\W4PS 2 3c 0
POLICY RESEARCH WORKING PAPER   23P 9                                                 2
The Social Rate of Return                                             Large potential benefits of
infrastructure may elude
on  Infrastructure                                                   identification and
Investments                                                           measurement by
conventional cost-benefit
analysis. Canning and
David Canning                                                         Bennathan estimate social
Esra Bennathan                                                        rates of return to paved roads
and electricity-generating
capacity. relative to the return
on general capital, by
examining the effect on
aggregate output and
comparing that effect with
the costs of construction.
The World Bank
Development Research Group
Public Economics
and
Private Sector Development and Infrastructure
Infrastructure Group
July 2000



POLICY RESEARCH WORKING PAPER 2390
Summary findings
Canning and Bennathan estimate social rates of return to     par with, or lower than, rates of return on other forms of
electricity-generating capacity and paved roads, relative    capital. But in a few countries there is evidence of acute
to the return on general capital, by examining the            shortages of electricity-generating capacity and paved
effect on aggregate output and comparing that effect         roads and, therefore, excess returns to infrastructure
with the costs of construction.                              investment.
They find that bot]h types of infrastructure capital are      Excess returns are evidence of suboptimal investment
highly complementary with other physical capital and    that, in the case of paved roads, appears to follow a
human capital, but have rapidly diminishing returns          period of sustained economic growth during which road-
if increased in isolation. The complementarities on          building stocks have lagged behind investments in other
the one hand, and diminishing returns on the other,          types of capital. This effect is accentuated by the fact that
point to the existence of an optimal mix of capital           the relative costs of road construction are lower in
inputs, making it very easy for a country to have too         middle-income countries than in poorer and richer
much -  or too little -  infrastructure. For policy          countries.
purposes, Canning and Bennathan compare the rate                As a rule, a tendency to infrastructure shortages -
of return for investing in infrastructure with the           signaled by higher social rates of return to paved roads
estimated rate of return to capital.                         or electricity-generating capacity than to other forms of
The strong compleimentarity between physical and           capital -  is symptomatic of certain income classes of
human capital, and the lower prices of investment goods       developing countries: electricity capacity in the poorest,
in industrial economies, means that the rate of return to    paved roads in the middle-income group. To the
capital as a whole is just as high in rich countries as in   extent that such high rates of return are not detected
the poorest countries but is highest in the middle-income    by microeconomic cost-benefit analysis, they suggest
(per capita) countries,.                                     macroeconomic externalities associated with
In most countries the rates of return to both              infrastructure.
electricity-generating capacity and paved roads are on a
This paper-a product of Public Economics, Development Research Group, and the Infrastructure Group, Private Sector
Development and Infrastructure-is part of a larger effort in the Bank to study the impact of public expenditures and the
growth effects of physical infrastructure. The study was funded by the Bank's Research Support Budget under the research
project "Infrastructure and Growth: A Multi-Country Panel Study" (RPO 680-89). Copies of this paper are available free
from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Hedy Sladovich, room MC2-609,
telephone 202-473-76;98, fax 202-522-1154, email address hsladovich@worldbank.org. Policy Research Working Papers
are also posted on the Web at www.worldbank.orglresearch/workingpapers. The authors may be contacted at d.canning
@qub.ac.uk or ebennathan@worldbank.org. July 2000. (48 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 narmes 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 Policy Research Dissemination Center



The Social Rate of Return
on Infrastructure Investments
David Canning
Professor of Economics at the Queen's University at Belfast
Northern Ireland, U.K.
Email: d.canning@qub.ac.uk
and
Esra Bennathan
Professor Emeritus of Political Economy, Bristol University, U.K
E-mail: ebennathan@worldbank.org
This paper has been produced as part of a World Bank research project "Infrastructure and
Growth: A Multicountry Panel Study" (RPO 680-89), sponsored by the Public Econornics
Division of the Development Research Group and by the Transport, Water, and Urban
Development Department.






1. Introduction
The construction of infrastructure has traditionally had a large public sector
component. For some kinds of infrastructure, the argument for public provision is that they
represent non-rival public goods, as is the case of rural roads, or a natural monopoly, as is
the case of electricity distribution systems and land-line telephone networks. Public sector
provision, often in the absence of market pricing mechanisms, has led to projects being
evaluated by the methods of cost benefit analysis, as is the practice of the World Bank in
its infrastructure projects. The average economic rate of return for World Bank projects
evaluated over the period 1983-1992 was 11 percent for electricity projects, and 29 percent
for road building(Wold Bank, 1994). Rates of that order might be described as adequate,
but not exceptional. Where they prevail there is an argument for infrastructure provision,
but no indication of a serious shortage of the infrastructure.
There are, however, a number of well-known problems with rates of return based
on cost benefit analysis. Actual practice of such studies often departs far from the
theoretically correct methodology (Little and Mirrlees (1990)). Even if done correctly,
however, microeconomic cost benefit analysis is likely to miss important benefits of
infrastructure if those occur in the form of extemalities. Transportation infrastructure may
have a profound impact on the extent of the market and the ability of producers to exploit
economies of scale and specialization. Widening the market then brings benefits in terms
of increased competition and contestability in markets. Transportation infrastructure also
allows greater dissemination of knowledge and technology. Models incorporating these
ideas are now common in the "new economic geography," and there is increasing
empirical evidence for these effects, see, for example, Krugman (1991, 1996), Borland and
Yang (1992), Krugman and Venables (1995), Kelly (1997), Porter (1998), Gallup, Sachs
and Mellinger (1999), Limao and Venables (1999).
Other infrastructure, such as electricity generating capacity, should be important in
the type of "big push" models of economic development as proposed by Murphy Shleifer
and Vishny (1989). If the takeoff in developing countries relies on a co-ordinated bout of
investment, the public provision of risky, large scale, infrastructure projects may provide a
trigger for private sector investment and escape from a poverty trap.
2



These arguments point to very large potential benefits of infrastructure which
nevertheless elude identification and measurement by conventional cost-benefit analysis.
Unless measured in a convincing way, however, we do not know whether the size of these
effects provides a case for expanding infrastructure beyond current levels, or even perhaps
for adopting a policy of infrastructure-led development. This remains true even under the
current trend of providing infrastructure though the private sector, or at least to have some
form of pricing mechanism. While public or private pricing schemes can recover at least
in part, the costs of a project, prices can only capture private benefits. If infrastructure has
large positive externalities, even under private provision we may wish to have a policy of
subsidies to ensure provision on an adequate scale.
Our approach to finding the benefits of infrastructure is to estimate an aggregate
production function for a panel of countries over the last 40 years, including as explanatory
variables physical capital and human capital as well as our infrastructure variables, paved
roads and electricity generating capacity. We can then calculate the marginal product of
infrastructure as its contribution to aggregate output. While this approach misses any
benefits to infrastructure that do not appear in Gross Domestic Product (for example, time
savings that lead to increased leisure) it should allow us to see if infrastructure has large
output effects.
Using aggregate production functions to estimate the contribution of infrastructure
has become quite common (for example, Andrews and Swanson (1995), Boarnet (1997),
Carlino and 'Voith (1992), DeFrutos, GarciaDiez and PerezAmaral (1998), GarciaMila,
McGuire and Porter (1996),or Pinnoi (1994)). The main problem with estimating these
function is reverse causality. An increase in income leads to increased demand for
infrastructure, and so a positive correlation between infrastructure stocks and output levels
may be simply due to increased demand, and may not reflect any supply side productivity
effect. To overcome this problem, we use the techniques developed in Canning (1999)
based on a panel data, cointegration, analysis, as outlined in 2.1 below. One appealing
feature of our approach is that the estimates we get for the productivity of human and
physical capital are close to those found in microeconomic studies of their private rates of
return. This suggests that the procedure does indeed remove the bias introduced by reverse
3



causality which we suspect to be just as great for investment in physical and human
capital as for infrastructure.
A major difference between the results in this paper and those in Canning (1999) is
that here we base our approach not on a Cobb-Douglas production function but on a trans-
log specification. The Cobb-Douglas production function imposes a declining marginal
product of each type of capital as the capital-labor ratio rises. This virtually imposes a
finding of a high rate of return to all capital goods in lower-income countries and a low
rate of return in high-income countries, which is greatly at odds with observed private rates
of return on physical and human capital and the pattern of capital flows between countries
(Lucas (1990)). The trans-log specification, on the other hand, allows for flexibility in the
pattern of rates of return across countries.
A further major reason for adopting this specification is to allow us to examine the
pattern of complementarity and substitutability between inputs into the production
function. We find that each type of infrastructure, on its own, has rapidly diminishing
returns, which implies little support for a policy of purely infrastructure led growth.
However, infrastructure is found to be strongly complementary with both physical and
human capital, giving it an important role in a process of balanced growth and the
possibility of acute infrastructure shortages if investment in other types of capital takes off
but infrastructure investment lags behind. We explain these relationships in sections 2.2
and 2.3 below. Together with the cost of infrastructure (section 3 below) these productivity
relationships enter into the determination of the social rates of return to infrastructure
(section 4 below).
For many countries, across the whole range of income levels, we estimate rates of
return to infrastructure that are in line with, or actually lower than, those found for physical
capital as a whole. Given the extra costs caused by the distortions involved in raising taxes
to fund public infrastructure projects, this gives little support for a general policy of
increasing infrastructure stocks.
However, the rate of return to infrastructure is found to be highest in countries with
infrastructure shortages, that is low levels of infrastructure relative to their levels of human
4



and physical capital, and countries that have low costs of infrastructure construction.
Among a subset of middle income countries , we find evidence of an acute shortages of
paved roads, coupled with very low costs of road building. This generates exceptionally
high estimated rates of return to paved road building in these countries. We find similar
evidence of high rates of return to electricity generating capacity, but this time mainly in a
subset of lower and lower-middle income countries.
It should be emphasized that for all higher income countries, and for the vast
majority of lower and middle income countries, we find that the estimated rates of return to
infrastructure are in line with, or below, those for capital as a whole. High rates of return
to infrastructure are the exception rather than the rule, making the case for large scale
investment in infrastructure depend on an analysis of a country's characteristics rather than
a blanket prescription or sector-specific rules or schemas.
2. The Effect of Infrastructure on Aggregate Output
2.1 Theory
We begin by examining the contribution of infrastructure to aggregate production.
The approach used is to argue that there is a common world-wide production function
given by
Yit = ai + bt + f (ki, hi, xi,) + sit              (1)
where y is log output per worker, a is a country specific level of total factor productivity,
and b is a time dummy capturing world-wide changes in total factor productivity while k, h
IOur countxy groups are based on World Bank definitions measured in US Dollar Purchasing Power
Parity terms; Low Incomes (43 of 123 countries) have an upper limit of (1985) $ 1,690 p.c.; Lower Middle
Income (40 countries) range from $1,890 to $4,735 p.c.; Upper Middle Income (16 countries, from
Venezuela to Romania), from $4,904 to $6,764 p.c., and 24 High Income countries (from Saudi Arabia to
Switzerland), from $6,765 to $17,000 p.c.
5



and x represent the log of per worker inputs of physical capital, human capital, and
infrastructure capital respectively. The term - represents a random error.
By defining everything in per worker terms we rule out economies of scale at the
aggregate level that may in fact be important for measuring the effect of infrastructure
(Morrison and Schwartz (1994)). For simplicity, we include infrastructure as a normal
factor of production, ignoring the possible effects of infrastructure on the long growth rate
of technology and total factor productivity that are examined in Duggal, Saltzman and
Klein (1999). We also assume random errors in output around our production function,
rather than allow for a stochastic frontier approach as used by Mullen, Williams and
Moonmaw (1996). The motivation for our straightforward approach to the production
function is that it allows us to use techniques that control for reverse causality. Since it is
reverse causality that is the major issue for the credibility of aggregate production
functions, this seems worthwhile, even if it is at the cost of using a simple functional form.
We allow the production function, f, to take two different forms. Our first
approach is to assume that the underlying production function is Cobb-Douglas, so that, in
logs, we have
f (ki,,hi,xi,) = aki, + fihi + xit.*          (2)
A second approach is to assume a more complex functional form given by
f (kit, hi,, xit) =a,ki, + fl h +ri xi + a2ki2t + 12 hit +r2 i2t +fkh kit hit + ,kxkitxi, + Vr,,h,,x,,. (3)
This variant of the trans-log production function allows for different degrees of
substitutability and complementarity between the different types of capital. However, by
using capital per worker variables we again impose constant returns to scale and are ruling
out the interaction effects between each type of capital and labor that would appear in a
standard trans-log specification. The larger the number of variables to be estimated, the
6



lower will be the precision of our estimates, so that (3) represents a trade-off between a
more general model and the parsimonious specification that one would like to have for
estimation purposes.
A major problem in estimating the production function as set out above is the
potential for reverse causation. If capital investments depend on income (for example,
through a savings function s;) we can write
A.Kit = Si (Yit ) - dKi,              (4)
where K is the capital stock, Y is the total GDP, and d is the depreciation rate. This gives
the steady state relationship
K_ Si (Yit)                           (5
d
This implies a feedback from income to the capital stock, making it difficult to
identify the results of regressions such as (2) or (3) as a production function relationship.
Tlhere is also obvious potential for a feedback from income to a demand for infrastructure.
If we follow a country through time, output will grow as capital accumulation proceeds,
but capital accumulation will follow income, making it very difficult to establish the causal
links in each direction. The positive feedback from higher income to greater capital
accumulation in infrastructure might lead us to expect an over- estimation of the
coefficients in a production function regression.
While this problem of reverse causality usually precludes simple direct estimation
of the production function, there are circumstances under which we can estimate a
relationship such as (1) using simple methods. As shown in Canning (1999), each of the
series that appear in (1) are non-stationary. We can, therefore, think of (1) as a long run
cointegrating relationship. Note however, that in each country, (5) may also be a
cointegrating relationship, holding even when we divide though by the number of workers.
7



It follows that when we estimate the "production function" as a cointegrating relationship
we will in practice estimate a mixture of a production function and an investment
relationship.2
However, in panel data the problem disappears, provided the long run relationship
(1) is homogeneous across countries while the investment relationship (5) differs across
countries. In a panel we can pool data across countries, and while (1) remains a
cointegrating relationship, when we pool the data and estimate a homogeneous form of
equation (5),
Kit  s(Y)                                 (6)
d
we find that the error term, due to actual investment behavior in each county being
different from the world average relationship, is given by
=Si (Y ) -s(Y)                         (7)
d
It follows that the error term in each country is non-stationary, and eventually
becomes very large, because the error produced by using a pooled relationship, rather than
the true country specific relationship, depends on the income level, which is non-
stationary. Even if we have a long run relationship between income and investment for
each country, pooling the data across countries allows us to identify the long run
production function relationship. This argument, of course, depends on the assumption
that our model (1) is correct and holds across countries. It also depends on the relationship
between income and investment being heterogeneous across countries but, as Chari, Kehoe
and McGratten (1996) point out, differences in the security of property rights and tax
policies are likely to produce very different investment rates even for countries at the same
level of income.
2Even if we adopt Johansen's (1991) technique, which allows the estimation of nmltiple cointegrating
vectors, the results depend on an arbitrary normalization and provide a basis for the subspace spanned by the
cointegrating vectors rather than the structural relationships themselves.
8



If we accept this argument, we can estimate equation (1) consistently by ordinary
least squares (OLS). However, OLS has poor small sample properties in this framework
and its reported t-values are not appropriate, even asymptotically. BaneriJee (1999) and
Phillips and Moon (1999) each give an overview of recent techniques for estimating long
run relationships using panel data that overcome these problems. In this paper we follow
Kao and Chiang (1999) who argue that a dynamic OLS estimator that includes leads and
lags of the first differences of the explanatory variables, has good small sample properties
and gives a method (based on the long run variance co-variance matrix of the innovations
and residuals') of estimating consistent t statistics.
The method used by Kao and Chiang (1999) is appropriate when we estimate a
Cobb-Douglas production function relationship, as in equation (2), since all the variables
appear to be I(1)3, and we postulate that the production function is a cointegrating
relationship4. However, estimation of the more complex production function (3) is
somewhat more problematic. The difficulty is that if capital stock and infrastructure
variables are I(1), the higher order squared and cross product terms cannot be I(1).
However, Chang, Park and Phillips (1999) show that estimating non-linear functions of
1(1) variables does not affect the consistency properties of the standard OLS estimator,
though it does affect the speed of convergence of the estimates5. In addition, while we
report adjusted t-statistics in the same way as for the linear case, it is not clear than these
are asymptotically consistent for the non-linear case. Therefore, while we have a
3 I(1) means integrated of order one; that is non-stationary, but stationary when first differenced. The
tests for non-stationarity are reported in Canning (1999). Here we use paved roads rather than paved roads
plus railway lines but this change makes little difference to the time series properties of the series.
4When estimating a single time series relationship with non-stationary variables it is important to test for
cointegration because the time trends in non-stationary variables can lead to a "spurious" regression
suggesting a close relationship, when in fact none exists. However, Phillips and Moon (1999) point out that
this does not occur in panel data, and we can safely estimate long run relationships relationships by OLS,
even without cointegration.
5 The speed of convergence of a parameter estimate to the true parameter value depends on the range of
variation of the explanatory variable, relative to the variance of the error term. Non-stationary variables tend
to have much greater variance than stationary variables, giving much faster convergence of parameter
estimates in cointegrating relationships than in standard regressions (so called "super-consistency"). Our
higher power terms exhibit even an greater range of values than I(1) variables, indicating that their para,meter
estimates will converge to the true value even more quickly.
9



consistent estimate of the parameters of equation (3), and regard these as our "best
estimates" of the long run relationship between inputs and aggregate output, we do not
carry out hypothesis testing of the significance of the estimates.
2.2 Cobb-Douglas Production Function Estimates
Data for output per worker and capital stock per worker are from the Penn World
Tables 5.6 (see Summers and Heston, 1991). For output per worker we use purchasing
power parity GDP per worker (chain index). Our physical capital measure is constructed
using a perpetual inventory method; assuming a capital-output ratio of three in the base
year (usually 1950) we update each year's capital stock by adding investment (from Penn
World Tables 5.6) and subtracting 7% depreciation from the previous year's capital stock.
Since our estimation only starts in 1960 this gives a reasonable period of time for our
capital stock estimates to lose their dependence on the arbitrary initial condition. The
results of this procedure for producing capital stock estimates are remarkably robust to
variations in the initial choice of capital-output ratio and the depreciation rate. Human
capital per worker is measured by the average years of schooling of the workforce, from
Barro and Lee (1993).
The two infrastructure stock variables used are kilowatts of electricity generating
capacity and the length of paved roads, (including urban paved roads), both taken from the
processed data in Canning (1998). These physical measures do not reflect quality
differences in infrastructure across countries and over time. These differences may occur
at the time of construction; roads differ enormously in terms of their capacity (number and
width of lanes) and durability. Electricity generating capacity comes in many forms (e.g.
oil fired, coal fired, nuclear, hydroelectric) with different construction costs and running
costs. In addition, the effectiveness of infrastructure may depend crucially on its quality,
both initially and in terms of maintenance (see Hulten (1997)). In particular, there is
evidence of wide variation in the quality of roads in different countries due to different
climatic conditions, as well as different levels of maintenance and repair. The lack of
comprehensive quality data means we use our simple quantity measures in our estimation;
10



however, it is worth noting that the fixed effect specification we use to capture cross
country differences in total factor productivity tends to net out any cross-country
infrastructure quality differences that are constant over time.
In table 1 we report the results of our estimates using the Cobb-Douglas
production function. All regressions in this paper include country specific intercepts and
world-wide year dummies (which are not reported). The regressions also include the
value, currerLt, as well as one lead and one lag, of the growth rate of each capital input per
worker (the first differences of the capital stock variables). The short-term effects of these
growth rates are estimated separately for each country, to allow for country specific
business cycle multiplier/accelerator effects. Estimating the short -run coefficients
separately for each country uses up a large number of degrees of freedom, but may
improve the small sample properties of the estimators considerably.
The first column reports results for a standard Cobb-Douglas production function
specification including only capital per worker and human capital per worker. Both
coefficients are statistically significant; they can be interpreted as the elasticity of output
with respect to each input. The coefficients found are consistent with what emerged from
the calibration of a Cobb-Douglas model using microeconomic studies on private rates of
return to physical and human capital (Klenow and Rodriguez-Clare (1997)). We take that
as indicating that any externalities to physical and human capital are small on average. On
the other hand, our results contrast with the finding in some macroeconomic studies of
much higher elasticities, particularly for human capital (e.g. Mankiw , Romer and Weil
(1992)). However, these earlier studies may be contaminated by a feedback from income
to savings (or savings rates) which biases their estimates upwards; the similarity between
our macroeconomic estimates and those based on micro-evidence on private returns
suggests that our econometric methods have overcome the feedback problem.
When we add electricity generating capacity (column 2 of table 1)we find a
significant, positive, coefficient. Since electricity generating capacity is already included
in total capital, we have a double counting problem in interpreting regression 2: an increase
in electricity generating capacity will have two effects, increasing the capital stock as well
as the stock of generating capacity. The coefficient on log electricity generating capacity
11



can be thought of as the effect of increasing generating capacity while holding capital stock
constant; that is, it is the effect of diverting resources from other types of capital to
investment in generating capacity. As shown in Canning (1999), a positive coefficient
implies a gain in output from shifting resources to generating capacity, provided that the
reallocation is carried out at world average prices. In general, therefore, a positive
coefficient on generating capacity implies a higher rate of return to generating capacity
than that for other types of physical capital, though this may not hold in countries where
the cost of generating capacity is relatively high compared with that of other types of
capital.
We find a similar result for paved roads, (column 3 of table 1), suggesting that
paved roads have, in general, higher rates of return than other types of capital. These
positive results retain their statistical significance when we add both types of infrastructure
together ( column 4 of table 1).
The result that paved roads and electricity generating capacity have higher returns
than found for capital in general is at odds with the results reported in Canning (1999),
where no evidence of significant excess returns was found. One difference between the
two studies is that here we use paved roads instead of transport routes (which include
railway line length). In addition, we drop Singapore and Hong Kong from the roads
sample. These city states have very high incomes, despite having very low road lengths,
and including them in the data set tends to produce a much lower estimate for the effects of
roads, since they suggest that roads are not required to generate a high income level. We
remove them from our estimation when we include paved roads as an explanatory variable
on the grounds that as city states their unusual geography, in particular their high
population densities, make them unrepresentative of the development process.
However, the main difference between the two studies is that in this paper, when
estimating the effect of paved roads and electricity generating capacity on output, we do
not include telephones as an extra explanatory variable. The difficulty with including
telephones in the regressions is that it has a very large estimated coefficient, and tends to
swamp the effect of the other variables. The large coefficient would still not justify
exclusion if it were a true reflection of the productivity of telephones, but the estimated
12



productivity effects are implausibly large (giving rates of return of over 10,000% per year)
and may well reflect the fact that the number of telephones is to a greater extent more
demand determined than the other types of infrastructure we are considering.6 To avoid
this difficulty we exclude telephones from our inputs in this study.
We could use the result in table 1 to compute the marginal product, and rate of
retum, to infrastructure. However, the Cobb-Douglas production function imposes the
assumption of a constant elasticity of output with respect to each type of input and ignores
the possibility that the elasticity may vary across countries. In table 2 we report the result
of splitting the sample into two equal sub-samples, based on each country's income per
worker in 1975.7 We find that the coefficients on the infrastructure terms in poorer
countries are very small, and statistically insignificant, but that they remain large, and
significant, in richer countries. This implies that infrastructure in the poorer countries
appears to have the same effectiveness in raising output as other types of physical capital,
while having a greater effectiveness than other types of capital in richer countries.
2.3 Trans-Log Production Function Estimates
We can investigate the production function relationship in greater detail by
adopting the more complex trans-log style of production function set out in equation (3).
The results of these trans-log regressions are shown in table 3. In the first column we
report results for capital and human capital on their own. In column 2 we add electricity
generating capacity while in column 3 we add paved roads. In all three regressions we add
6Formlly, the problem may be that the feedback from income level to the number of telephones is fairly
homogeneous across countries, so that this is what our estimation procedure picks up, while the institutional
structures of roads building and installing electricity generating capacity are more varied across countries,
and so do not bias our results.
7 Splitting the sample on the basis of income tends to bias the results slightly because of the correlation
between sample selection and the disturbance terms. However, table 2 is intended for illustration purposes
only and is not used in calculating rates of return.
13



short run adjustment terms including current, lagged, and a lead of each capital stocks
growth rate, again estimated separately for each country.8
In the trans-log specification the important points are the size and sign of the higher
power terms. In our base specification (column 1), the squared term in capital is positive.
The elasticity of output with respect to capital is, therefore, rising, giving capital a greater
effectiveness in countries that already have a great deal of it. On the other hand, the
squared term for human capital is negative, implying rapidly diminishing returns to
investment in human capital. The interaction effect between human capital and physical
capital in column 1 is positive, suggesting that the two are complements, which is
consistent with the complementarity between capital and skilled labor found by Bemdt and
Christensen (1974).
In column 2 of table 3 we add electricity generating capacity, EGC, to the
specification. The squared term in EGC is negative, indicating rapidly diminishing returns
to investment in electricity taken in isolation. However, the interactive terms between
electricity and physical capital, and electricity and human capital, are both positive. This
implies that electricity generating capacity is complementary to, physical and human
capital, with its effectiveness increasing in their presence. Since we measure the various
capital stocks each per worker, the effectiveness of EGC is found to be rising with capital
deepening.
We find the same pattern for paved roads, with the squared term in roads being
negative, but both interaction terms, between roads and the other forms of capital, being
positive. These results, for both kinds of infrastructure, indicate that infrastructure
investments are not sufficient by themselves to induce large changes in output. However,
infrastructure can be a productive investment in economies with high levels of physical
8 As noted above, it is unclear we should put much weight on the estimated t statistics in table 3 because
of the non-linearities in the specification. In addition, it should be noted that the large increase in the R
squared between table 1 and table 3 is an artifact of the fact that in table 3 we include in the R squared the
explanatory power of the country specific fixed effects and the worldwide time trend, while in tables 1 and 2
these effects are removed from the data before estimation.
14



and human capital, and infrastructure itself, in turn, raises the productivity of investment in
those other types of capital.
A clearer picture emerges from calculating the elasticity of output with respect to
each capital input. Since the elasticities vary with the amount of each input, we begin by
doing this for three fictitious countries, one with median inputs of physical capital, and
human capital and infrastructure per worker, one with each input at the lower quartile and
one with each input at the upper quartile. The results, using input measures taken in 1985,
are reported in table 4. Notice that, in general, the actual country with median amount of
physical capital in that year will not be the one with median levels of human capital or
infrastructure. The table therefore does not represent elasticities in actual countries but in
the hypothetical ones that we construct to represent an average, a moderately poor, and a
moderately rich country.
For physical capital we find a consistent pattern of rising elasticities. Based on the
results in column 1 of table 3, we find that the elasticity of output with respect to capital
would be 0.5 for a country at the first quartile in terms of its input levels of human and
physical capital, rising to 0.65 in a country that was at the third quartile. On the other
hand, the elasticity of output with respect to human capital is fairly steady as we change
input levels. I'urning to the regressions that include our infrastructure variables we find
that the elasticity of output with respect to infrastructure seems to be higher in middle
income countries, that have input levels per capita around the world median, than in
countries with higher or lower input levels. This reflects the fact that infrastructure in
middle income countries benefits from the presence of complementary inputs in the form
of physical and human capital, but is not yet extensive enough to have entered the phase of
rapidly diminishing returns (reflected in the negative coefficients on infrastructure squared
in table 3).
In figures 1 through 4 we plot the elasticities of output with respect to each input,
estimated using each country's actual input mix in 1985, and plot the result against its
income per capita (at purchasing power parity, from the Penn World tables) in that year.
Figures 1 and 2 are based on regression 1 in table 3. Figure 1, for the elasticity of output
with respect to physical capital, tells the same story as table 4, with poorer countries
15



having low elasticities while richer countries (with higher levels of input per worker) have
high elasticities. For education, shown in figure 2, there is some evidence of a U shaped
relationship with elasticities higher in poorer and richer countries and lower in middle
income countries. However, the non-linearity in the relationship is not statistically
significant.
Adding together the elasticities of output with respect to these two types of capital
produces a figure that rises with income, and is close to 0.9 in the most developed
countries. This implies that while we have diminishing returns to capital as a whole over
the entire income range, these diminishing return may occur very slowly in developed
countries. If this is so, then the developed world may have self sustaining "endogenous
growth" while developing countries live in a neoclassical paradigm.
Figures 3 and 4 (based on regressions 2 and 3 of table 3 respectively) plot the
estimated elasticity of output with respect to our two kinds of infrastructure, in each
country against that country's income per capita in 1985. In both cases we see an inverted
U shape, with elasticities being higher in middle income countries and somewhat lower in
the poor and rich extremes of the income distribution. It is notable that in figures 3 and 4
we find a relatively large number of countries that have negative elasticities of output with
respect to paved roads or electricity generating capacity. This does not imply that adding to
the stock of these types of infrastructure reduces output; as before, the elasticities refer to
the effect of adding to the stock of infrastructure while holding total capital constant. That
is, we have the effect of diverting spending away from other physical capital and into the
relevant infrastructure; the negative coefficient therefore means that infrastructure
spending is less productive than spending on other types of capital (at world prices).
This heterogeneity in the response of output to increases in infrastructure, holding
the total capital stock constant, agrees with the results found in Canning and Pedroni
(1999) who use a different technique to estimate the sign of this elasticity on a country by
country basis. It is notable that in both figures 3 and 4 the heterogeneity of the estimated
elasticities is higher for the lower income countries than for the higher income countries.
Variations in the elasticity are caused by differences in the relative proportions of physical
capital, human capital and infrastructure capital across countries, so that the greater
16



heterogeneity implies that the mix of capital varies more among the less developed
countries. We see this as a characteristic of countries in the process of development, that
they are in general further from the optimal mix of capital than the richer countries (or just
the mix characteristic of developed countries), though the way in which the mix varies
differs between countries.
The production function estimates allow us to calculate the impact of infrastructure
investment on output, and indeed also the marginal product of infrastructure. To calculate
rates of return, however, we need data on construction costs.
3. The Cost of Infrastructure
3.1 Measuring the Cost of Infrastructure
Cost data on infrastructure investment are relatively scarce. Our two main sources
are electricity generating capacity costs from the World Bank study by Moore and Smith
(1990) and the cost of constructing transportation routes from the United Nations
International Comparison Project (ICP). In addition we compared the cost of constructing
transport routes from the ICP with data from World Bank projects.
There are several difficulties involved in measuring the cost of constructing
infrastructure to go along side our physical measures of infrastructure stocks. One
fundamental problem with comparing data across countries is differences in the type and
quality of the infrastructure being built. For electricity generating capacity the figures give
averages over many different types of capacity that may reflect different combinations of
capital and running costs.
In theory, the ICP data for the cost of construction of transport routes are for a
common basket of goods and so should adjust for quality differences. However in practice
the adjustment may not be complete. Our World Bank project data are superior in that
they measure road building costs (rather than transport routes in general) but they are not
adjusted for road quality. For these projects we count kilometers of road, not lane
kilometers, nor can we distinguish between roads according to the strength of the surface
17



or the width of the lane. A kilometer added in a high-middle-income country is likely to be
of higher quality than a kilometer in a low-income country, which may introduce a
systematic bias into the data. In addition, the coverage of the ICP data set is much broader
than that from World Bank projects and in what follows we rely exclusively on the ICP
data though it is worth noting that the ICP and World Bank data, when we have data on
both, are in broad agreement.
There is the additional problem for paved roads that our cost figures refer
exclusively to construction, without any allowance for the cost of the land. Land costs are
one reason why Hong Kong and Singapore are such outliers in terms of their road stocks:
not only is the productivity of transport systems likely to be different in such a densely
populated environment, but these city states also have notoriously high land costs.
As with cost benefit analysis, the cost of infrastructure construction we use should
be the real resource cost. However, our data are actual costs, including any price
distortions caused by the tax system or import controls. If we were to take the view that
most of the cross country differences in infrastructure projects are due to such distortions,
it would be appropriate to take a common world cost for each type of infrastructure as its
real resource cost. On that view, the elasticity results in table 4, and in figures 1 to 4,
would indicate whether there is an excess return to paved roads over and above that found
for other forms of capital. However, while this type of assumption may be appropriate for
internationally traded goods, infrastructure projects often involve large scale labor inputs in
the country concerned. This makes the real resource cost depend on the productivity of
labor in other employment, which can vary dramatically across countries. In what follows
we use the actual costs as indicative of the real resource cost.
A problem specific to our data on the cost of electricity generating capacity is that
these are measured in US Dollars while our marginal productivities are measured in
constant international (ICP)dollars. The value of the international dollar is normalized so
that the GDP of the United States is the same in either unit. However, in other countries
the two are not equivalent. In poorer countries, where prices (measured at the nominal
exchange rate) tend to be lower than in the U.S., the real purchasing power of a U.S. dollar,
and so the real resource costs of spending on infrastructure, is high. Before carrying out
18



our rate of return calculations we therefore convert our costs from U.S. Dollars to
international dollars by dividing through by the country's 1985 price level (its purchasing
power parity exchange rate divided by its nominal exchange rate) taken from Summers and
Heston (1991).
3.2 Exploring the Data
Table 5 reports our data on costs by country. We take 1985 as base year for
comparisons because this gives us a fairly wide range of data in nearby years that can be
deflated to 1985 values. Column (1) of table 5 gives data on the cost of construction of
transport routes from the 1985 International Comparison Project. These price indices
represent the nominal price of a basket of transport routes deflated by the country's
purchasing power parity price level. The indices have been converted in a dollar cost per
kilometer of paved road by taking a figure of $627,580 for the U.S.A.. While transport
routes are a more general category of infrastructure than paved roads, roads make up a
large component of transport routes and these figures do reflect the price of a common
basket of routes which tends to lessen the problem of measuring costs for different
infrastructure qualities. For a small number of countries we also have data on the cost of
road construction from World Bank projects. These data are roughly in line with those
from the International Comparison Project but more vulnerable to differences in road
quality between countries.
Summers and Heston (1991) show that there is a tendency for capital goods to be
relatively more expensive in developing countries than developed countries. When we plot
the data on the cost of transport routes in figure 5 we see some evidence of a U- shaped
relationship, with costs being high in the poorer developing and in the developed countries,
but substantially lower in middle income countries. Regressing log cost on log income per
capita and the square of log income per capita gives the result
LogCostperkm=   25.9  -3.517logy  +0.226 (logy)2              (8)
(4.66)  (2.59)      (2.76)
N= 53, R2 = 0.26
19



This gives a minimum cost at an annual income level of around $2,300 (in 1985
International Dollars) per capita, which lies in the bottom half of the income range spanned
by the 40 countries classified as lower-middle income in 1985 by the World Bank. The
World Bank data on costs of paved road construction give a similar picture of U shaped
costs. In our calculations of rates of return we use the larger data set based on the cost data
we have calibrated from the ICP costs of route construction.
One reason for this U shaped cost structure would be that middle income countries
have lower labor costs than developed countries, but also more of the skills and industry
required to produce construction materials and equipment than the majority of the low-
income countries. Where road construction and paving depend on importing equipment
and even raw materials, costs in the poorer countries can rise to levels found in
industrialized countries.
Our cost data for electricity generating capacity come from Moore and Smith
(1990). Cost, in US dollars, per kilowatt of electricity generating capacity and the
corresponding extension of transmission and distribution, in 1989 is shown in column (2)
of table 5. The figures are deflated to 1985 values using the GDP deflator and then
converted to International Dollars, using the country's purchasing power parity price level,
before being employed in our rate of return calculations. These figures are reported in
column (3) of table 5.
Looking at the costs in US Dollars, there is clearly an outlier: Senegal has
construction costs in excess of US$13,000 per kilowatt, which is substantially higher than
for any other country. In fact, the next two most expensive countries, also in Africa, are
Niger and Mozambique, with costs that are about half those found in Senegal. However,
when we look at the costs in international dollars we find high real costs of generating
capacity in many developing countries due to their low price levels relative to their
exchange rate. The relative consistency of prices at nominal exchange rates suggests that
developing countries are not, in general, able to exploit their low wage costs to achieve low
costs of installing electricity generating capacity.
20



Once again there is evidence that the cost of installing electricity generating
capacity falls with the level of income. A regression of log cost per kilowatt of capacity on
log income per capita (both in international dollars) gives the result
LogCostperkW.=   11.18 -        0.287 log y                      (9)
(16.4)       (3.58)
N=63 R2 =0.061
t statistics in parenthesis
For both electricity generating capacity and paved roads the difference in
construction costs between the cheapest and most expensive countries is a factor of almost
10, while cost differences on the order of a factor of 3 are not unusual. Cost differentials
are therefore likely to play an important role in determining rates of return to infrastructure
investment.
4. The Rate of Return to Infrastructure
4.1 Rate of Return estimates and Infrastructure Policy
The rationale for our approach is that there may be externalities to infrastructure
projects that are not caught in micro-economic cost benefit studies. The inclusion of these
externalities potentially allows us to capture the total social rate of return to infrastructure.
There are, however, a number of caveats that must be borne in mind when looking at our
results.
Firstly, our approach is to look at the impact of infrastructure on aggregate output
as measured by GDP. This measure of aggregate output has the potential to capture some
of the externalities that microeconomic cost-benefit analysis may miss, it yet has
conceptual drawbacks of its own. For example, cost-benefit analysis can estimate the
travel time saved by a road project and calculate the value of this time. An analysis using
aggregate output will only pick up the time saved if it is devoted to productive uses, time
21



saved that is spent in leisure activities will not be accounted for. In addition, as
Haughwout (1998) points out, an analysis which relies on aggregate output may neglect
relative price effects of infrastructure construction that can have a significant welfare
impact.
A second problem is that our estimate of the effect of infrastructure on output is its
long run steady state effect. In calculating rates of return we assume that this long-run
effect occurs immediately, and lasts for ever, and we depreciate infrastructure stocks at 7
percent a year to allow for the cost of maintaining the infrastructure in the long run. This
creates a difficulty because, when calculating rates of return, the discounting of future
flows means that returns in the early years tend to dominate the calculations. It follows
that, if it takes several years for infrastructure to reach its full potential, we may be
overestimating its rate of return. However, a similar consideration applies to our estimates
of the rate of return to private capital, so that when we compare infrastructure rates of
return to those found on general capital, we might expect both to be overestimated in
similar proportions. While both are probably overestimates, the problem may be worse for
infrastructure where there is considerable evidence that construction may sometimes lead
demand for infrastructure services, either due to its "lumpy" nature, or to over-optimistic
demand projections (World Bank (1994)).
Our macroeconomic estimates of the rate of return to infrastructure also ignore any
"crowding in" effects that it may have on other types of capital. While an increase in
infrastructure raises the return to other forms of capital, and can lead to an increase in
investment, with consequent effects on output and economic growth, these induced
changes in investment may have only a very small impact on welfare. As Baldwin (1992)
points out, if the marginal product of capital is close to the rate of discount, the marginal
benefit and marginal cost of extra investment are roughly the same, implying little or no
gain in welfare from the extra investment.9
9 This is simply an application of the envelope theorem.
22



However, there are two cases in which this negative result does not hold. If, instead
of a small increase in infrastructure, we are analyzing a large change, then marginal
analysis is no longer appropriate, since the new infrastructure may raise the marginal
product of capital substantially above the discount rate. Alternatively, if we have reason to
believe that the marginal product of capital already exceeds the discount rate, owing, for
example, to a tax wedge, induced increases in investment can have large welfare effects.
In our calculation we ignore any "crowding in" effect, implicitly assuming that we are
analyzing relatively small infrastructure projects and that the existing allocation of
resources to other forms of capital is reasonably efficient (though we do in fact present
evidence that in some countries the rate of return to capital is considerably in excess of any
reasonable discount rate).
There are also several caveats about the use of our rate of return estimates for
policy purposes. First of all, for evidence of externalities to infrastructure to emerge, we
have to subtract from our figures the private returns to infrastructure projects. Only if all
the returns to infrastructure captured in cost benefit analysis are private benefits (and none
externalities)would we arrive at a measure of externalities by subtracting those benefits
that are measured in aggregate GDP (private benefits not measured in GDP should be
added to aggregate productivity estimates to find social rates of return).
However, it is not clear that we ought to focus on externalities: when the
government is the main supplier of infrastructure there is no presumption that it will be
setting infrastructure at the optimal level in terms of private benefits. There may be capital
misallocation in infrastructure even without externalities. Instead, we shall focus on the
rate of return to infrastructure relative to that on other forms of capital. Where this ratio
exceeds one, there is a case for arguing that there should be a reallocation of resources to
infrastructure.
Note that this is somewhat different from the normal cost-benefit approach which
looks at the rate of return to a project in relation to a threshold level that is set by the cost
of funds. We find that, in many countries, the rate of return to capital as a whole appears
to be considerably higher than the commonly used threshold levels (or test discount rates).
In this case there is an argument for encouraging investment in general, and in particular,
23



for removing any distortions that are keeping investment rates low. However, if the rate of
return to infrastructure, while high, is lower than that for other capital, the optimal policy is
to encourage investment in capital other than infrastructure. Infrastructure investment in
those circumstances is very much a second best policy, and would depend on an argument
that investments in other types of capital are not feasible for some reason.
4.2 Calculating Rates of Return
In order to estimate the marginal product of infrastructure we must take account of
the fact that it appears twice in our production function, once on its own in the form of X,
but also as a part of aggregate capital, K. Let Z be non-infrastructure capital, then we can
write
Kit = ZitP +XitP-                              (10)
Pk
The aggregate capital stock is the value of total capital (we sum the volume of each
type of capital times its price) divided by the price of capital. To construct these volume
measures we use world prices of investment goods; all prices are expressed relative to
output, which is taken to be the numeraire. For simplicity, we use the approximation
PZ = Pk, taking the price of non-infrastructure capital as equal to the price of capital as a
whole. Given that infrastructure capital is a relatively small component of the total capital
stock (certainly less than 20% of the total in each case), this approximation seems
reasonable.
Using equation (3) it is easy to derive the country and time specific elasticities
ek =a1 +2a2kit +  hhit +  kXit                 (1 1)
e. _= y, + 2y2kit + y'ikkit + tVfhhit           (12)
The elasticity of output with respect to infrastructure that we estimate is actually
the elasticity found when increasing infrastructure but holding aggregate capital (including
infrastructure) constant. It can therefore be interpreted as the result of diverting a unit of
24



physical capital from other purposes to infrastructure. From these elasticities, and the
definition (10), we can calculate the marginal products of a unit of physical and
infrastructure capital - MPK and MPX respectively - as
MPKi, =ek-   MPXit=  MPKit Px + e.-*                 (13)
Kit                pk    xit
Note that the marginal product of infrastructure consists of two terms, the first
representing the effect of infrastructure on aggregate capital and a second, representing the
distinctive infiastructure effect.
These equations for marginal productivity highlight an important feature of using
aggregate data. To find the marginal product per dollar spent on an input, the estimated
elasticity must be multiplied by the ratio of output to the stock of the relevant capital, each
measured in dollar tenns. For capital as a whole this ratio is quite small (typically less
than one third) but for sub-categories of capital this ratio may be large. Multiplying the
estimated elasticities by a large number also multiplies up any errors in estimation.
It follows that this method is unlikely to be good for determining the marginal
product of small components of the capital stock. Paved roads and electricity generating
capacity, valued at replacement cost, each make up around 20% of the capital stock on
average, implying that they should have observable effects on aggregate output. Not so,
however, telephone main lines which make up less than 2 percent of the capital stock by
value. For reasons stated in 2.2 above, we omitted telephone main lines from our analysis.
A further reason for that decision is their low share in the total capital stock. We would,
therefore, expect to see them have only a small effect on aggregate output . And to find
their marginal product, we would have to multiply a badly estimated elasticity by a huge
number (the ratio of output to the value of the telephone stock) .
The marginal products measure the output effect of an extra unit of capital. In the
case of infrastructure this is the marginal product of an extra kilowatt of electricity
generating capacity or an extra kilometer of paved road. To find the rates of return, we
need the information on the cost of a unit of capital, its marginal product, and its rate of
depreciation. We take the price of investment goods from the Penn World Tables (Mark
25



6.5) as the price of capital goods and measure both marginal products and costs in a
common unit, 1985 international dollars.
Formally, we can find the rates of return to infrastructure type x in country i,
given by ri. by solving for the internal rate of return in the formula
E MPX  -dp                                  (14)
The left hand side of this equation is the discounted flow of benefits from a unit of
infrastructure, minus depreciation (or maintenance costs) which occur at a rate d per unit of
infrastructure per year. The right hand side represents the cost of the unit of infrastructure.
Assuming that the marginal product of infrastructure, and the price at which depreciation is
replaced (or maintenance costs), Pix,t are constant over time, and taking a depreciation rate
of 7 percent per annum10, equation (14) simplifies to
MPX  -0.07.                          (15)
Pix
An equivalent simple formula holds for the rate of return to capital as a whole.
In the following two sections we use these equations to estimate the rate of return
to electricity generating capacity and paved roads. It is, nevertheless, worth noting that if
the relative price of capital and infrastructure are the same in every country, equation (15)
simplifies to
'0The simple path of initial expenditure, followed by positive returns to the project, ensures the existence
of a unique internal rate of return for the project. The result is exactly the same if, instead of replacing
depreciation as it occurs, we assume that we let the capital stock decay to zero over time with proportional
reductions in the benefits of the project.
26



rgX=rLk+eX    it.(6
r. --~~ rik + ex  yi,         (16)
In this case the infrastructure has an excess return over and above that found for
capital in general if and only if it has a positive elasticity as given by equation (12). As
pointed out by Prittchett (1996), however, and as is also evident in our data, the relative
prices vary enormously across countries. We should therefore use equation (15) based on
our two stage procedure of estimating the marginal product of a physical unit of
infrastructure and then relating this to its price. We could still use equation (16) and figures
3 and 4 to indicate the pattern of the sign of excess social returns to infrastructure, so long
as we were to believe that our cost data reflect rents and distortions while the real resource
costs of infrastructure relative to other forms of capital are roughly constant across the
world. However, it what follows we shall concentrate on the rates of return using actual
cost data for the construction of infrastructure. Estimates based on this approach are
reported in tables 6 and 7. All data refer to 1985.
4.3 Estimates of the Rate of Return to Electricity Generating Capacity
Table 6 reports the estimated rate of return to electricity generating capacity,
physical capital in general, and the ratio between the two rates of return in all countries for
which we have the necessary data. The elasticity estimates that underlie these calculations
-- that is the elasticities of both electricity generating capacity and capital in general --
come from regression (2) in table 3. There are a wide range of rate of return estimates,
from well in excess of 100% a year ( in 1985, for Bangladesh, Kenya, Bolivia and China,)
to quite low figures ( Brazil and Zimbabwe,) and even a negative rate of return for
Mozambique. Note that a small negative rate of return does not imply that infrastructure
does not benefit output, only that its benefits do not cover the costs of depreciation or
maintenance.
One might simply use these rates of return to indicate whether or not investment in
electricity generating capacity was a good use of funds. However, the real issue being the
allocation of investment between projects, it is more relevant to compare the estimated rate
27



of return to electricity generating cap4city with that of physical capital in general. The rate
of return to capital (again based on regression (2) of table 3) is reported in column (2) of
table 6 while column (3) gives the ratio of the rate of return on electricity generating
capacity to the rate of return on capital in general. This ratio takes on a wide range of
values. In figure 7 we plot the estimated ratio of rates of return against log income (in
purchasing power parity terms) per capita in 1985. There is a clear downward trend in the
relationship; the poorer countries, on average, have much higher rates of return for
electricity generating capacity than for other capital, while the middle income countries
show rates of return to electricity generating capacity that are roughly the same as for
capital in general. Unfortunately, our cost data are all for developing countries so that we
cannot see how the relationships changes as one moves to high-income levels .
Just as the average rate of return is higher in the poorer countries than in middle
income countries, the variation in the rate of return to electricity generating capacity as
seen in figure 7 is also greater in the poorer countries. High rates of return in the poorer
countries are based on low stocks of electricity generating capacity relative to the stocks of
complementary inputs, that is, physical and human capital. A line is drawn across figure 7
at a ratio of one; at this point the returns to infrastructure are equal to those on capital in
general. As we can see, it appears to be quite possible for a developing country to have a
excessive investment in electricity generating capacity, relative to its stocks of other
physical capital and of human capital, driving its rate of return down below that on other
forns of capital.
On average, therefore, we find a tendency for returns to electricity generating
capacity in the poorer developing countries to exceed the returns to other forms of capital.
The heterogeneity of the rates of return in the poorer countries suggests independently that
these countries tend to be further from an optimal mix of investment than middle, or
higher, income countries, perhaps reflecting greater market failure, possibly externalities,
or state failure, and thus prime issues for country analysis.
28



4.4 Estimates of the Rate of Return to Paved Roads
Our cost data for paved roads cover a wider incomes span of countries, especially
high-income countries, than those for electricity generating capacity. The first column of
table 7 reports the rates of return to paved roads, based on regression 3 of table 3. For
some developing countries (in 1985, notably South Korea, Colombia, Bolivia and the
Philippines) we find exceptionally high rates of return to paved roads. In some others,
(such as Tunisia and Botswana, again in 1985), rates of return are low. Low rates of return
are also found in most developed countries, with negative returns being present in Austria
and Australia.
Rates of return to capital (this time based on the productivity effects from the same
regression 3 of table 3) are reported in column 2 of table 7. These estimated rates of return
show much less variation than those for paved roads, partly, no doubt, because of the much
larger value of the total capital stock which makes for greater accuracy in the
macroeconomic estimates of the marginal product .
In figure 7 we plot the estimated ratio of the rate of return to paved roads to that
found on capital in general in each country, against the country's log income per capita.
The first point to note is that in most countries, notably in all the developed and high-
income countries, but also in the poorer developing countries, the ratio is less than one. In
these countries the rate of return to paved roads is lower than that on capital in general .
However, in a group of middle income countries the ratio exceeds one by a long way.
These countries get the benefit of a high marginal product of roads coupled with a low cost
of road building. However, even among middle income countries the rates of return to
roads are sometimes lower than the rates of return to capital as a whole.
Once again we find a great deal of heterogeneity across countries in rates of return
to paved roads relative to other forms of capital, and once again the heterogeneity is
greatest among the low-to upper- middle income countries. However, if we are looking for
high rates of return to investment in paved roads, it is in that very class of middle-income
countries that we have to look.
29



4.5 The Rate of Return to Capital
The rate of return to capital as a whole has been used extensively in the last two
sections as a benchmark by which to judge the attractiveness of infrastructure investment.
We now take the rate of return to capital on its own, basing our estimates on regression 1 (
with physical and human capital as the only independent variables) in table 3. Plotting
this in fig.9 against log income per capita for a cross section of countries in 1985 we
obtain, a graph with an inverted U shape. The highest rate of return is found in middle
income countries; and the maximum on the curve corresponds to an income per capita of
$3,600 (international dollars) which is in the top half of the 1985 lower-middle income
class. This result contrasts starkly with the very steeply downward sloping graph for the
rate of return to capital that we obtain from a Cobb-Douglas specification.
The relationship we find is consistent with the observation that actual private
returns to capital are quite low in the poorer developing countries and that capital does not
flow from the rich to the poor (see Lucas (1990)) but rather to middle-income countries. In
developed countries, diminishing returns to capital set in quite slowly because they can
keep their marginal productivity of capital up by having large amounts of human capital.
We nevertheless find some evidence that returns to capital are higher in middle income
developing than in the developed (industrialized) countries, a finding that makes the very
high relative returns to paved roads even more interesting.
5. Conclusion
The use of an aggregate production function allows us to calculate rates of return to
infrastructure that should capture any externalities that escape microeconomic cost-benefit
studies. The model could be improved upon, in particular by estimating a more general
production function, including, for example, the effects of industrial structure and
geography on the productivity of infrastructure.
30



Though our results depend on a number of simplifying assumptions, they appear
plausible. They suggest that as a rule, infrastructure shortages, signaled by high social rates
of return to electricity generating capacity or paved roads, relative to other capital, are
symptomatic of limited groups of countries identified by the income per capita class that
they belong to, essentially the lower-middle and upper-middle income classes of
developing countries. To the extent that such high rates of return are not detected by a
microeconomic cost-benefit analysis, they point to macroeconomic externalities associated
with infrastructure.
References
Andrews, K. and J. Swanson (1995), "Does Public Capital Affect Regional Perfonnance?"
Growth and Change, Vol. 26, pp 204-216.
Baldwin R.E. (1992), "The Measurable Dynamic Gains from Trade," Journal of Political
Economy., Vol. 100, pp 162-174.
Banerjee, A. (1999), "Panel Unit Roots and Cointegration: An Overview," Oxford Bulletin
of Economics and Statistics, Special Issue, pp 607-629.
Barro, R.J. and J.W. Lee (1993), "International Comparisons of Educational Attainment,"
Journal of Monetary Economics, Vol. 32, pp 363-394.
Berndt, E.R., and L.R. Christensen (1974), "Testing for the Existence of a Consistent
Aggregate Index of Labor Inputs," American Economic Review, Vol. 64.
Boarnet, M.G. (1997), "Infrastructure Services and the Productivity of Public Capital: The
Case of Streets and Highways," National Tax Journal, Vol. 50, pp 39-57.
Borland, J. and X.K. Yang. (1992), "Specialization and a New Approach to Economic-
Organization and Growth," American Economic Review, Vol. 82, pp386-391.
Canning, D. (1998), "A Database of World Stocks of Infrastructure, 1950-1995,"
World Bank Economic Review, Vol. 12, pp 529-547.
----------------- (1999), "Infrastructure's Contribution to Aggregate Output," World Bank
Policy Research Working Paper, No. 2246, Washington, D.C.
------------------ and P. Pedroni (1999), "Infrastructure and Long Run Economic Growth,"
mimeo, Queen's University of Belfast.
Carlino, G.A. and R. Voith (1992), "Accounting for Differences in Aggregate State
Productivity," Regional Science and Urban Economics, Vol. 22, pp 597-617.
31



Caselli, F., G. Esquivel and F. Lefort (1996), "Reopening the Convergence Debate: a New
Look at Cross Country Growth Empirics," Journal of Economic Growth, Vol. 1,
363-389.
Chang Y., J.Y. Park. And P.C.B. Phillips.(1999), "Nonlinear Econometric Models with
Cointegrated and Deterministically Trending Regressors," Cowles Foundation for
Research in Economics, Yale University.
Chari, V.V., P.J. Kehoe and E.R. McGratten. (1996), "The Poverty of Nations: A
Quantitative Exploration," NBER Working Paper, No. 5414.
DeFrutos, R.F., M. Garcia-Diez and T. Perez-Amaral (1998), "Public Capital Stock and
Economic Growth: an Analysis of the Spanish Economy," Applied Economics, Vol.
30, pp 985-994.
Duggal, V.G., C. Saltzman and L.R. Klein (1999), "Infrastructure and Productivity, a
Nonlinear Approach," Journal of Ecocometrics, Vol. 92, pp 47-74.
Gallup J.L., J.D. Sachs and A.D. Mellinger (1999), "Geography and Economic
Development, "International Regional Science Review, Vol. 22, pp 179-232.
Garcia-Mila, T., T.J. McGuire and R.H. Porter (1996), "The Effect of Public Capital in
State Level Production Functions Reconsidered," Review of Economics and
Statistics, Vol. 78, pp 177-180.
Haughwout, A.F. (1998), "Aggregate Production Functions, Interregional Equilibrium, and
the Measurement of Infrastructure Productivity," Journal of Urban Economics,
Vol. 44, pp 216-227.
Hulten, C.R. (1997), "Infrastructure Capital and Economic Growth: How Well You Use It
May Be More Important Than How Much You Have," mimeo, University of
Maryland.
Johansen, S. (1991), "Estimation and Hypothesis Testing of Cointegration Vectors in
Gaussian Autoregressive Models," Econometrica, Vol. 59, pp 1550-1580.
Kao, C. and M.H. Chiang (1999), "On the Estimation and Inference of a Cointegrated
Regression in Panel Data," mimeo, Syracuse University.
Kelly, M. (1997), "The Dynamics of Smithian Growth," Quarterly Journal of Economics,
Vol. 112, pp939-964.
Klenow, P.J. and A. Rodriguez-Clare (1997), "The Neoclassical Revival in Growth
Economics: Has it Gone Too Far?," mimeo, Graduate School of Business,
University of Chicago.
Krugman, P. (1991), "Increasing Returns and Economic Geography, " Journal of Political
Economy, Vol. 99, pp 483-499.
----------------- (1996), "Urban Concentration: The Role of Increasing Returns and
Transport Costs," International Regional Science Review, Vol. 19, pp 5-30.
------------------- and A.J. Venables (1995), "Globalization and the Inequality of Nations,"
Quarterly Journal of Economics, Vol. 110, pp 857-880.
32



Limao, N. and A.J. Venables (1999). "Infrastructure, Geographical Disadvantage and
Growth," World Bank, Policy Research Working Paper, No. 2257.
Little, I.A.M. and J.A. Mirrlees (1990), "Project Appraisal and Planning 20 Years On,"
World Bank Economic Review, Special Supplement, pp 351-382.
Lucas, R.E. (1990), "Why Doesn't Capital Flow from Rich to Poor Countries," American
Economic Review, Vol. 80, pp 92-96.
Mankiw, N.G., D. Romer. and D.N. Weil. (1992), "A Contribution to the Empirics of
Economic Growth," Quarterly Journal of Economics, Vol. 107, pp. 407-437.
Moore, E.A. and G. Smith (1990), "Capital Expenditures for Electric Power in the
Developing countries in the 1990s," World Bank, Industry and Energy Department,
Energy Series Working Paper No.21.
Morrison, C .J. and A.E. Schwartz (1994), "Distinguishing External from Internal Scale
Effects - The Case of Public Infrastructure," Journal of Productivity Analysis,
Vol.5, pp 249-270.
Mullen, J.K., M. Williams and R.L. Moomaw (1996), "Public Capital Stock and Interstate
Variations in Manufacturing Efficiency," Journal of Policy Analysis and
Management, Vol. 15, pp 51-67.
Murphy, K.M., A. Shleifer and R.W. Vishny (1989), "Industrialization and the Big Push,"
Journal of Political Economy, Vol. 97, pp 1003-1026.
Phillips, P.C.B. and H.R. Moon (1999), "Nonstationary Panel Data Analysis: An Overview
of Some Recent Developments," Cowles Foundation for Research in Economics,
Yale University.
Pinnoi, M. (1994), "Public Infrastructure and Private Production: Measuring Relative
Contributions," Journal of Economic Behavior and Organization, Vol. 23, pp 127-
148.
Porter, M.E. (1998), "Clusters and the New Economics of Competition," Harvard Business
Review, Vol. 76, pp 77-82.
Pritchett, L. (1996), "Mind Your P's and Q's, The Cost of Public Investment is Not the
Value of Public Capital," World Bank, Policy Research Department, Policy
Research Working Paper No.1660
Summers, R. and A. Heston (1991), "The Penn World Table (Mark 5): An Expanded Set
of International Comparisons, 1950-1988," Quarterly Journal of Economics, Vol.
CVI, pp 327-336.
World Bank (1994), World Development Report 1994, Infrastructure for Development,
New York, Oxford University Press.
-------------- World Development Report, various years.
33



Table 1. The Cobb Douglas Production Function with Infrastructure
Dependent Variable: Log GDP per Worker 1960-1990
Total Factor Productivity  Year         Year         Year         Year
Dummies,    Dummies,    Dummies,    Dummies,
Fixed Effects Fixed Effects  Fixed Effects  Fixed Effects
Short Run Dynamics     2 lags, 1 lead  2 lags, 1 lead  2 lags, 1 lead  2 lags, 1 lead
Log Capital per Worker     0.455        0.404        0.417         0.392
(14.7)       (14.6)       (11.7)       (11.9)
Log Human Capital per      0.125        0.051        0.079        0.059
Worker                    (3.73)        (1.43)       (1.77)       (1.54)
Log Electricity                         0.085                      0.057
Generating Capacity per                (5.83)                     (3.13)
Worker
Log Paved Roads per                                  0.083         0.048
Worker                                               (4.06)       (2.30)
R squared adjusted         0.729        0.678        0.716         0.685
Countries                 97           90           67            62
Observations            2674         2473         1671          1534
Average T                 28           27           25            25
t ratios in parentheses are calculated based on the long run auto-covariance matrix and are
asymptotically N(0, 1).
34



Table 2. The Cobb Douglas Production Function with Infrastructure
In Low Income and High Income Countries
Dependent Variable: Log GDP per Worker 1960-1990
Low Income         High Income
Full Sample         Countries          Countries
Total Factor         Year Dummies,      Year Dummies,       Year Dummies,
Productivity          Fixed Effects      Fixed Effects       Fixed Effects
Short Run          2 lags, 1 lead     2 lags, 1 lead      2 lags, 1 lead
Dynamics
Log Capital per          0.392               0.371              0.365
Worker                  (11.9)              (8.58)             (6.41)
Log Human Capital        0.059               0.035              0.112
per Worker              (1.54)              (0.64)             (1.57)
Log Electricity          0.057               0.012              0.117
Generating Capacity      (3.13)             (0.50)             (3.73)
per Worker
Log Paved Roads          0.048               0.003              0.134
per Worker              (2.30)              (0.12)             (4.05)
R squared adjusted       0.685               0.582              0.478
Countries               62                  31                 31
Observations          1534                 781                753
Average T               25                  25                 24
t ratios in parentheses are calculated based on the long run auto-covariance matrix and are
asymptotically N(0, 1).
Sample split on the basis of income per capita in 1975
35



Table 3. The Translog Production Function with Infrastructure
Dependent Variable: Output per Worker
Regression                   (1)               (2)              (3)
Log Input(per worker)
Capital                        0.072           -0.038             0.017
(0.70)          (0.20)            (0.10)
Human Capital                 -0.151            0.992             0.569
(1.39)          (6.31)            (3.27)
Electricity                                    -0.869
(7.47)
Paved Roads                                                      -0.398
(2.98)
Capital Squared                0.026            0.034             0.027
(3.57)          (3.50)            (2.75)
Human Capital Squared         -0.064           -0.114            -0.062
(5.78)          (5.76)            (2.92)
Electricity Squared                            -0.061
(10.9)
Paved Roads Squared                                              -0.054
(6.36)
Capital*Human Capital          0.049           -0.049            -0.039
(3.81)          (3.13)            (1.89)
Capital*Electricity                             0.069
(6.07)
Capital*Paved Roads                                               0.044
(2.87)
Human Capital*Electricity                       0.152
(9.31)
Human Capital*Paved
Roads                                                             0.101
(6.12)
R squared adjusted             0.993            0.995             0.996
N                           2674             2473              1671
Countries                     97               90                67
Number of Short Run          582              810               603
Parameters
36



Table 4. Elasticity of Output
Regression   Elasticity of Output       Inputs Per Worker in 1985
No.           with Respect to  Lower Quartile   Median   Upper Quartile
(1)   capital                    0.50          0.59        0.65
human capital             0.09           0.11        0.11
(2)   capital                    0.35          0.52        0.65
human capital             0.08           0.08        0.13
electricity               0.06           0.09        0.07
(3)   capital                    0.43          0.52        0.61
human capital             0.14           0.09        0.14
paved roads                0.05          0.09        0.04
37



Table 5. Unit Costs of Construction
Paved Roads,
International $ per    Electricity    Electricity International
Infrastructure units     kilometer        US $ per kilowatt     $ per kilowatt
Year                       1985                1989                 1985
Algeria                                       2347                 2193
Angola                                        3400                 3257
Argentina                80223                 1902                2780
Australia               869154
Austria                 506012
Bangladesh                                    2815                17833
Belgium                 402887
Bolivia                 180458                 1740                3177
Botswana                256089
Brazil                  639203                2655                 5447
Cameroon                278808
Canada                  500760
Central Afr.R.                                7786                15407
Chile                   143840                 1924                4126
China                                         1502                 4695
Colombia                169987                2564                 5401
Congo                                         2429                 4934
Costa Rica              131966                2301                4143
Cyprus                                        2655                 3982
Denmark                 400378
Dominican Rep.          253455                 1914                4850
Ecuador                 366371                2439                 4581
Egypt                                          1590                3498
El Salvador             540362                3971                 7127
Ethiopia                712160                2689                 6128
Fiji                                          2923                 4924
Finland                 477889
France                  386139
Gambia                                        1769                 3929
Germany, West           443177
Ghana                                         2460                 3274
Guatemala               631965                4719                 6785
Honduras                771088                2144                 3006
Hong Kong               305218
Hungary                 159311                3439                 7878
India                   143306                2061                 6504
Indonesia               200008                 1829                4736
Ireland                 399348
Israel                  337680
Italy                   296089
Ivory Coast             288277                 1680                3048
Jamaica                                       2023                 4196
Japan                   339714
38



Table 5. (continued)
Paved Roads                             Electricity
International $ per    Electricity     International $ per
Infrastructure units     kilometer       US $ per kilowatt       kilowatt
Year                       1985               1989                1985
Jordan                                        1797                2846
Kenya                   285128                1717                3779
Korea, Rep.              92072                2990                4651
Lesotho                                       2918               14928
Liberia                 426839
Luxembourg              402887
Madagascar               176712               4882               11174
Malawi                  282163                1990                5499
Malaysia                                      1746                3057
Mali                                          1957                6145
Mexico                                        1949                3729
Morocco                 270454                2145                6040
Mozambique                                    6250               15957
Myanmar                                       2719                7646
Nepal                                        *4346               22989
Netherlands             529989
New Zealand             456604
Nicaragua                                     3229                5280
Niger                                         7000               14977
Nigeria                                       2793                2560
Norway                  438496
Pakistan                434650                1390                4550
Panama                   187551               3417                4423
Papua N.Guinea                                1925                3737
Peru                                          3393                8273
Philippines              111343               2043                4708
Poland                                        1851                3404
Portugal                236770                2330                4858
Senegal                 306742               13600               32856
Sierra Leone                                  3038                6304
Somalia                                       3268                5413
Spain                   236990
Sri Lanka                65277                4451               19930
Sudan                                         2422                5293
Sweden                  522244
Syria                                         1539                3458
Tanzania                221723
Thailand                                      2034                5823
Tunisia                 313404                1189                2415
Turkey                  228506                1849                4555
U.K.                    777133
U.S.A.                  627580
39



Table 5. (continued)
Paved Roads                           Electricity
International $ per   Electricity    International $ per
Infrastructure units.   kilometer      US $per kilowatt       kilowatt
Year                     1985               1989               1985
Uruguay                 95440               1778               3776
Yugoslavia                                  1702               3591
Zambia                 144577
Zimbabwe               277287               1927               3660
40



Table 6. Rates of Return to Electricity Generating Capacity and Capital
Rate of Return    Rate of Return      ROR EGC/
to EGC           to Capital          ROR K
Algeria                   0.63              0.15               4.20
Argentina                 0.46              0.29               1.59
Bangladesh                0.61              0.80               0.77
Bolivia                   0.92              0.19               4.74
Brazil                    0.10              0.58               0.16
Central Afr.R.            0.40              0.12               3.25
Chile                     0.41              0.73               0.56
China                     0.54              0.41               1.31
Colombia                  0.28              0.55               0.50
Congo                     1.14              0.25               4.58
Costa Rica                0.25              0.36               0.69
Cyprus                    0.36              0.31               1.19
Dominican Rep.            0.25              0.61               0.42
Ecuador                   0.45              0.50               0.91
Egypt                     0.45              0.50               0.90
El Salvador               0.17              0.42               0.40
Fiji                      0.32              0.30               1.06
Gambia                    1.05              0.23               4.49
Ghana                     0.25              0.18               1.37
Guatemala                 0.18              0.34               0.52
Honduras                  0.95              0.27               3.56
India                     0.24              0.53               0.44
Indonesia                 1.06              0.62               1.70
Jamaica                   0.11              0.20               0.54
Jordan                    0.40              0.42               0.96
Kenya                     1.25              0.19               6.63
Korea, Rep.               0.31              0.45               0.68
Malawi                    0.54              0.18               3.00
Malaysia                  0.77              0.44               1.76
Mali                      0.51              0.24               2.16
Mexico                    0.51              0.52               0.98
Mozambique               -0.07              0.17               -0.42
Myanmar                   0.34              0.33               1.03
Nepal                     0.40              0.56               0.72
Nicaragua                 0.20              0.30               0.67
Niger                     0.12              0.13               0.92
Pakistan                  0.18              0.95               0.19
Panama                    0.21              0.38               0.55
Papua N.Guinea            0.06              0.24               0.26
Peru                      0.21              0.40               0.51
Philippines               0.44              0.35               1.25
Portugal                  0.07              0.46               0.14
41



Senegal                 0.06              0.24              0.27
Sri Lanka               0.27              0.86              0.31
Syria                   0.35              0.80              0.44
Thailand                 0.42             0.61              0.69
Tunisia                  0.40             0.37               1.07
Turkey                   0.32             0.72              0.45
Uganda                   0.80             0.02              46.26
Uruguay                  0.30             0.51              0.59
Yugoslavia               0.24             0.34              0.72
Zimbabwe                 0.05             0.38              0.14
42



Table 7. Rates of Return to Paved Roads
Rate of Return     Rate of Return    ROR Paved Roads/
to Paved Roads       to Capital         ROR Capital
Argentina                  3.85                0.29               13.33
Australia                 -0.01                0.30               -0.02
Austria                    0.00                0.29               -0.02
Belgium                    0.06                0.40                0.14
Bolivia                    7.96                0.21               37.09
Botswana                   0.20                0.58                0.34
Brazil                     0.61                0.57                1.07
Cameroon                   1.88                0.35               5.31
Chile                      5.24                0.73                7.15
Colombia                   9.47                0.54               17.53
Costa Rica                 1.96                0.37                5.24
Denmark                    0.12                0.30                0.40
Ecuador                    1.97                0.51                3.85
El Salvador                1.11                0.47                2.38
Finland                    0.15                0.22                0.68
Germany, West              0.16                0.29               0.55
Guatemala                  0.76                0.38               2.01
Honduras                   0.39                0.34                1.15
India                      0.74                0.78                0.96
Indonesia                  2.03                0.83                2.45
Ireland                    0.06                0.36                0.15
Italy                      0.26                0.34                0.76
Japan                      0.62                0.20                3.05
Kenya                      0.53                0.35                1.51
Korea, Rep.               15.76                0.43               36.95
Liberia                    1.04                0.15                6.82
Malawi                     0.60                0.40                1.50
Netherlands                0.15                0.32                0.46
New Zealand                0.08                0.36                0.23
Norway                     0.02                0.21                0.08
Pakistan                   0.52                1.17                0.45
Panama                     2.18                0.38                5.76
Philippines                7.19                0.40               17.99
Senegal                    0.48                0.45                1.07
Sweden                     0.06                0.29                0.21
Tunisia                    0.16                0.43                0.36
Turkey                     1.58                0.78                2.03
U.K.                       0.13                0.39                0.32
U.S.A.                     0.07                0.29                0.26
Zambia                     0.65                0.24                2.69
Zimbabwe                   0.15                0.45                0.33
43



Figure 1
Elasticity of Output w ith Respect to Capital
0.7
0.6
~0.5
04
~0.3
0.2
0.1
0
6            7             8            91
log Income per Capita               10
Figure 2
Elasticity of Output with Respect to Education
0.4
0.3
t> 0.2  +  ;w
0.
w  O. i
0
6            7             8            9            10
log Income per Capita
44



Figure 3
Elasticity of Output with Respect to EGC
0.3
0.2
0.1
0
-0.2
-0 .3*
6           7           8           9          10
log Income per Capita
Figure 4
Elasticty of Output with Respect to Paved Roads
0.2.*.*
.50.1
0
-0.2
6           7           8            9           10
Log Income per Capita
45



Figure 5
Cost of Paved Roads
1000000l
800000
E  600000*
400000    *    *
6         7          8          9         10
log income per capita
Figure 6
Cost of Eecricity Generating Capacity
45000
40000
35000
30000
25000
~20000
o 15000    *  
10000
6          7          8           9          10
log income per capita
46



Figure 7
The Rate of Return to EGC Relative to Capital
20
15
0
-   10
0
-5
6          7           8           9          10
log Income per Capita
Figure 8
Rate of Return to Paved Roads Relative to Capital
15
'I-  10
o
U)
0
-5   .  . .
6            7            8            9            10
log income per capita
47



Rgure 9
Rae of Fbturn to Capita
1C0
~80
,E 60.@                              w<w -
Eo
0~~~~~~~4
~40
0
6     6.5    7      7.5    8     &5    9       9.5   10
log incore per cpita
48






Policy Research Working Paper Series
Contact
Title                           Author                  Date              for paper
WPS2365 Leading Indicator Project: Lithuania  Stephen S. Everhart  June 2000        M. Geller
Robert Duval-Hernandez                    85155
WPS2366 Fiscal Constraints, Collection Costs, Keiko Kubota       June 2000          L. Tabada
and Trade Policies                                                         36896
WPS2367 Gender, Poverty, and Nonfarm      Constance Newman       June 2000          M. Clarke
Employment in Ghana and Uganda    Sudharshan Canagarajan                   31752
WPS2368 Seeds of Corruption: Do Market    Harry G. Broadman      June 2000          S. Craig
Institutions Matter?            Francesca Racanatini                       33160
WPS2369 How the Proposed Basel Guidelines  Giovanni Ferr         June 2000          E. Mekhova
On Rating-Agency Assessments    Li-Gang Liu                                85984
Would Affect Developing Countries   Giovanni Majnoni
WPS2370 A New Model for Market-Based      Marcelo Giugale        June 2000          M. Geller
Regulation of Subnational Borrowing: Adam Korobow                          85155
The Mexican Approach            Steven Webb
WPS2371 Shock Persistence and the Choice   Marcelo Giugale       June 2000          M. Geller
of Foreign Exchange Regime: An  Adam Korobow                               85155
Empirical Note from Mexico
WPS2372 Financial Openness, Democracy,    Mansoor Dailami        June 2000          W. Nedrow
and Redistributive Policy                                                  31585
WPS2373 Reciprocity across Modes of Supply  Aaditya Mattoo       June 2000          L. Tabada
in the World Trade Organization:  Marcelo Olarreaga                        36896
A Negotiating Formula
WPS2374 Should Credit Be Given for        Aaditya Mattoo         June 2000          L. Tabada
Autonomous Liberalization in    Marcelo Olarreaga                          36896
Multilateral Trade Negotiations?
WPS2375 Asset Distribution, Inequality,   Klaus Deininger        June 2000          M. Fernandez
and Growth                      Pedro Olinto                               33766
WPS2376 The Effect of Early Childhood     Michael M. Lokshin     June 2000          P. Sader
Development Programs on Women's  Elena Glinskaya                           33902
Labor Force Participation and   Marito Garcia
Older Children's Schooling in Kenya
WPS2377 Reforming the Water Supply in     Claude Menard          June 2000          H. Sladovich
Abidjan, C6te d'lvoire: Mild Reform    George Clarke                       37698
in a Turbulent Environment



Policy Research Working Paper Series
Contact
Title                            Author                  Date              for paper
WPS2378 Disintegration and Trade Flows:   Simeon Djankov          June 2000          R. Vo
Evidence from the Former Soviet  Caroline Freund                           33722
Union
WPS2379 India and the Multilateral Trading  Aaditya Mattoo        June 2000          L. Tabada
System after Seattle: Toward a   Arvind Subramanian                        36896
Proactive Rolie
WPS2380 Trade Policies for Electronic     Aaditya Mattoo          June 2000          L. Tabada
Commerce                         Ludger Schuknecht                          36896
WPS2381 Savings and the Terms of Trade    Pierre-Richard Agenor    June 2000         T. Loftus
under Borrowing Constraints     Joshua Aizenman                            36317
WPS2382 Impediments to the Development and Thorsten Beck          June 2000          E. Mekhova
Efficiency of Financial Intermediation                                     85984
in Brazil
WPS2383 New Firm Forrnation and Industry  Thorsten Beck           June 2000          E. Mekhova
Growth: Does Having a Market- or    Ross Levine                            85984
Bank-Based System Matter?
WPS2384 Are Cost Models Useful for Telecoms Daniel A. Benitez     July 2000          G. Chenet-Smith
Regulators in lDeveloping Countries?  Antonio Estache                      36370
D. Mark Kennet
Christian A. Ruzzier
WPS2385 The Rise, the Fall, and ... the   Antonio Estache         July 2000          G. Chenet-Smith
Emerging Recovery of Project    John Strong                                36370
Finance in Transport
WPS2386 Regulators and the Poor: Lessons    Richard Green         July 2000          G. Chenet-Smith
from the United Kingdom                                                     36370
WPS2387 The Long and Winding Path to Private Antonio Estache      July 2000          G. Chenet-Smith
Financing and Regulation of Toll  Manuel Romero                            36370
Roads                            John Strong
WPS2388 The Role of Special and Differential   Constantine Michalopoulos July 2000   L. Tabada
Treatment for Developing Countries in                                       36896
GATT and the WNorld Trade Organization
WPS2389 Vietnam: On the Road to Labor-    Patrick Belser          July 2000          H. Sutrisna
Intensive Growth?                                                          88032