, Kala Krishna, Jenny Lanjouw, Bee Roberts,
James Tybout, David Wheeler and, especially, Mark Schankerman.



I.    Introduction
A country's export demand function relates its export volume to the
relative price of its products and to the incomes of international buyers. Price
and income elasticities of demand estimated from such functions are used both for
predicting exports and prescribing effective export strategies.
The focus on price and income elasticities has, however, led to the
neglect of an important empirical regularity: a strong persistence in the growth
rate of a country's exports. Persistence can arise from a slow adjustment to
short-term demand fluctuations, lasting typically for several quarters. Such
inertia is of limited interest to us. In this paper, we are concerned with a
persistence of much longer duration, implying the influence of institutional
features that exert long-lasting effects on export growth.
Evidence on persistence can be seen in different versions of export
demand functions. When the variables are expressed in levels, export demand
functions tend to systematically over- or under-estimate export levels: in other
words, the "residuals" (actual minus estimated exports) have a high degree of
positive serial correlation, reflected in Durbin-Watson statistics of the order
of 0.75 (see, for example, Krugman and Baldwin 1987, Landesmann and Snell 1989,
and Bhalla 1989). This same characteristic of export growth is seen more sharply
when the variables of the demand function are represented as rates of growth: in
addition to growth explained by price and world income changes, a non-zero,
country-specific growth rate (fixed effect) is observed.
More often, persistence in export growth rates is obscured due to the
use of ad hoc procedures when estimating export demand functions. First, long-
term persistence is misread as a short-term adjustment to excess supply and
demand conditions and is accounted for (incorrectly, in our judgement) by the
inclusion of lagged export volume as an "explanatory" variable (for a recent



2
example, see Marquez and McNeilly 1988).
Second, persistent evolution of export volumes is subsumed in high
income elasticities of demand. For some industrial countries (notably Japan) and
many developing countries, income elasticities are in the range of 2.5 to 5.0,
i.e., a one percent increase in world income increases their exports by 2.5 to
5 percent (see Marquez and McNeilly 1988 for a review of selected studies).
Recently, Muscatelli, Sr ivasan, and Vines (1992) estimated Hong Kong's income
elasticity of demand to be 4.2. Most authors are generally uncomfortable when
reporting such high elasticities. Muscatelli, Srinivasan and Vines (1992), for
example, note that the high elasticities are due to: "a failure of conventional
models of export flows (including our own) to identify important forces causing
shifts in export demand: 'income effects' thus probably subsume a variety of
other non-price factors."   We will show that while their instinct on shifts in
demand are right, their interpretation of high income elasticities is probably
faulty.
Finally, Helkie and Hooper (1988) provide a more explicit accounting
of persistence, using the stock of capital in the exporting country as a proxy
for secular changes in the capability to supply an increasing range of products.
Their defense for "this unabashedly ad hoc adjustment is that the existing price
indexes do not adequately capture the price effects of the introduction of new
product lines."  Similarly, Krugman and Baldwin (1987) add a time trend variable
to their export demand equation to account for long-term changes.
Our purpose is to cast a spotlight on the long-term persistence found
in the data, examine its robustness, seek statistical proxies that may account
for the persistence, and provide an interpretation of the observed patterns. In
pursuing this investigation, we believe that we have identified a much richer set



3
of export determinants than are implied in the traditional models t1at focus on
income and price elasticities and short-run adjustment.
Our first effort is to identify and examine sources of errors and
biases that may lead to exaggerated estimates of the persistence effect.  A
specific concern is the existence of errors in the measurement of relative price.
Aw (1992) and Feenstra (1992) have taken the approach that such errors are
minimized when the demand equation is estimated for narrowly defined products
rather than for manufactured goods as a whole.  Although measurement errors
obviously exist, and the choice of instruments used to account for the errors has
an influence on estimated price elasticities, these considerations are not
sufficient to explain away the long-term country-specific persistence.!
We suggest that the observed persistenci reflects a diffusion of
demand from industrial'zed to newly industrializing economies and zould be
considered an evolution by developing country exporters from "outsider" to
"insider" status. An outsider is a marginal supplier; an insider is a supplier
with whom the buyer has a long-term relationship in which both parties have made
(tangible and intangible) investments. Insiders are part of an extensive network
of buyer-supplier relationships and draw on this capital to maintain high growth
rates.
We provide ind4rect evidence of an "insider-outsider" phenomenon in
world markets for manufactured goods by examining asymmetries in the income
elasticity of demand for different groups of countries. Specifically, we find
that the magnitude of the export response depends upon whether the buyers'
incomes rise or fall.    When world income rises, exports rise relatively
Y/    Benhabib and Jovanovic (1991) also observe persistence over 15 to 25 years
in the growth rates of per capita incomes in a wide range of countries.



4
uniformly for different country groups; the decline in exports with world income
contraction is larger and especially sharp for certain countries.   Suppliers
facing high elasticities on the down-side are marginal to the buyer. When the
distinction between the rise and fall of world income is not made, high (average)
elas:,icities are often incorrectly interpreted as a sign of successful export
performance.
Countries that have profited from the shift to insider status have
not been passive beneficiaries. Racher, they have invested in improving their
transactions infrastructure,  making them easier to do business with.   The
development of a country's telecommunications network appears to be a partial
proxy for the ability to deliver time- and communication-sensitive services that
are relevant for developing country exports of such goods as garments, shoes,
bicycles, consumer electronics, and auto parts.
The paper is organized as follows. In section II, we describe how
the degree of persistence has varied across (and within) countries over time.
We also discuss and evaluate issues relating to mismeasurement and the choice of
proper instruments.   In Section III, we examine asymmetries in the income
elasticity of demand. The use of telecommunications as a proxy for the quality
of a country's transactional infrastructure is described in Section IV.   In
Section V, we present a model of demand diffusion and long-term contracts that
is consistent with the observed evidence. The conclusion evaluates the evidence
and comments on its policy relevance.



II. Patterns of Persistence
We begin with the following demand equation:
(1)       A.ogE,d = X + A1og(PidP,)   al Alog(PX.1/Pl'.A) * PAlogy,,
Equation 1 is designed to estimate a set of parameters from pooled observations,
for a number of countries and several years. Each variable is defined for a
country i and a time t.  Ed is the demand for a country's exports, px is the
price of the country's exports, pw is the price of exports by competitors
(proxied by a world price index) and yw is the world income relevant to the
country (the weighted sum of the purchasing countries' GDP, where the weight is
the average share of each purchasing country in the total exports of the country
in question for the sample period).V   The coefficients a and S are the price
and income elasticities of demand, respectively.y The symbol A indicates that
the equation is specified in first differences (which approximates to the rate
of growth when variables are expressed in their logarithmic value).
I/    We choose to use the average export share for the sample period, rather
than the export share for the corresponding year to avoid the possibility of
introducing endogeneity into our world income variable.
v     Riedel (1988a) has argued that the results are substantially different
when, in contrast to the procedure adopted here, price is used as the dependent
variable and export volume is the independent variable. Muscatelli, Srinivasan,
and Vines (1992) show, however, that the normalization (or the choice of the
dependent variable) does not matter once serial correlation and endogeneity are
accounted for.   Riedel's results, therefore, appear to arise from the non-
stationarity of the variables, leading to a "spuriously" strong correlation
between the country's export price and the world price and eliminating all other
partial correlations.



6
Of specific interest is y;, which in our pooled cross-section time-
series setting summarizes, for each country i, the effect of country features
that we do not observe. These unobserved country features include variables that
are difficult to observe,  such as the strength of international marketing
relationships between suppliers and their international buyers, or can in
principle be observed but can be measured only imperfectly, such as the quality
of a country's infrastructure. These features are a potential cause of export
growth persistence, and hence we begin our description of the data by treating
y as our measure of persistence. As constructed, y1 remains unchanged over time;
however, by considering overlapping slices of time, we are also able to follow
the evolution of yi.
The questions of interest are: first, are the yi's different from
zero, i.e., is there persistence in export growth rates; and, second, are the
yj's different from each other, or do the influences causing persistence vary by
country?
If the yi1s are different from zero and from each other, then the
unobserved differences across countries apparently have a significant influence
on export growth. Benhabib and Jovanovic (1991) note that persistence in per
capita income growth rates across countries could either reflect country-specific
features or all countries could be influenced by the same stochastic forces but
the specific realization for different countries could vary and cause long-term
differences.  They favor the latter interpretation for its parsimony.  While
empathizing with this view, we choose to focus on the specific country correlates



7
of persistence rat-her than attempting to identify the common stochastic structure
of knowledge and institutional evolution.
It is common in such estimations to allow for lags in the response
of exports to the variables influencing them. We allow throughout for lags in
response to price changes. Hence the previous year's price (with the subscript
"t-1") is included as an explanatory variaile.  Landesmann and Snell (1989! and
Krugman and Baldwin  (1987) s"ow empirically that lags in the world income
variable do not have much explanatory power. Landesmann and Snell argue that
this is to be expected since changes in relative prices require shifting to new
buyers and hence imply d lag, whereas changes in world income do not require such
shifts and hence lags are not likely. Our estimations confirmed this result and
we do not report them here.
A.    TestinQ the Specification
The export demand function is estimated for a set of 20 developing
countries, from 1972 to 1985. Results for 13 developed countries over the same
period are used wherever relevant. The data for developing -;ountries are pooled
when estimating the demand function. The advantage in pooling the data is that
we have sufficient degrees of freedom to estimate the coefficients with some
accuracy; the disadvantage is that the price and income elasticities are assumed
to be equal across the countries. We show in the next section that the basic
finding of persistent country-specific fixed effects remains unaltered even when
price and income elasticities are allowed to vary across the countries in t}._



8
sample.
To allow for the possibility of simultaneous determination of export
volume and relative prices, we use the two-stage least squares procedure (2SLS),
where the instruments used for the endcgenous relative price variable are: lagged
exports (AEt1,t2), lagged relative export prices  (A(PV/P"') t.IJ2), current world
prices (APWt), current and lagged wages (A'Wt,l), current and lagged world income
(AY.1W,t_2), and lagged imports of capital goods (AN-I.t-2) -- all variables are
expressed in logs.
In the first equation of table 1, yi are free to take on any value,
allowing for the possibility that yi differ by country. This is our most general
model. The hypothesis that the yi are equal to zero is rejected very strongly
(the p-value for the X2-statistic is 0.005).  The hypothesis that all yl are
equal, though not necessarily zero, is also rejected with a p-value of 0.016.
Thus the evidence on country-specific fixed effects terms, and hence
on the persistence of growth rates in specific countries, is strong. Differences
in fixed effects between countries are also evident; as we shall see below, two
groups of countries have very different fixed effects and also face different
buyer behavior.
The second column in table 1 shows that the 20 developing countries
as a group experienced a statistically significant persistent growth of 4.4
f     In the presence of heteroskedasticity the least squares standard errors are
biased. In our estimations we use a consistent estimator of the covariance
matrix (White 1980). We use the Wald test for the joint significance of the
coefficients, the statistic for which has a Chi-square distribution.



9
percent a year, over and above that explained by relative price and world incone
changes. It is sobering to reflect that during this period the average rate of
growth of manufactured exports from these countries was 12 percent; thus about
one-third of export g.owth depended upon factors not conventionally accounted
for. When the second and third columns i5 table 1 are compared, a--counting for
persistent growth rates substantially lowers the income elasticity of demand from
4.1 to 3.1.
For individual countries, such as Turkey, Indonesia, and Republic of
Korea, the proportion of growth explained by the persistent country-effect was
much larger than the average of one-third for all countries (see column 4 in
table 2). At the other extreme, a few countries that had negative underlying
persistent growth rates could have doubled their export growth if they had lost
their handicap. Another way to assess the importance of country-specific effects
is to subtract them from the actual growth rate to arrive at the growth rate that
would have occurred if the underlying persistent country-specific export growth
rate  had  been  zero  (column  3).    Though  the  actual  growth  rates  vary
substantially, the growth rates net of country-specific fixed effects are much
closer to each other. In other words, if the countries with high fixed effects
terms did not have their unobserved advantages, and the countries with low fixed
effects terms did not have their unobserved disadvantages, the export growth
rates of different countries would have been fairly closel The implication is
that the degree of relative price changes or the choice of specific export
destinations had a much smaller bearing on export performance than the factors



10
that caused the persistent growth rates.
A similar experiment with the developed countries yielded interesting
results. The residual growth rates for these countries are relatively small,
generally between -2 and 3.6 percent, and these could not be considered
statistically different from zero.y As expected, Japan has a positive residual
growth rate, but it is small (less than 3 percent).  Somewhat surprisingly,
Germany has a small negative residual growth rate. As we discuss below, the
relatively small size of the residual growth rate for developed countries
suggests that the impact of non-price and non-income factors on the demand for
exports tends to diminish as the country secures its position as an insider in
international markets.
B.    Country Groups
One potential problem with our estimates of y1 is that the price and
income elasticities have been constrained to be equal across countries. Hence,
if the countries with rapidly growing exports had larger elasticities, averaging
across countries would result in high positive estimated fixed effect
coefficients; similarly, it would not be surprising if countries with negative
fixed effect coefficients are losing market shares as a result of below-average
income and price elasticities.
Ideally, the export demand function should be estimated for each
The residual growth rates for developed countries are not reported in a
separate table, as they are not the main focus of the paper.



11
country. However, the limited degrees of freedom make the estimates imprecise
as well as unstable.   To overcome this limitation we adopt two strategies.
First, we allow the price elasticities for specific countries to differ from the
average -- for example, we allow the price elasticity of Turkey and Indonesia,
which have the highest fixed effects estimates, to differ from that of other
countries in Group I (by creating dummy variables for each of these countries and
interacting the dummy with the relative price).   The results show that income
and price elasticities for these countries are not statistically different from
the elasticities for other developirng countries (with a p-value of 0.76), and
that large fixed effects remain.
The second strategy was to split the countries into two groups --
thopq with positive fixed effects (Group I) and those with zero or negative fixed
effects  (Group II).   Using the Chow specification test, we tested whether
splitting the countries in this fashion is supported by the data.  The data
strongly reject the restrictions imposed by pooling all developing countries in
the sample, with an F-statistic of 6.88 and a p-value of 0.02 percent, thus
supporting the split. Certain countries on the margin were not easy to classify;
however, the exact composition of the two groups did not alter the results. To
be precise, the general observations from the regression results remain
unaltered; the interpretation of the performance of the specific countries on the
margin, however, does change.
The following differences between the two groups emerge. First, the
average fixed effect for Group I is 9.4 percent and significantly different from



12
zero with a t-value of 4.69, whereas the average fixed effect for Group II is
-1.0 percent and statistically insignificant with a t-value of -0.38. The price
elasticities for the two groups (the sum of the current and lagged values) are
quite similar.   The point estimate of the income elasticity for Group II is
actually higher than that for Group I. Though the difference between the income
elasticities of the two groups is not statistically significant, the larger point
estimate for Group II is unexpected and we return to this issue below.
Table 3 shows that country-specific fixed effects are significantly
different from each other in Group I, the p-value of the X2-statistic is 0.006.
As can be expected, the fixed effects terms obtained for individual countries are
now different from the ones obtained from pooling all developing countries.
However, the orders of magnitude and relative rankings are verv similar (compare
the first and the last columns in table 2).  The persistent growth rates of
Indonesia and Turkey are 19 and 17 percent, respectively, whereas Korea's is 12
percent. When equation 1 is estimated for Group I countries, Portugal's fixed
effects term becomes positive (in contrast to the result when it is estimated for
all developing countries, see Table 2).
Group II countries have negligible country-specific fixed effects
when considered as a group, though some have individually negative residual
growth rates.   India,  for example,  records -3 percent.   Venezuela has a
relatively high 3.3 percent; however, the time pattern of the growth rate is
erratic, in part because its exports tend to be dominated by petroleum-related
products. Accordingly we keep it in Group II.



13
Note also from table 3 that the income elasticity of demand is lower
for the products from developed countries than the ones from developing countries
-- this is commonly observed and attributed to the inclusion of high-growth
countries in the developing country sample. However, Table 3 shows that the
difference between developed and developing countries' income elasticities
originates in part from the high elasticity for the Group II countries (3.53),
which have low export growth rates and, typically, negative fixed effects. The
paradox of high income elasticities in Group II countries is discussed in Section
III, leading to a new interpretation of conventionally estimated income
elasticities.
C.    Persistence
To study changes in the country-specific fixed effects over time, we
create seven-year overlapping "windows" in our sample period. The first window
covers 1972-1978, the second: 1973-1979, the third: 1974-1980, and so on. We
thus have eight windows. For each window we estimate the export demand equation.
For each country, therefore, we obtain eight yis. (See Figure 1, where it should
be noted that year refers to the final year of the window.)
Despite fluctuations, most countries in Group II (Argentina,
Colombia, Chile, India, Pakistan, Yugoslavia, and Israel) had low fixed effects
throughout the period (Figure la) ; in contrast, most countries in Group I
(South Korea, Singapore, Brazil, Malaysia, Thailand, Philippines, Spain, and
Greece) had relatively high fixed effects (Figure lb).



14
However, both increasing and decreasing trends are also discernible,
indicating that the groups are not closed.   Indonesia,  Turkey,  Portugal,
Venezuela, and Mexico have steadily increased the size of their fixed effects
(Figure ic).  Significant realignments occurred from 1981 to 1983, during a
severe downturn in global economic activity. In these years, some of the East
Asian newly industrializing economies (such as Korea, Singapore, and Taiwan)
experienced rapid wage growth.  Countries that increased their fixed effects
coefficients during those years have continued to increase them. A number of
countries suffered a sharp decline in fixed effects coefficients during that
period and have not recovered: most of these were countries that already had low
fixed effects -- India, Israel, Argentina, Chile, Colombia, and Yugoslavia;
however, Greece and Philippines also suffered.
Further, in the early 1980s, estimated fixed effects coefficients of
countries that performed very well in the 1970s, including South Korea,
Singapore, Brazil, Malaysia and Thailand began to decline (Figure lb). If the
high fixed effects term represents the transition from outsider status to the
ranks of the insiders, a decline in the fixed effect toward zero represents their
maturity as insiders.
We draw three inferences from these observations. First, the growth
of exports due to unobserved factors tends to persist over time within a country.
Second, when underlying cost conditions change, and for instance, Group I
countries become expensive producers, new entrants are likely to gain. Third,
these shifts take place over time but can be accentuated by downturns in the



15
world economy.  During such periods international buyers seek new suppliers.
Firms and countries that are well-positioned in such years stand to make large
gains.
D.    Mismeasurement and Incorrect Instruments
Since the country-specific persistent influences could be merely a
reflection of errors in measuring the relevant price and income variables, it is
necessary to take into account the proposition that if all variables were
correctly measured, the observed persistence would disappear. The implication
would be that price and/or income elasticities are much higher than typically
estimated. A similar argument would hold if the simultaneous determination of
export prices and volumes was not fAlly accounted for. The use of incorrect
instruments for relative price in the export demand function would lower the
(absolute) value of the price elasticity of demand.
A common problem in estimating export demand functions is that the
price variable is not measured correctly. The unit values typically used, as is
the case here, do not account adequately for changes in the composition of
exports. Thus, for countries that shift toward products with higher prices per
unit of product sold (from t-shirts to televisions), the unit value index
understates the price increase (See Alterman 1991, who finds that this has been
the case for U.S. imports from some developing countries).
Other factors will lead to an overstatement of the price change by
the unit value index.   If new products include an increasing fraction of a



16
country's exports, the true price index for the enlarged bundle of goods will be
lower than the conventionally measured index (see Feenstra 1992).   The effect
of the increased bundle of goods is identical to unmeasured quality improvements
or greater "taste" for that country's goods in world markets. Feenstra (1992)
has attempted to construct price indices that reflect the introduction of new
products and the exit of old products in the goods supplied by developing
countries. As noted, Helkie and Hooper (1988) use a more direct (though more
approximate) approach by including in their demand function the capital stock of
the supplying country as a proxy for the ability to supply new products.
In our estimates, the measurement problem is alleviated by the use
of instrumental variables and by data transformation.   In correcting for
simultaneity, we use wage rates and other proxies for production conditions as
instruments. In general, the solution for simultaneity is the same as that for
measurement error, and we have, in principle, corrected for simultaneity. The
question is whether our correction is adequate. Specifically, have we adequately
accounted for influences on the supply of exports? If not, the persistence being
picked up in the demand function could well be the result of ignoring supply
factors rather than a feature of the demand function.
We do not believe that it will be possible to fully resolve the
question of whether the observed persistence derives from supply or demand
factors.   However, additional results suggest that while we have not fully
accounted for all supply influences, efforts at refining the export supply



17
equation are not likely to have much power in eliminating the persistence effects
observed.
To test the sensitivity of our results to the choice of instruments,
we experimented with dropping instruments individually or in groups. The overall
conclusion is that neither the elasticities nor the fixed effects coefficients
change significantly. Only when the world price was dropped from the list of
instruments, the estimated price elasticity of the Group I countries increased,
with no significant effect on the country-specific constant terms.
These results are similar to those obtained by Feenstra (1992). In
a more sophisticated correction of price changes, he finds that the quality-
adjusted price for many developing countries rose more slowly than conventional
estimates suggest.   The correction leads,  in his case,  to a higher price
elasticity of demand and a lower income elasticity, although the changes are
limited in magnitude.
A second approach to dealing with measurement errors is through the
transformation of data. At least since Griliches and Hausman (1986), it has been
known that specific transformations of panel data can be used to minimize the
influence of measurement errors; it is also the case, however, that certain
transformations of the data can exacerbate measurement errors. We measure the
variables as rates of change (first differences of log values), a procedure that
is generally considered to increase the "noise-to-signal" ratio, if the variable
under consideration is serially correlated. However, Griliches and Hausman (p.
100) note that if the measurement error, rather than the variable itself, is



18
serially correlated, then first-differencing helps to reduce the noise and to
increase the signal. In our situation, we can expect the measurement errors to
be highly positively correlated, since quality and compositional changes in
exports are not random effects that vary from year to year, but represent changes
over time.  This is at least partially borne out by the figures in Alterman
(1991), in which unit value indices are compared with "true" price indices: the
errors show significant positive serial correlation. Thus first-differencing is
likely to be an effective method of reducing measurement errors.
Griliches and Hausman (1986) propose a test to determine whether the
presence of measurement errors is a source of bias in parameter estimates. The
test involves a statistical comparison of the GLS (random effects) and the fixed
effects (within) estimators.
Adapting their framework and allowing for the possibility that the
unit values reflect export prices only partially, we write the percentage change
in the export unit value (AIh) as the sum of the percentage change in the
unobserved export price (,Pit) and the measurement error (v1t): AIt = AP,t+vk. When
the export demand equation (equation 1) is estimated with AI;, and AI; t-. rather
than the true price variables, the residual term will incorporate  a vt + a1vi.,
which will be correlated with AIft and AIj,t-l.  If the measurement errors are
negligible and there are no other sources of correlation between the residuals
and the unit values, both models (GLS and fixed effects) will be unbiased and
consistent; more importantly the parameter estimates from the two models (GLS
and fixed effects) will be asymptotically equal. However, when the correlation



19
between the composite residual term and the observed export unit value is
statistically significant, both models will produce biased parameter estimates
and these estimates will not be close to each other. As a result, it is possible
to test for the statistical significance of measurement errors indirectly via a
test for the equivalence of the GLS and the fixed effect estimates using the
Hausman specification test (Griliches and Hausman 1986).
P-values for the Hausman test (which has a x2 distribution under the
null hypothesis) are reported in Table 3 for both the first difference and the
levels estimation. The test results show that the first-differences model is not
plagued by the use of unit'value indices. In contrast, however, we cannot reach
the same conclusion for the levels estimation. Hausman tests for all country
groups have very small p-values, which implies that the levels estimation
produces biased estimates due to measurement errors;  the highest p-value is
obtained for the developed countries. This result is consistent with the fact
that the composition of exports from developed countries is more stablP over time
compared to exports from developing countries, and that the quality of the data
from developed countries is higher.
III. Asvmmetries in Income Elasticity: Insiders and Outsiders
Traditionally, the effect of non-price factors is thought to be
captured by the response of exports to changes in world income (summarized in P,
the income elasticity of demand).   A high income elasticity of demand is



20
considered a measure of superior quality, although as noted, the relationship
does not seem obvious from table 3. The income elasticity of demand for products
from developing countries is much higher than that from developed countries.
Should that be read to imply that developing countries export higher quality
products? Within developing countries, it is much higher for the lagging Group
II countries than for the dynamic Group I countries.
Krugman (1989) suggests that the observed high income elasticity of
demand for Japanese products reflects that country's ability to rapidly increase
the variety of products sold on world markets. Muscatelli, Srinivasan, and Vines
(1992) similarly suggest that Korea, Taiwan, and Hong Kong have benefitted
through expanding the range of products for sale.   However, as noted, our
estimates show that the slow-growing (Group II) countries have a higher income
elasticity than the Group I countries, which is contrary to what would be
expected if income elasticity were a good measure either of product quality or
of expanding product variety.
The paradox is resolved when we consider the possibility that exports
respond asymmetrically to changes in world demand. We test the proposition that
income elasticity is different when world income rises than when it falls in the
following equationYV:
More refined non-linear responses could be tested but the results presented
are striking enough. The years in which world income fell were: 1974, 1975, and
1982.



21
(2)          AlogLie = yi  £0Aog(Pt/P¢) + a1Alog(e ,w,/P'.)
+ V (AlogY7eV + + 1*- [AlogY7eV
where:
AlogY'  if AlogY P0
(AlogYl =
O        otherwise.
AlogYr   if AlogYl<O
(AlogY'V ={(
O        otherwise.
The effects of world income change are sharply asymmetric (table 4). In years
of rising income, the elasticities (P+) for the different groups of countries
are fairly close -- between 1.51 and 1.84.' The response to a decline in world
income, however, is much higher and more heterogenous.
The difference between income elasticities when incomes rise ana fall
is small and statistically insignificant. In the case of developed countries,
the p-value for the x2 test is 0.22, substantial for Group I countries but not
significant at the 5 percent significance level, and large and significant for
Group II countries with a p-value of 0.023.    When world income rise by 1
7/    For Group II countries, P+ are barely statistically different from zero at
the 10 percent level of significance; however, it is not statistically different
from the corresponding elasticity for Group I and developed countries at the S
percent level of significance. The data continue to support the estimation of
demand equations separately for Group I and Group II countries, rather than for
all developing countries. However, compared to the original demand equation the
F-statistic is low at 2.41 with a p-value of 0.049, which implies that the model
with pooled developing countries can be rejected only at the 5 percent
significance level.



22
percent, exports of Group II countries increase by almost 2 percent (0+=1.84);
but when world income falls by 1 percent, exports from these countries decline
by 14 percent (P-=14.0).   Thus, in times of rising incomes, all countries gain
equally in terms of export growth.  When world income declines, however, the
Group II countries are hit the hardest; since they do not enjoy any special
advantage when the upturn occurs, they lose market share over time.
If product characteristics were the main factor in determining
elasticity, one would expect that when incomes fall, expenditure on luxury
products would register the sharpest decline. Thus we would conclude that 1-
should be the higheet for developed countries, instead of which, it is actually
the lowest. Similarly, if the Group II countries are supplying basic products
that account for a low share of the incomes of consumers, their exports would
hardly be affected by a decline in world income.
The ordering of a- can be better explained in terms of the demand for
adherence to delivery standards and long-term buyer-supplier relationships, as
discussed in Section V. when global demand declines, profit margins shrink and
buyers are more sensitive to product reliability and delivery schedules. Thus
they rely to a much greater extent on suppliers with a strong track record than
they would in periods of strong demand and higher profit margins.
Moreover, buyers invest in long-term relationships.   The cost of
breaking a relationship is particularly high when ties are strong. Buyers in
industrial countries, who generate about two-thirds of the world demand for
manufactured exports, have strong relationships with other industrial countries



23
(Hakansson 1987); relationships are weakest in Group II countries, and are easily
broken. The marginal supplier is the first to lose an order. Such asymmetric
effects, which reflect the different costs of breaking relationships, have also
been noted in the context of the insider-outsider theory of labor markets
(Lindbeck and Snower 1988, Chapter 9).
For countries seeking to enhance their presence in international
markets, therefore, it is particularly critical to perform well during periods
of slowing or declining world demand. Growth in exports when world demand is
growing is relatively easy and does not contribute significantly to increased
market share. Maintaining market share when world income is growing slowly or
falling is of much greater value in terms of increasing long-term buyer interest.
These results also show that conventional measures of income
elasticity that do not disentangle the asymmetry discussed can lead to erroneous
interpretations. Landesmann and Snell (1989), for example, find that the income
elasticity of demand for manufactured exports from the United Kingdom has been
rising, which they assume implies greater non-price competitiveness.   It is
curious, however, that during this period (late 1981 and 1982), the U.K. Is share
of exports in world trade fell quite precipitously and declined even further in
the next five years (Landesmann and Snell 1989, figure 3).
Our interpretation is that U.K. exports have performed poorly during
global downturns. The arithmetic average of a low elasticity when world income
expands and very high elasticity when income falls can be quite large. In this
case high income elasticity suggests that the U.K.'s non-price competitiveness



24
fell during the last decade. Our measure of non-price factors, y;, shows a clear
decline in the case of England.
In terms of method, therefore, this paper shows that commonly
estimated measures of income elasticity of demand can be very misleading. If the
asymmetric response of demand to rises and declines in income is not considered,
the high estimated income elasticity will be incorrectly interpreted as Implying
high product quality or an increasing variety of products.
IV.   Telecommunications Penetration: A Proxy for Service Oualit ?
There are no simple physical correlates of a process by which new
buyers are drawn to a supplier.   However, we attempt to demonstrate that
transactional quality (as proxied by the telecommunications infrastructure) is
important. We find that discontinuous shifts in transactional quality are needed
to make the jump from outsider to insider.
The number of telephone lines per capita is used, as a proxy for the
ability to communicate easily with buyers and to respond rapidly to their
requests. Note that this measure is mainly an indicator of the quality of the
transactions environment and not necessarily of the quality of the goods being
sold. To the extent that these are different, the measure is only a partial one.
Table 5 reports the regressions when the telecommunications variable
is added to the demand equation. To examine the possibility of endogeneity, both
the contemporaneous growth of telecommunications and growth in the previous year
are considered.



25
For Group I countries, we find the contemporaneous growth rate of
telecommunications penetration had no effect on export growth but growth in the
previous year had a positive effect on the growth of manufactured exports. The
elasticity of export growth with respect to growth in telecommunications is 0.54,
which is significantly different from zero at the 5 percent significance level.
More importantly, when telecommunications availability is introduced
as an independent explanatory variable, the fixed effects become much weaker.
As before, we use the X2-statistic to test the significance of the persistence.
Recall that when the telecommunications variable is not included as a right-hand
side variable, the X2-test strongly suggests the significance of the constant
term (or the average persistent growth rate) as well as the significance of the
individual fixed effects coefficients. When the telecommunications variable is
introduced directly into the demand equation, however, the p-values for both of
the X2-tests rise to 0.08.   The significance levels, however, are sensitive to
the specification of the demand function. Thus when we allow for asymmetric
income elasticities, the p-value of the X2-test for the significance of country-
specific fixed effects is 0.02.   Thus, while telecommunications growth is
strongly related to export growth in Group I countries, it does not conclusively
account for the country-specific effects.
The result is useful, however, because it points us in a direction
that provides an intuitively acceptable explanation of persistence.    Two
observations are relevant in this regard.   First,  regardless of the exact
specification, with the inclusion of telecommunications as an explanatory



26
variable, the magnitude of the fixed effects coefficients falls for all countries
and in some--notably Korea--the fixed effects term almost disappears.   In
Indonesia and Turkey, the two most dynamic countries in the sample, the fixed
effects decline but remain large. Second, table 6 shows that countries with high
export persistence have high average growth rates of telecommunications networks.
For Group II countries (with low and negative) fixed effects, table
5 shows that the contemporary telecommunications penetration has a positive
effect on export growth and the coefficient is significant at the 10 percent
level. A contemporaneous link could well be the result of more rapid exports
resulting in greater telecommunications investment. If that is the case, the
contemporaneous coefficient will be biased upward. To test for that possibility,
we also examine the relationship between the lagged value of telecommunications
penetration and export growth. Lagged telecommunications growth appears to have
no effect on export growth in this set of countries; the coefficient is negative
and insignificant with a t-statistic of -0.35. The evidence is, therefore, very
suggestive of the possibility that indeed the causality runs from export growth
to telecommunications investment.
To summarize, while the direction of causality in either case remains
uncertain, Group I countries have had high export and telecommunications growth
rates, whereas Group II countries have experienced low growth in both respects.
It is tempting to infer that a certain acceleration in growth, or a "push," in
either of the two variables is required to move from the low-growth to the high-
growth cycle. Although such a push could be successful in situations where the



27
capability of exporters to respond consistently with internationally marketable
goods is high, it could also lead to waste.
Note that the approach adopted in this section is similar to that of
Helkie and Hooper (1988) who take the stock of capital in the economy as a proxy
for the supply of a larger variety of goods. We focus on just telecommunications
because it is likely to be more germane to export activity. Moreover, for the
same reason, we prefer to think of telecommunications variable in the demand
equation as a measure of transactional quality rather than as a measure of
ability to supply an increased variety of goods.
V.    A Descrintive Model of Diffusion and Lona-Term Contracts
If persistent growth rates are not artifacts, what omitted export-
inducing explanatory variables do they reflect? Why are the influences of the
omitted variables not captured in high (or low) income elasticities of demand?
And why do they instead appear as unidentified country-specific residuals?
The country fixed effects term reflects the influence of non-price
factors that are not captured by world income.  As distinct from short-term
changes in demand because of world income changes, the fixed effects term is a
measure of the secular increase in a country's market share. Thus, the existence
of a positive country-specific fixed effect represents a diffusion of demand for
t.e country's products, that occurs when new buyers are attracted or when
existing buyers shift demand from other suppliers.
Product-cycle theories (Vernon 1966 and Krugman 1979) would predict



28
that the demand for specific croods shifts from developed country suppliers to
developing country suppliers who operate with lower wage costs; in our estimates,
however, a shift in the market share of all manufactured exports is implicit.
The shift, therefore, represents more than a simple product-cycle effect;
countries that benefit from this shift in demand are being transformed from
"outsiders" to "insiders" in the world trading system: thus, they sell to a
greater number of buyers and they sell a larger variety of goods.
Buyers "sample" products from different firms in different countries
and make inferences about quality and delivery standards.   In purchasing
manufactured goods (especially time- and fashion-sensitive items, such as
clothes, shoes, bicycles, televisions) the dimensions of quality sought by a
buyer are: adherence to specifications, rimely delivery, and ease of
communications; good communications are essential for quick changes in production
and delivery (see Boatman 1991 and Pashigian 1988).
A few samples of poor quality can break the links between buyer and
supplier (even a single bad sample will do this, if the prior reputation is very
unfavorable). Good samples increase confidence in the supplier. The buyer, of
course, is anxious to choose a source as costs are incurred during the sampling
process. At some point, the buyer will select from those he perceives to be good
suppliers.  Mistakes are possible, however, because sampling is not a perfect
predictor.
Once the decision is made, buying offices will be established, local
agents will be recruited, and relationships with suppliers will be formed. At



29
this stage, the buyer has an incentive to improve the performance of his supplier
and is likely to provide information and support to reinforce the suppliers,
strong points. Not only is there a significant cost to making these commitments,
but it is also costly to abandon them. If the long-term relationships are not
used, their value depreciates rapidly and must be rebuilt. This process creates
the persistence in relationships that we have observed in our field work (Egan
and Mody 1992), and is consistent with the relative stability of the country-
specific fixed effects.F
This line of reasoning is closely related to the Roberts and Tybout
(1992) analysis of sunk costs in exporting. Those firms that have exported are
treated as having incurred the sunk costs, raising the probability of exporting
in the current period and in the future. Our findings suggest that in addition
to exporters incurring sunk costs, buyers (conditional on transactional quality)
also make sunk investments in the establishment of long-term relationships with
their suppliers.   Both buyers and sellers, therefore, have an incentive to
maintain their relationship, contributing to persistence in export growth.
Persistence in trade flows has also been examined in the context of
Y     In the statistical literature,  the process of sampling described is
referred to as the two- (or multi-) armed bandit problem. The analogy
refers to slot machines which have more than one arm, each with a
different pay off. It is optimal in this situation to sample from both
arms for a limited period and then invest all resources on the one that
seems to have a better performance (see Degroot 1970). Cowan (1988) has
extended the multi-arm bandit problem to the situation we consider here:
once a choice is made, further investment in the relationship reinforces
the decision to stick to one's choice.



30
U.S. trade by Krugman and Baldwin (1987), and their interpretation is similar to
ours. They argue that export prices react only with long lags to changes in
exchange rates and that trade volumes react only with long lags to export prices.
They use the "Book of the Month Club" analogy to explain these lags: once buyers
subscribe to a particular club, relative price changes do not cause them to
abandon the club unless these price changes persist.
A similar pattern of persistence is found in labor markets.
"Insiders" in the labor market are experienced workers who are hard to fire
because they are more productive, have legal contracts that are expensive to buy,
and work as a team. "Outsiders" are costly to identify and train. A consequence
of insider power is that wage and employment levels tend to persist. The degree
of persistence is not unlimited and if wage differentials between insiders and
outsiders become too large, a reshuffling can occur (see Lindbeck and Snower
1988).
VI.   Conclusions
Growth rates of manufactured exports from developing countries tend
to be persistent,.although not immutable. In addition to the price and income
elasticities, a country's exports are partially influenced by the quality of its
transactional infrastructure (proxied by telecommunications penetration). More
importantly, we find indirecr evidence of "insiders" and "outsiders" in the world
market for manufactured goods, a condition that tends to perpetuate itself.
An asymmetry in the income elasticity of demand is related to the



31
persistence of export growth. The magnitude of the decline in exports following
a fall in world income is greater than the rise following an increase in world
income. The degree of asymmetry is greatest for marginal exporters (outsiders),
who are hardest hit from a drop in world income; these are also the countries
that face persistently low or negative growth rates. In contrast, developing
countries with persistently high underlying growth rates suffer a milder setback
when world income drops. Thus, countries with high persistent growth rates are
beneficiaries of a secular shift in demand, which reflects their changing status
from marginal to long-term suppliers.
This descriptive account is consistent with the hypothesis that
international buyers invest in long-term relationships to ensure product quality
and reliable delivery, and use these relationships to transfer production and
marketing knowledge to their suppliers. The resulting network of relationships
creates social capital that makes it costly to change buying patterns. Hence the
stability in buying relationships is broken mainly when production conditions
change significantly. The limited number of countries that are able to change
their status from outsiders to insiders experience explosive growth. Once their
insider status is established growth tends to level off. The evidence also shows
that countries that appear to have acquired such social capital have further
benefitted from physical investment in their transactional infrastructure.
Many theoretical models and export development strategies assume that
developing and newly industrializing economies, which are small in relation to
the world economy, can rapidly expand exports if domestic supply constraints are



32
removed. The results here do not support the notion that demand for developing
country exports is infinitely price elastic; neither do they imply elasticity
"pessimism." A price elasticity of about 1 is observed, large enough that policy
efforts to change relative prices should have a significant effect on export
growth.   From a methodological point of view, the paper supports Feenstra's
conclusion that efforts at refining measures of relative price lead to only small
changes in the elasticities.
The paper's main point is that price-related measures, such as
devaluation or export subsidies, will have limited effects unless backed up by
infrastructure that makes the country an efficient and reliable supplier. These
attributes are of great importance for international buyers. The argument for
the overwhelming role of price as a determinant of export competitiveness is
certainly open to question. Riedel (1988b) points to Turkey as an example of the
benefits of trade liberalization and export subsidies, but it is worth noting
that Turkey's exports have also been supported by a massive effort to improve the
predictability of suppliers and expand the country's telecommunications network.
This is not to infer that a sophisticated telecommunications network
is the simple answer to export development. Telecommunications penetration is
not a fully satisfactory proxy for service quality. Moreover, in the case of the
slower-growing, Group II countries, the evidence suggests that exports induce
telecommunications investment rather than the other way around. The implication
for these countries is that mere reliance on telecommunications expansion as an
instrument for expanding exports is likely to result in disappointment. Building



33
long-term export relationships with international buyers also require the
capacity to deliver reliable products on time.
The findings imply that effective microeconomic institutional reforms
that improve the ease with which transactions are accomplished, together with an
enhanced ability to meet the exacting demands of world markets are critical to
export success. Such reforms cannot be achieved at once, and self-correction may
be required to improve performance.



34
REFERENCES
Aw, Bee Yan. 1992. "An Empirical Model of Mark-ups in a Quality Differentiated
Export Market." Journal of International Economics 33: 327-344.
Alterman, William. 1991. "Price Trends in U.S. Trade: New Data, New Insights.",
In Peter Hooper and J. David Richardson. eds. International Economic
Transactions:  Issues in Measurement and Empirical Research.   Chicago:  The
University of Chicago Press.
Benhabib, Jess, and Boyan Jovanovic. 1991. "Externalities and Growth Accounting."
American Economic Review 81: 82-113.
Bhalla, Surjit. 1989. "Indian exports, imports, and exchange rates: a comparative
quantitative analysis." mimeo. World Bank, Washington.
Boatman, Kara T. 1991. Telecommunications and Export Performance: An Empirical
Investigation of Develorincr Countries.   Ph.D Dissertation:  University of
Maryland, College Park, Maryland.
Cowan, Robin. 1988. "Backing the Wrong Horse: Sequential Choice Among
Technologies of Unknown Merit." mimeo. Stanford University.
DeGroot, Morris H. 1970. Optimal Statistical Decisions. McGraw Hill Book Company.
Egan, Mary Lou and Ashoka Mody. 1992. "Buyer-Seller Links in Export Development.
World Development 20 (3): 321-334.
Feenstra, Robert C. 1992. "New Product Varieties and the Measurement of
International Prices." University of California, Davis.
Griliches, Zvi and Jerry A. Hausman. 1986. "Errors in Variables in Panel Data."
Journal of Econometrics 31: 93-118.
Hakansson, Hakan. 1987. Industrial Technolocrv: A Network AoDroach. Croom Helm.
Helkie, William H. and Peter Hooper. 1988. "The U.S. External Deficit in the
1980s: An Empirical Analysis." In R.C. Bryant, G. Holtham, and P. Hooper. eds.
External Deficits and the Dollar: The Pit and the Pendulum. Washington D.C.: The
Brookings Institution.
Krugman, Paul. 1979. "A Model of Innovation, Technology Transfer and the World
Distribution of Income." Journal of Political Economy 87: 253-266.



35
Krugman, Paul. 1989. "Income Elasticities and Real Exchange Rates." European
Economic Review 33: 1031-54.
Krugman, Paul and Richard Baldwin. 1987. "The persistence of the U.S. trade
deficit." Brookings Paners on Economic Activity 1: 1-55.
Landesmann, Michael and Andrew Snell. 1989. "The consequer.ces of Mrs. Thatcher
for U.K. Manufacturing Exports." Economic Journal 99, 1-27.
Lindbeck, Assar and Dennis J. Snower. 1988. The Insider-Outsider Theory of
Emtlovment and UnemDlovment. Cambridge: MIT Press.
Marquez, J. and C. McNeilly. 1988. "Income and Price Elasticities for Exports of
Developing Countries." Review of Economics and Statistics 70: 306-314.
Muscatelli, V.A., T.G. Srinivasan, and D. Vines. 1992. "Demand and supply factors
in the determination of NIE exports: a simultaneous error-correction model for
Hong Kong." Economic Journal 102, 1467-1477.
Pashigian, B. Peter. 1988. "Demand Uncertainty and Sales: A Study of Fashion and
Markdown Pricing." American Ero Cmic Review. 78: 936-953.
Riedel, James. 1988a. "The Demand for LDC Exports of Manufactures: Estimates from
Hong Kong." Economic Journal 98: 138-148.
Riedel, James. 1988b. "Strategy Wars: The State of Debate on Trade and
Industrialization in Developing Countries." Presented at Symposium in Honor of
Jagdish Bhagwati, Erasmus University, Rotterdam.
Roberts, Mark and James Tybout. 1992. "Sunk Costs and the Decision to Export in
Colombia." mimeo. Pennsylvania State University and Georgetown University.
Vernon, Raymond. 1966. "International Investment and International Trade in the
Product Cycle." Ouarterly Journal of Economics 80: 190-207.



36
APP8NDIX: Definition of Variables and Data Sources
P' : Unit value index for manufactured exports,
World Tables (WT), 1980100, US dollar based.
pW: Unit value index for worldwide exports of manufactured goods,
GATT: International Trade, 1987-1988, 1980=100.
W  : Nominal wage rate, ( - RW * GDP deflator ),
RW : real earnings per employee in manufacturing, WT, 1980=100;
GDP deflator : WT, 1980=100, local prices.
e  : Nominal exchange rate, index, WT, conversion factor(annual average).
B  : Exports of manufactured goods, WT, constant 1980 US dollars.
Yw : World GDP, defined as
94
r siyi
where YJ is the GDP of the ith partner of the country i, in constant
1980 US dollars, and sJi is the average share of partner j in the total
exports of i throughout the period:
and E'i : Manufactured goods exports from ith country to its jth
partner, constant 1980 US dollars. Manufacturing sector covers all
products from SITC 5 to SITC 8 minus SITC 68. Source: United Nations,
Comtrade Database.
K  : Capital good imports, constant 1980 US dollars.
Data from Comtrade database ( = 71 + 72, SITC ) in current prices, US
dollars, are deflated by the price index for worldwide capital good imports
from UN Statistical Bulletin, 1970-87.
T     Telephone stations (sets) of all kinds per 100 inhabitants.
Data series 9.1, from The Yearbook of Common Carrier Telecommunications
Statistics, various issues.



37
Table 1: Export Demand Fur.ctions for Developing Countries
Dependent Variable:  Rate of Growth of Export Volume (&log E,)
Country-specific        Yi = zj            Fi - 0
7'Y
Alog (P/PW)t              - 0.95             - 1.01             - 1.08
(- 3.62)           (- 3.46)           (- 3.66)
slog (px/pw)11            - 0.20             - 0.21             - 0.12
(- 1.42)           (- 1.58)          (- 0.87)
Alog ywt                   3.10               3.10               4.14
(6.81.)           (6.38)             (12.45)
Constant (7)                                 20.044                _
(2.67)
No. of observations         253                 253               253
D-W statistic               2.06               1.85               1.79
R-bar squared               0.22               0.18               0.16
p-value (X21)                                  0.016                _
p-value (X 2)                                                    0.005
Notes:  1. p-value (X2i) - marginal significance level for HO: 7y = yj.
2. p-value (X22) - marginal significance level for Hb: yi   yj =0.
3. t-statistics in parenthesis, based on standard errors robust
to general cross-section and time-series heteroskedasticity
(White 1980).
4. i and j refer to countries.
5. Instruments include:  AE1.1t.2, AYw,tl.2 APw   A (pX/Pw)t - .2'
AW, t-.1, A*Ktl, t.2- All variables are expressed in logs.



38
Table 2: Quantitative Importance of Country Specific Fixed Effects
Country-specific     Actual          Growth rate      Relative importance    Country-specific
growth rate      growth rate     without country-    of country-specific     growth rate
(pooled)                        specific effects         effect          (separate groups)
(2)                               (4) = (1)/(2)            (5)
(1)                  ~~~~~~~~(3) =(2) - (1)
Brazil                  6.6               16.5              9.9                 0.40                 9.2
Greece                  2.1              10.4               8.3                 0.20                 4.2
Indonesia               16.9             26.3              9.4                  0.64                 19.2
South Korea             8.9              17.0              8.1                  0.52                 10.9
Malaysia                10.4              15.4             5.0                  0.68                 11.8
PhDippines              14.3             18.7              4.4                  0.76                 16.1
Porugal                 -0.6              7.3              7.9                 -0.08                 1.5
Singapore               6.2              12.6              6.4                  0.49                 7.8
Spain                   1.4               9.9              8.5                  0.15                 3.7
Thailand                6.7               13.7             7.0                  0.49                 8.5
Turkey                  16.2             21.7               5.5                 0.75                 17.6
Argenine               -0.5               2.8              3.3                 -0.18                 -1.7
Chile                   -0.1              7.5              7.6                 -0.02                 -1.1
Colombia               -0.25              3.8               6.3                -0.67                 -3.9
India                   -2.3              3.7              6.0                 -0.61                 -3.6
Israel                  3.3              11.7              8.4                  0.28                 2.6
Mexico                  -3.2              3.1              6.3                 -1.04                 -4.5
Pakistan                1.3               8.7              7.4                 0.15                  0.2
Venmzuela               3.6               14.4             10.8                 0.25                 2.7
Yugoslavia              0.6               5.6               5.0                 0.11                 -0.7
Note: Country-specific growth rates (column 1) are the corresponding fixed-
effect terms from the pooled estimation of equation 1 for 20 developing
countries in our sample. Country-specific growth rates in Column 5 are the
fixed effects terms from separate estimations for Group I and II countries.



39
Table 3: Persistence Effects in Different Country Groups
Dependent Variable:  Rate of Growth of Export Volume (Alog Et)
Developing Countries             Developed
Group I      Group II        All        Countries
Alog (px/pw)              - 0.57        - 0.93       _ 0.95         - 0.28
(- 1.83)     (- 2.76)      (- 3.62)       (-3.38)
Alog (PK/PW)11            - 0.29        - 0.11       - 0.20         - 0.16
( 2.32)       (- 0.5)      (- 1.42)       (1.79)
Alog Ywt                   2.29          3.53         3.10           1.78
(3.91)       (4.52)        (6.81)        (8.67)
No. of observations          141          112           253           169
D-W statistic               2.02          2.14         2.06           2.15
R-bar squared               0.18          0.24         0.22           0.38
Hausman - difference        0.71          0.26         0.91           0.94
Hausman - levels           2x10-5         0.0          0.001         0.022
p-value (X21)              0.006          0.99         0.016          0.22
p-value (X22)              7x104          0.99         0.005          0.27
Notes:  1. p-value (x 21) * marginal significance level for Hi: y =  j
2. p-value (X22) = marginal significance level for HO: yl - Tj =0.
3. t-statistics in parenthesis, based on standard errors robust
to general cross-section and time-series heteroskedasticity
(White 1980).
4. i and j refer to countries.
5. Instruments include: AE_t.2, AYWt.ttlt2, APwt,  A (P[/PW) t-1. t-2'
AWtt1,  AKt1. t-2*  All variables are expressed in logs.



40
Table 4: Asymmetries in Income Elasticity
Dependent Variable: Rate of Growth of Export Volume (&log Et)
Developing Countries          Developed
Group I    Group II        All      Countries
Alog (px/pw) t           0.71      - 0.88       - 1.02      - 0.28
(- 2.23)    (- 2.69)    (- 3.93)    (- 3.31)
Alog (p/pw) t-         - 0.29      - 0.10       - 0.20      - 0.14
(- 2.36)    (- 0.48)    (- 1.52)      (1.55)
[Alog yw1]+            l.Sl         1.84        1.64         1.55
(2.05)      (1.78)       (2.65)      (5.14)
[Alog Ywtl-            8.69        14.00        3.2.68       3.65
(2.43)      (2.97)       (_.97)      (2.45)
No. of observations      141          112         253          169
D-W statistic            2.06        2.18         2.12        i.lO
R-bar squared            0.18        0.27         0.24        0.38
p-value (X2;)           0.004        0.96        0.007        0.17
p-value (X22)           0.0001       0.90        0.0001       0.19
p-value (X 3)           0.069        0.023       0.002        0.22
Notes:  1. p-value (X21) = marginal significance level for HO: 'yi = yj.
2. p-value (X22) = marginal significance level for HO: yi =   =0°
3. p-value (X23) = marginal significance level for go : B+ -.
4. t-statistics in parenthesis, based on standard errors robust
to general cross-section and time-series heteroskedasticity
(White 1980).
5. i and j refer to countries.
6. Instruments include: AEt_.t2,   t AYwtt   tA    (1 / PW)t_. t.2,
AWt. t-l   AKt-1. t-2 -All variables are expressed in logs.



41
Table 5: Telecommunications as a Proxy for Service Quality
Dependent Variable:  Rate of Growth of Export Volume (Alog Et)
Group I                    Group II
Alog (PX/pw)1 t          0.67        - 0.70          0 0.90      - 0.87
(- 2.09)     (- 2.16)      (- 2.52)      (- 2.27)
Alog (PVP/w)tI         - 0.30        - 0.35        - 0.15        - 0.10
(- 2.37)     (- 2.97)      (- 0.57)      C- 0.4)
Alog YWt                2.32          2.43          3.51          3.27
(3.93)       (4.14)        (4.15)        (3.85)
Alog Tt                 0.13           -            0.85           _
(0.52)                     (1.63)
Alog Tt.i                 -           0.54           -             0.19
(2.13)                    (- 0.35)
No. of observations       138          137           100           103
D-W statistic            2.05          2.00          1.97          2.07
R-bar squared            0.17          0.21          0.28          0.23
p-value (%21)            0.062        0.077           .99          .99
p-value (X27)            0.035        0.083           .86           .99
Notes:  1. p-value (X'I) - marginal significance level for HO: y   yj.
2. p-value (X22) - marginal significance level for HO:  '=j yj =0.
3. t-statistics in parenthesis, based on standard errors robust
to general cross-section and time-series heteroskedasticity
(White 1980).
4. i and j refer to countries.
5. Instruments include: AEt,.It2 AYw t-t.,t2 APwt, A(PX/PW)t-l.t-2
AWt, t- AI S4, t-2; and ATt-I or ATt, depending on the equation
estimated. All variables are expressed in logs.



42
Table 6: Telecommunications Penetration Rates
Average          1971           1985           Average     Group
1971-85                                      growth rate
Korea                                 9.67            2.10           25.50           0.15        I
Singapore                             24.65           6.80           44.24           0.11        I
Turkey                                4.21            1.62            9.07           0.10        I
Malaysia                              4.66            1.61           9.08            0.09
Yugoslavia                            8.10            3.59            NA             0.07        n
Thalnd                                0.96            0.43            NA             0.09        I
Brazil                                5.28            2.09            9.30           0.09        1
Mexico                                6.20            3.12            NA             0.08        U
Greece                                26.96           11.97          41.34           0.07        1
Indonesia                             0.33            0.17           0.52            0.07        I
Spain                                 27.52           13.46          39.59           0.06        I
Israel                                30.02           17.40          46.89           0.06        n
India                                 0.37            0.22           0.57            0.06        U
Philippines                           1.20            0.65            NA             0.06
Pakistan                              0.42            0.79           0.68            0.05        U
Portugal                              13.65           8.50           20.48           0.05        I
Colombia                              5.85            3.77            8.04           0.04        n
Argentina                             9.10            6.81           11.60           0.03        n
Chile                                 5.03            4.01            6.51           0.03        n
Venezuela                             6.13            3.70            8.19           0.02        n
All                                   9.51            4.64            17.6           0.159
Note:   Number of  telephone  sets of all kinds per  100  inhabitants.



Figure 1: Residual Rate of Export Growth (1973-1985)
(a) Low
0.35
0.3
0.25
0.2-
2 0.1
s                   \             ~~~~ ~       ~ ~~~~~~~India   tw
0.05
0
-0.05'
4.0.
0  1970    1980         1981        1962        1983         184         1WS
(b) High
0.35
0.3
0.25,
0.2
8  0.15
20.1  
3 0.05-
O
.0.05
-0.1
1979       1980         1981        1982        1983        19t4        1985
(c) Increasing
0.35
0.3_
0.05
0.2-
el
80.1
-0.05.
1979        1980        1981        1982        1983         1984        1988



Policy Research Working Paper Serles
Contact
Title                           Author                   Date              for paper
WPS1249 Competitiveness and Environmental Piritta Sorsa           February 1994      P. Kokila
Standards: Some Exploratory Results                                        33716
WPS1250 Explaining Miracles: Growth       William Easterly        February 1994      R. Martin
Regressions Meet the Gang of Four                                          39026
WPS1251 Excise Taxes                      John F. Due             February 1994      C. Jones
37699
WPS1252 On the Dangers of Decentralization   Remy Prud'homme      February 1994      TWUTD
31005
WPS1253 Can Competition Policy Control 301?  J. Michael Finger    February 1994      M. Patena
K. C. Fung                                37947
WPS1254 What Are OECD Trade Preferences   Alexander J. Yeats      February 1994      J. Jacobson
Worth to Sub-Saharan Africa?                                               33710
WPS1255 Intrahousehold Resource Allocation: Lawrence Haddad       February 1994      P.Cook
An Overview                     John Hoddinott                             33902
Harold Alderman
WPS1256 World Fossil Fuel Subsidies and   Bjom Larsen             February 1994      C. Jones
Global Caibon Emissions in a Model                                         37699
with Interfuel Substitution
WPS1257 Old-Age Security in Transitional  Louise Fox              February 1994      E. Vincent
Economies                                                                  82350
WPS1258 Decentralizing Infrastructure:    Richard Bird            February 1994      WDR
For Good or for III?                                                       31393
WPS1259 The Reform of Fscal Systems in    Robin Boadway           February 1994      C. Jones
Developing and Emerging Market    Sandra Roberts                           37754
Economies: A Federalism Perspective Anwar Shah
WPS1260 When Is a Life Too Costly to Save?   George L. Van Houtven    February 1994  A. Maranon
Evidence from U.S. Environmental   Maureen L Cropper                       39074
Regulations
WPS1261 A Political-Economy Analysis ot   Arvind Panagariya       March 1994         N. Artis
Free Trade Areas and Customs    Ronald Findlay                             37947
Unions
WPS1262 Flexibility in Sri Lanka!s Labor Market Martin Rama       March 1994         P. Cook
33902
WPS1263 The Effects of Barriers on Equity    Stijn Claessens      March 1994         F. Hatab
Investment in Developing Countries  Moon-Whoan Rhee                        35835



Policy Research Working Paper Series
Contact
Title                            Author                  Date               for paper
WPS1264 A Rock and a Hard Place: The Two    J. Michael Finger      March 1994         M. Patenia
Faces of U.S. Trade Policy Toward                                           37947
Korea
WPS1265 Parallel Exchange Rates in        Miguel A. Kiguel         March 1994         R. Luz
Developing Countries: Lessons from  Stephen A. O'Connell                    34303
Eight Case Studies
WPS1266 An Efficient Frontier for International Sudhakar Satyanarayan   March 1994    D. Gustafson
Portfolios with Commodity Assets    Panos Varangis                          33732
WPS1267 The Tax Base in Transition: The Case Zeljko Bogetic        March 1994         F. Smith
of Bulgaria                      Arye L. Hillman                            36072
WPS1268 The Reform of Mechanisms for      Eliana La Ferrara        March 1994         N. Artis
Foreign Exchange Allocation: Theory Gabriel Castillo                        38010
and Lessons from Sub-Saharan     John Nash
Africa
WPS1269 Union-Nonunion Wage Differentials   Alexis Panagides       March 1994        I. Conachy
in the Developing World: A Case  Harry Anthony Patrinos                     33669
Study of Mexico
WPS1270 How Land-Based Targeting Affects   Martin Ravallion        March 1994         P. Cook
Rural Poverty                    Binayak Sen                                33902
WPS1271 Measuring the Effect of External   F. Desmond McCarthy     March 1994         M. Divino
Shocks and the Policy Response to  J. Peter Neary                           33739
Them: Empirical Methodology Applied Giovanni Zanalda
to the Philippines
WPS1272 The Value of Superfund Cleanups:   Shreekant Gupta         March 1994         A. Maranon
Evidence from U.S. Environmental   George Van Houtven                       39074
Protection Agency Decisions      Maureen L Cropper
WPS1273 Desired Fertility and the Impact of    Lant H. Pritchett   March 1994         P. Cook
Population Policies              Lawrence H. Summers                        33902
WPS1274 The New Trade Theory and Its      Asad Alam                March 1994         A. Alam
Relevance for Developing Countries                                          87380
WPS1275 Female-Headed Households,         Ricardo Barros           March 1994         K. Binkley
Poverty, and the Welfare of Children  Louise Fox                            81143
in Urban Brazil
WPS1276 Is There Persistence in the Growth   Ashoka Mody           March 1994         M. Pateffa
of Manufactured Exports? Evidence  Kamil Yilmaz                             37947
from Newly Industrializing Countries