WPS3858
Estimates of Government Net Capital Stocks for 26
Developing Countries, 1970-2002
Florence Arestoff and Christophe Hurlin
Abstract
We provide various estimates of the government net capital stocks for a panel of 26 developing
countries over the period 1970-2001. Two kinds of internationally comparable series of public
capital stocks are proposed. Firstly, we provide estimates based on the standard perpetual
inventory method and various assumptions on initial stocks and deprecation rates. Secondly, we
propose to take into account the potential inefficiency of public investments in creating capital
with a non parametric approach. Three estimates of net capital stocks are then proposed for three
assumptions on the public investment efficiency.
Key Words: Public Capital, Capital Stocks, Developing Countries. J.E.L Classification
Numbers: C82, E22, E62.
World Bank Policy Research Working Paper 3858, March 2006
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the
exchange of ideas about development issues. An objective of the series is to get the findings out quickly,
even if the presentations are less than fully polished. The papers carry the names of the authors and should
be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely
those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors,
or the countries they represent. Policy Research Working Papers are available online at
http://econ.worldbank.org.
We would like to thank Santiago Herrera for his support and his comments on a previous version of this
work.
EURIsCO, University of Paris Dauphine. Place du Maréchal De Lattre de Tassigny. 75016 Paris.
France. Email address: florence.arestoff@dauphine.fr.
LEO, University of Orléans. Rue de Blois. BP 6739. 45067 Orléans Cedex 2. France. Email
address: christophe.hurlin@univ-orleans.fr.
1 Introduction
Since the seminal works of Aschauer (1989a,b), the measure of the productivity and the
efficiency of infrastructure and public capital has been the subject of many empirical
studies, for the OECD countries and especially for the United States1 but also for the
case of developing countries. One of the main differences between the papers concerning
the OECD countries and those concerning the developing countries is that the first ones
use rather the notion of "public capital" or "public spending" while the second ones
generally emphasize the notion of "infrastructure". Thus, if we consider only the titles
of the most recent papers devoted to the OECD countries and published in the most
famous economic reviews, the terms of "public capital" (Hulten and Peterson, 1984 ;
Aschauer, 1989b, Ram and Ramsey, 1989 ; Lynde and Richmond, 1992 ; Evans and
Karras, 1994 ; Holtz-Eakin, 1994 ; Garcia-Mila, McGuire and Porter, 1996; Otto and
Voss, 1998 ; Fernald, 1999) or "public spending" (Aschauer, 1989a ; Munnell, 1990a ;
Sturm and De Haan, 1995 ; Devarajan, Swaroop and Zou, 1996 ) appear more often than
the notion of "infrastructure" (Morrisson and Schwartz, 1996 ; Nadiri and Mamuneas,
1994). Moreover, when the notion of "infrastructure" is used in the title, the definition
used to construct the data is sometimes equivalent to the definition of the public capital
(Ford and Poret, 1991). On the contrary, the empirical studies devoted to developing
countries are generally based on the past flows of public investments or on physical
measures of infrastructure (private, but also public). It is for instance the case in
the World Development Report for 1994 (Infrastructure and Development), Canning
(1999), Canning and Pedroni (1999), Canning and Bennathan (2000), or in the recent
study done by Easterly and Serven (2004) for Latin America.
There are many possible explanations of this dichotomy in the literature. Undoubt-
edly, one of them is the lack of homogeneous data on public capital stocks for the
developing countries. Consequently, as noted by Pritchett (1996), "there are few good
estimates of the productivity of public capital [for developing countries] and what few
there are, are of limited comparability" (Pritchett, 1996, page 36). This last point, i.e.
the comparability of the estimates of the public capital productivity, and consequently
the homogeneity of the definition used for the capital stocks, is essential. This prop-
erty of comparability is required if we want to assess the robustness of the estimates
of the public capital productivity for a country and to compare it directly with the
estimates obtained for OECD countries or for the other developing countries. It is
notably the case when physical measures are used to estimate the social rate of return
of infrastructure investments, as in Canning and Bennathan (2000) for instance.
In this study, we propose to estimate internationally comparable net public capital
stocks for a panel of 26 developing countries over the period 1970-2000. The public
1For more details on this subject see the surveys of Gramlich (1994), Sturm (1998) or Romp and
De Haan (2005).
2
capital is considered here as mean of production. Given the definition of the System
of National Accounts (SNA, 1993), the concept of capital stocks encompasses fixed
capital as a set of tangible or intangible assets produced as outputs from the produc-
tion processes that are themselves repeatedly or continuously used in other production
processes for a period longer than one year. It is important to note that this definition
also includes the intangible assets (e.g. computer software), contrary to the definition
proposed in the 1968 SNA. The public capital stocks defined in this manner does not
correspond to the notion of "infrastructure". It is obvious that the gross fixed capital
formation of the public sector includes non-productive investments. Even if it would be
possible to measure the public investments corresponding to the most important cate-
gories of infrastructure, "it would be naive to believe that everything called infrastructure
spending in the fiscal accounts is necessarily productive or that such spending should be
the only -or even the main- indicator of public infrastructure performance" (Calderon,
Easterly and Serven, 2004, page 3). On the contrary, some infrastructure investments
are financed by the private sector. This point is important since in many develop-
ing countries, as those in Latin America for instance, the private part of investments
in infrastructure has recently increased, as an effect of the periods of macroeconomic
adjustments.
Given these differences between the notion of public capital and that of capital in-
frastructure, we intend to evaluate specifically the public capital stocks in developing
countries. In order to propose an international comparison, we consider the same envi-
ronment of analysis as that generally used for the OECD countries. Naturally, the level
of stock is not in itself helpful to policy development for a variety of reasons. The nature
and the type of investments involved, the organizational and institutional arrangements,
the nature of the available private capital, and so on, will influence the action to be
taken. However, in an international perspective, such an evaluation provides limited
guidance for evaluating the situation of under-provision, or on the contrary oversup-
ply, of public investments. It is recognized that the situations of oversupply of public
investments have a negative impact on the economy, as the situations of infrastructure
shortage, since it draws scarce resources away from maintenance and operation of exist-
ing stocks. In this context, the use of a homogenous definition of public capital stocks
(as it was the case for the homogenous definition of infrastructure stocks), whatever it
is based on physical measures or on monetary flows, is the main ingredient required in
the comparative study of the social rates of return of public investments.
For the same reasons, a similar work has been recently done by Kamps (2004) who
has proposed various estimates of the government net capital stocks for twenty-two
OECD countries . But, if we propose to evaluate the public capital productivity in
developing countries as it is done for the OECD countries, it does not imply that we
propose to estimate the capital stocks according to the same methodology. Except for
the United States (Bureau of Economic Analysis), there are a few available databases
of public capital stocks even for the major OECD countries2. Consequently, several
2In his study devoted to 22 OECD countries, Kamps (2004) found 13 countries for which public
capital stocks series are available from national sources. But, the definitions of public capital used are
generally not homogeneous.
3
studies have been devoted to the estimation of net stocks using the investments flux:
Sturm and De Haan (1995) for the Netherlands, Berndt and Hansson (1992) for Sweden,
Ford and Poret (1991) for France and Japan and more recently Kamps (2004) for 22
OECD countries. All these studies use the same methodology to estimate the net capital
stocks: the Perpetual Inventory Method (PIM, thereafter). This well known method
consists in cumulating historical series of past investments and in deducting assets which
were retired. However, it is recognized that this method based on monetary value of
investments cannot be directly adopted in the case of developing countries. Pritchett
(1996) showed that in many poor countries the problem is not that governments do not
invest, but that these investments do not create productive capital. The cost of public
investments does not correspond to the value of the capital stocks. In other words, in
some developing countries, one dollar's worth of public investment spending does not
create one dollar's worth of public capital. Pritchett estimates that only slightly more
than half of the money invested in investment projects will have a positive impact on the
public capital stocks in the developing countries. Consequently, under this assumption,
the use of the PIM leads to overvalue the public capital stocks effectively available. This
issue is generally related to the debate concerning the efficiency of public investments
in the developing countries (World Bank, 1994).
That is why we propose two kinds of estimates of public capital stocks for each
country of the panel. Firstly, we consider the estimates of net capital stocks based on the
PIM. These estimates will be considered as benchmark in our econometric estimations
of the public capital productivity in order to compare our results to those generally
obtained for the OECD countries. This method requires a variety of assumptions
concerning the depreciation patterns and the treatment of the initial stocks. We assume
a geometric depreciation pattern with a time varying depreciation rates. For a given
asset, the depreciation rate is generally considered to be constant. But, the structure
of public investments in developing countries has deeply changed over the last decades.
As the relative importance of different kinds of assets changes with time, so the average
depreciation rate defined on the total stocks does. Consequently, the depreciation rate
of the aggregated stocks cannot be considered constant. In this study, we use the
structure of public investments in LA studied by Calderon, Easterly and Serven (2004)
as a benchmark for the 26 developing countries of our panel. Given this decomposition,
we estimate an aggregated time varying depreciation rate. The initial stocks are chosen
according to the same methodology as that used by Kamps (2004) for OECD countries.
For each country of our panel, an artificial investment series is constructed since 1880.
These series are used as inputs to estimate the initial capital stocks. Finally, we analyze
the robustness of our estimates to these two main assumptions.
Secondly, we propose to evaluate the capacity of public investment to generate pub-
lic capital in developing countries. For that, we intend to link the monetary flows of
sectoral investments with the corresponding physical measures of capital stocks effec-
tively available and taken from the Canning' database (1998). We consider four sectors
(electricity, telecommunications, roads and railways) of two Latin American reference
countries (Mexico and Colombia). For each country and each sector, the analysis is
done over a period in which the public infrastructure investments represent more than
85% of the total sectoral investments. Then, we estimate a relationship between the
4
increase in the monetary value of the stocks and the current monetary value of public
investments. This relation gives us the value of the public capital produced by one
dollar's worth of government investment spending. If the PIM gives a correct approxi-
mation of the public stocks, we should observe the following relationship: one invested
dollar should increase the stock value with one dollar. On the contrary, if it is observed
that the stocks value is increased with less than one dollar, it implies that the PIM
overvalues the public stocks. But, more originally, this approach also gives us an esti-
mate of the part of the public investments that are really "productive", i.e. that are
efficient in creating capital.
Since no natural specification of this efficiency function can be justified, we propose
to estimate it by a non parametric method. We consider the road sector in the United
States as a benchmark and we compare the monetary valorization of the number of road
kilometers (Canning, 1998) to the flows of public investments in road and highways.
The results show that, in this case, the cumulated investment flows represent a good
proxy for the capital stocks. On the contrary, the same methodology applied to our
two reference developing countries shows that there is a large discrepancy between the
amount of investments and the value of the increase in stocks. Moreover, we show that
the estimated efficiency function in these countries is almost linear, i.e. the component
of the public investments really productive can be approximated by a constant fraction
in the total investments. We assume that this last result can be generalized to all the
countries of our panel and we estimate new series of net public capital stocks by using
only a fraction of investments to improve the stocks value. Naturally, we do not assume
that all developing countries have the same ratio of "productive" investments to the
total investments. For each country, we propose a sensitivity analysis and build three
estimates corresponding to three different calibrated values of this ratio.
The first section of this study is devoted to the estimates of the net capital stocks
using the PIM. In this section, the depreciation profiles, the choice of initial stocks and
the robustness of our estimates to these two assumptions are successively presented.
The second section is devoted to the measure of public investments' efficiency. The
conclusions are presented in the last section.
2 Government Net Capital Stocks in Developing
Countries
In this article, we propose to estimate government net capital stocks at constant prices
for a panel of developing countries. Our choice to estimate net capital stocks rather
than gross capital stocks is clearly related to our general aim which is to evaluate the
productivity of public capital in these countries. The gross capital stock measures the
value of assets at their new prices regardless their age. On the contrary, the net capital
stock corresponds to the value of assets held by the government at the price of new assets
minus the cumulative value of depreciation or, more generally minus the consumption
of fixed capital. The depreciation corresponds to the decline of the current value due
to the physical deterioration of assets. But, the net capital stock accounts also for
5
economic depreciation, i.e. the loss of value. Thus, the net public capital stock is an
estimate of the market value of the fixed assets held by government. More precisely, if
markets for used capital assets were perfect, the net capital stock would be a measure
for the present value of future capital services. As noted by Kamps (2004), this market
value corresponds to the so-called productive capital stock defined in dynamic general
equilibrium models or used in empirical measures of productivity (OECD, 2001b).
In order to estimate these net public capital stocks, in a first step we propose to
estimate benchmark series based on the perpetual inventory method.
2.1 The Perpetual Inventory Method
The PIM consists in cumulating historical series on past investments and deducting
assets which are retired. To our knowledge, it is the only method3 used in the literature
devoted to public capital or infrastructure capital stocks, if we except the Canning's
database (1998) based on physical measures of infrastructure. The estimation of the
net capital stocks by the PIM requires the following information:
· Sufficiently long gross investment time series, valued at constant prices.
· The law of depreciation of capital goods, and more precisely the time profile of
the depreciation rate.
· An evaluation of the initial net capital stocks.
The estimation is carried out for the 26 following developing countries: Botswana,
Burkina Faso, Cameroon, Colombia, Costa Rica, Dominican Republic, Fiji, Ghana,
India, Indonesia, Kenya, Malaysia, Mauritius, Mexico, Morocco, Pakistan, Panama,
Paraguay, Peru, Philippines, Thailand, Tunisia, Turkey, Uruguay, Venezuela, RB and
Zimbabwe. These countries have been selected for their data availability and for their
relative political and historical stability in order to avoid taking into account the dam-
ages of particular phenomena such as war, civil war etc. in the estimates of capital
stocks. Two countries (Mexico and Turkey) are OECD members, but their capital
stocks are not estimated in Kamps (2004). Our data on public investment are taken
from the World Bank World Development Indicators (WDI thereafter). The series cor-
respond to the capital expenditure for central government only4. Capital expenditure is
spending to acquire fixed capital assets, land, intangible assets, government stocks, and
nonmilitary, nonfinancial assets. Also included are capital grants. The investments ex-
pressed in constant prices are computed by deflating these series with the GDP implicit
price deflator5. Indeed, the price indexes to derive stock estimates for current-cost and
3 An alternative method consists in estimating directly the stock for a benchmark year using disag-
gregated values for each assets. However, with this method, only gross capital stocks can be estimated.
The net capital stocks are always estimated by a variant of PIM. Since in this paper, we are interested
in net capital stocks, the PIM approach cannot be avoided.
4 WDI Code: GB.XPK.TOTL.CN. Source International Monetary Fund, Government Finance Sta-
tistics Yearbook.
5 WDI Code: NY.GDP.DEFL.ZS. The base year is not the same for all the countries (see WDI,
2004).
6
real cost estimates are generally the same as those used to derive real GDP and its
components (see for instance, Bureau of Economic Analysis, 2003, page M-8).
Figure 1 displays the average government capital expenditure - GDP ratio for a
reduced sample of 19 developing countries6 over the period 1974-1997 (left hand side
scale) and the standard deviation (right hand side scale). The first remark is that,
on average, the public investment ratios observed for these countries are very close to
the values observed for OECD countries. The mean of the capital expenditure ratios
in our sample is 4.42%. This value roughly corresponds to that observed on average
for the OECD countries before the middle of the 1970s (Romp and De Haan, 2005).
As it was stated for OECD countries (Ford and Poret, 1991; Sturm, 1998; Kamps,
2004), the average public investments - GDP ratio also decreases for the developing
countries of our sample. The average ratio was 5.15% for the period 1974-1984, and
only 3.80% after 1985. Two remarks can be made at this step. Firstly, the decline
of public capital spending in the developing countries of our sample is, on average,
comparable to the decline observed in the 1970s in the OECD countries. For instance,
the public investment to GDP ratio for the United Sates was equal to 6.10% in 1961
and was only slightly superior to 3.00% in 1997. In France, this ratio was relatively
stable, around 3.5%, for the period 1960-1990. However, in the same time the average
part of the other public expenditures in the GDP increased from 40% during the period
1960-1979 to more than 50% during 1990-1997 (Hurlin, 2000). Secondly, this decline
has occurred later in developing countries than in OECD countries. It seems that, on
average, the decline in the ratios of government capital expenditure to GDP started
after the beginning of the 1980s, and mainly occurred in this decade. This observation is
identical to that made by Calderon, Easterly and Serven (2004) for nine Latin American
economies where the decline was partly due to the context of fiscal austerity of the 1980s
and 1990s. On the contrary, it is largely acknowledged that the decline of OECD public
investments mainly occurred in the 1970s and even at the early beginning of 1970s for
the United States if we consider infrastructure investments in roads (Fernald, 1999).
Finally, it is important to remark that the average ratio reached at the end of 1990s
in the developing countries of our sample, i.e. 3.70%, remains superior to the average
ratio observed in OECD countries at the same period, i.e. around 3.00%.
These conclusions based on average ratios hide a very heterogeneous reality. Figure
2 displays the government capital expenditure - GDP ratios for the 26 countries of our
sample. This ratio varies considerably across countries. Over the period 1974-1997,
the average ratio ranged from 1.73% for India to 9.26% for the Morocco. Moreover,
the trends are very heterogeneous. For instance, the decrease in the public investment
ratios is very pronounced during the 1980s for most of the Latin American countries
of our sample (Mexico, Panama, Peru, and Venezuela), except Colombia and Paraguay
where the decline mainly occurred during the 1970s. In the same time, the govern-
ment capital expenditures are relatively stable in Thailand. This opposition between
the Latin America and one of the East Asian Miracle Economies (World Bank, 1994)
6Since our panel is unbalanced, the average ratios displayed on Figure 1 are computed using a
reduced and balanced sample of 19 countries over the period 1974-1997. The excluded countries are
the followings: Botswana, Burkina Faso, Cameroon, Fiji, Ghana, Mauritius and Zimbabwe.
7
reflects the oppositions observed by Calderon, Easterly and Serven (2004), when they
consider physical measures of infrastructure. The only global and common observation
that we can make at this step concerns the volatility of the public investment ratios.
The variances of these ratios in the developing countries are clearly superior to those
observed in the OECD countries. Using Kamps's database (2004), Romp and De Haan
(2005) found that the standard deviation of the government investment ratio for OECD
22 countries during the period 1960-2001 ranged from 1.00% to 1.80%. On the Figure 1,
it can be observed that the corresponding standard deviation for developing countries
during the period 1974-1997 ranges from 1.75% in 1996 to 4.28% in 1977. This high
volatility is clearly observed on Figure 2 for the majority of the countries of our panel.
This result is undoubtedly linked to the particular status of the public investments
component in the total public expenditures. Much, although not all, of the public
spending decline can be traced to fiscal adjustment. As it was earlier noted by Oxley
and Martin (1991) "the political reality is that it is easier to cut-back or post-pone
investment spending than it is to cut current expenditures". Consequently, the public
investment is generally more reduced during times of fiscal stringency than the other
public expenditures. This observation has been largely documented for OECD countries
(Sturm, 1998), but also for Latin American countries (Calderon, Easterly and Serven,
2004) or for some African countries (World Bank, 1994). The developing countries, as
other countries, lower their budget deficit mainly by cuts in capital spending. However,
the amplitude of these adjustments and their frequencies seem to be more important
in the developing countries.
As it was previously mentioned, two other elements are now required to estimate
the net capital stocks: the time profile of the depreciation rate and an evaluation of
the initial stocks.
2.2 Depreciation Profiles
In the PIM, the pattern of depreciation charges for a given assets is determined by
its "depreciation profile". This profile describes the pattern of how, in the absence of
inflation, the price of an asset declines as it ages. In this study, we consider a geometric
pattern for the depreciation of public capital. For any given year, the constant-dollar
depreciation charge is obtained by multiplying the depreciation charge in the preceding
year by one minus the annual depreciation rate. It corresponds to the depreciation
profile generally used in the literature devoted to the estimation of net capital stocks
(private or public) and in particular used by the BEA for the United States to evaluate
the majority of assets, including government assets (Bureau of Economic Analysis, 1999
and 2003.). This choice is also justified by the fact that the available data suggest that
geometric patterns closely approximate the actual profile of price declines for a number
of assets7 (Hulten and Wykoff, 1981). That is why, the geometric pattern was used to
measure the depreciation of public capital by Kamps (2004a,b) for OECD countries,
Sturm and De Haan (1995) for the Netherlands, Berndt and Hansson (1992) for Sweden,
Ford and Poret (1991) for France and Japan. The only exception is Declercq (1994)
7Hulten and Wykoff show that in a collection of assets, if each asset depreciates as an on-hoss shay,
the aggregated stock depreciates with a near-geometric profile.
8
who used physical measures and technical mortality tables to estimate the net capital
stocks of infrastructure in France.
In this context, the issue is to choose the appropriate geometric depreciation rate
and mainly to decide if this depreciation is constant over time. Given the level of
aggregation of our public investments data, we decide to consider a time varying depre-
ciation profile8 as it was done by Kamps (2004) for OECD countries or Declercq (1994)
for France. The reason of our choice is the following: generally, for a given asset, the
depreciation rate is assumed to be constant over time, except for some assets related
to information and communication technology (BEA, 1999). However, the structure of
public investments has deeply changed over the last decades. As the relative impor-
tance of the different kinds of assets changes with time, so the average depreciation rate
defined on the total stock does. Consequently, the depreciation rate of the aggregated
stock cannot be assumed constant. For instance, Kamps (1994) shows that the average
depreciation rate for public capital in the United States, approximated by an implicit
scrapping rate defined as the quotient of total depreciation to the total net stock, has
slowly increased during the period 1960-2001. He explains this increase by the increase
of the proportion of the assets having relatively short lifetime comparing to all the other
assets. Thus, the time varying depreciation profile of the aggregated stock of capital is
directly related to the changes in the composition of public investments. These changes
and their consequences on the public capital productivity have been largely studied for
the United States (Gramlich, 1994; Cain, 1997; Fernald, 1999; etc.), mainly because
sufficiently disaggregated data are available in the "Fixed and Reproducible Tangible
Wealth" BEA's database. In this database, the total net stock of government capi-
tal (Federal, State and Local) is divided in two parts, equipment and structures, and
this last component represents roughly 85% of the total stocks. For the structures,
the BEA distinguishes six broad categories9 of assets: Buildings (including residential,
industrial, educational, hospital and other), Highways and Streets, Military facilities,
Conservation and Development, Sewer systems structures, Water supply facilities and
Other structures. The most important categories during the period 1990-1996 are the
buildings (including educational buildings) and the highways and streets (respectively
around 30% and 26% of the total stocks). We can observe on Figure 3 that the part
of these two main components in the total government net stocks has slowly decreased
since the 1970's. At the same time for instance, the part of the net capital stocks corre-
sponding to water and sewer systems has increased. There are numerous reasons which
could explain these modifications in the composition of the public capital stocks of the
United States: the achievement of the great highways projects such as the Interstate
Highway System (Fernald, 1999), the changes in the demography structure which have
affected all the investments related to education (Sturm, 1998) or more originally, the
legal constraints or the federal subvention10 (Hulten and Schwab, 1997).
8Kamps (2004) and Declerq (1994) are the only examples of the use of a time varying depreciation
profile for public capital stocks.
9More precise sub-categories are also available.
10Hulten and Schwab (1997) point out the influence of the Federal Water Pollution Act (1972) and
the Clean Water Act (1977) on the amount of subventions devoted to the water and sewer systems
investments.
9
To the best of our knowledge, there is no similar study devoted to the composition
of public capital stocks in developing countries or even in the major OECD countries
except the United States. However, it is reasonable to assume that, in developing
countries, the structure of public capital stocks has changed over the last decades, even
if these changes are not similar to those observed for the United States. This evolution
of stocks can be pointed out using the evolution of the composition of investments.
There are few studies devoted to the structure of public investments in developing
countries over long period of time. One exception is the recent study proposed by
Calderon, Easterly and Serven (2004) for nine Latin American countries11 over 1980-
1998. Their decomposition is based on six broad categories: road, railways, electricity,
gas, water and telecoms. On Figure 4, the average proportions of these categories in
the total public investments for these nine countries are reported. Even if this structure
of investments cannot be directly compared to the structure of the net stocks of United
States, we still can observe some important differences. For instance, the proportion
of investments in roads in LA increased during the 1980's, while we observe a decrease
of the part of this kind of investments in US during the same period (Fernald, 1999).
The part of investments in railways and electricity in LA decreased quickly at the end
of 1980's and the early of 1990's due to privatisations (Calderon, Easterly and Serven,
2004) and naturally we do not have the same evolution for the US investments during
the same period. These observations confirm our conviction that we cannot use the
structure of public investments / stocks in US or other OECD countries as a proxy
for the developing countries. Consequently, it implies that it is not reliable to use the
same time profile of depreciation for developing countries as for OECD countries since
this time profile is directly related to the composition of stocks. For these reasons, we
propose to estimate a time profile of depreciation rates specific to developing countries12
and based on the changes of the composition of public investments observed in LA.
In order to estimate a time profile of the depreciation rates for the aggregated capital
specific to developing countries, we propose the following definition. The time varying
depreciation rate is defined as a weighted average of the constant depreciation rates
of the six components of public investments defined in Calderon, Easterly and Serven
(2004). This estimate requires two inputs: constant depreciation rates for the six types
of assets used in the decomposition and the weights of these assets in the total stocks.
For the first input, we propose to use the BEA depreciations rates. This is appropriate
insofar as country-specific factors influence service lifetime. In particular, recurrent
infrastructure expenditures for operation and maintenance are generally cut along with
investments, and this decline may accelerate the depreciation of stocks. However, we
do not have information about the potential reduction of the service lifetime of public
investments in developing countries. Moreover, the literature devoted to public capital
stocks does not assert that services lifetime of equipments or structures is shorter in
developing countries than in other countries. The main particularity raised in the case
11These countries are: Argentina, Bolivia, Brazil, Chile, Colombia, Ecuador, Mexico, Peru,
Venezuela, R.B. de.
12So we will assume that the time profile is the same accross countries, as it was done in Kamsp
(2004) or Maddison (1995).
10
of developing countries is that public investments may not create productive capital
(Pritchett, 1996) and not that the depreciation of stocks is more accelerated.
The BEA computes net capital stocks from a detailed division of assets (Bureau of
Economic Analysis, 1999, tables C). Ideally, depreciation profiles should be based on
empirical evidence on used asset prices in resale markets. Such an evaluation is possible
only for some assets (autos, computers for instance). In this case, the profile can be
estimated using prices in the corresponding resale market. For all other assets, geo-
metric profiles are used. For a given type of asset, the estimated rate of depreciation
is derived from the estimates of Hulten and Wykoff (1979, 1981). The depreciation
rate is determined by dividing the declining-balance rate13 by the corresponding as-
sumed service life expressed in years. Hulten and Wykoff obtained that, on average,
the declining-balance rate for producer's durable equipment was 1.65 and that for non-
residential structures was 0.91. Given the BEA classification, we choose the assets that
are the closest to the decomposition of investments proposed by Calderon, Easterly
and Serven (2004). The geometric depreciation rates, the underlying declining-balance
depreciation rates and the services lives used by BEA for these assets are reported in
Table 1. For each type of investment two depreciation rates are proposed by the BEA:
one for equipment and one for structures. The only exceptions are the investments
in roads, gas and water systems for which only structure assets are reported. Taking
into account these information, we compute a weighted average of the rates on struc-
tures and equipment for all the six components of public investments used in Calderon,
Easterly and Serven (2004). The weights are defined by the average part of equipment
assets (respectively structure assets) in the total government net stocks of the United
States over the period 1950-1996. The weights used are then defined by 83.17% for
structures and 16.83% for equipments. The corresponding depreciation rates used for
the six components of public investments are reported in Table 2. The mean of these
depreciations rates is equal to 2.68% which is roughly the same value used by Kamps in
his study of OECD countries until the 1970's. Logically, the lowest values of deprecia-
tion rates are obtained for roads and water systems and the highest value corresponds
to the telecommunication sector where the service lifetime of assets is the shorter.
The second input required to estimate the depreciation rate for the aggregated
stocks is the weight of each type of assets in the total stocks. Let us consider the general
case with m types of assets. For each asset, the capital stock at the beginning of period
t is denoted Kit, with i = 1,..,m and t = 1,..T. The corresponding depreciation rate,
denoted i, is assumed to be constant over time. If we consider a perpetual inventory
method, and if we note Ii,t the gross investment in the ith asset for the current period,
we have:
Ki,t+1 = (1 - i)Kit + Iit i = 1,..,m (1)
Then, the process which generates the total stocks Kt = i=1 Kit is defined as:
Kt+1 =
e X(1
m e Pm
(2)
i=1
13 - i)Kit + Iet
The declining balance rate is equal to the multiple of the comparable straight-line rate.
11
where Iet =
be rewrittenPm a the time varying depreciation rate, denoted t, as:
i=1 Iit denotes the total investments. This process of accumulation can
using
Identifying equations (1) ande(2), it is possibleeto derive the aggregated depreciation
Kt+1 = 1 -et´
³ Kt + Iet e
time profile by the variable as a weighted average of constant specific depreciation rates.
In this case, the weights, denoted Wit, are defined by the proportion of the capital stocks
Kit in the total stocks Kt at time t.
e e Kit
t = Xim
Wit = (3)
Kt
i=1 Xi
m
i=1
approximating the weights Wit by the ratio of investments Iit/Iet. In thisecase, the time
However, we do not have some information about the ratios Kit/Kt since there
e
are no available data on sectoral public capital stocks. A solution would consist in
profile of the depreciation rates would be estimated by:
e
t ' t =
e Xi
m
Iit
(4)
i=1
The sequence of estimated depreciation rates t based on this approximation over
the period 1980-1998 is reported on Figure 5. The ratios Iit/Iet are computed using
eIet
the data of Calderon, Easterly and Serven (2004) for the following six components
of investments: road, railways, electricity, gas, water and telecoms. The constant
depreciation rates i, i = 1,..6, correspond to the values reported in Table 2. We can
observe that the estimated depreciation time profile slowly decreased from 2.74% in
1980 to 2.38% in 1998. This decrease mainly occurred after the end of 1980's. This
stability of the depreciation rates in the 1980's (between 2.75% and 2.80%) and the
slight decrease in the 1990's is clearly visible when the depreciation rates are smoothed.
Figure 5 displays the depreciation rates smoothed by a simple Hodrick and Prescott
(1980) filter. Two values of the smoothing parameter recommended for annual data are
used in order to assess the robustness of our results: the standard value of 100 proposed
notably by Backus and Kehoe (1992), but also a value of 6.25 recommended by Ravn
and Uhlig (2002). Using these parameters, the smoothed time profiles of depreciation
are similar. The smoothed depreciation rate ranges from 2.77% in 1980 to 2.50% in
1998 and we can distinguish two periods: a period of stability in the 1980's and a
period of decrease in the 1990's. Similar conclusions (see Figure 1 in Appendix B.1)
are obtained if we use a local regression on time (or Loess regression, Cleveland and
Devlin, 1988) with an optimal smoothing parameter chosen according to a modified
AIC criterion.
This slight decrease of the depreciation rates is in opposition to the depreciation
profile observed by Kamps (2004) for OCDE countries after 1960. Indeed, Kamps
(2004) determines the depreciation profile for the OECD countries by approximating
an implicit scrapping rate calculated for the United States over the period 1960-2000.
12
He finds that this scrapping rate has tended to rise over time. Consequently he assumes
a constant depreciation rate of 2,50% for the period 1860 to 1960, and an increasing
rate for the period 1960-2001. During this last period, the depreciation rates for OECD
countries are assumed to increase gradually from 2.50% to 4.00%. He explains this in-
crease of depreciation rates by the increasing weight of short lifetime assets. However,
considering the Calderon, Easterly and Serven (2004) decomposition in LA countries
since the 1980's, we observe an increasing weight of assets with relatively long service
lifetime, as roads and water systems (Figure 4). That is why we found that the ag-
gregated depreciation rates for developing countries had decreased. This result does
not exclude the possibility of an increase in the next decades due, for instance, to the
investments in information and communication technologies. At the same time, our
results could be interpreted as an increase in the assets lifetime imposed by a lack of
replacement investments.
One of the major drawbacks of this first approximation comes from the fact that the
weights are defined by the ratios of investments. These ratios reflect the composition of
new equipments or structures, and not the composition of the stock determined by the
past investments. That is why, we propose a second approximation of the weights used
in the definition (3) of the aggregated depreciation rates. Let us consider a perpetual
inventory method for each asset (equation 1) . For t 2, we have:
Kit = X(1
t-2 - i)h Ii,t-1 -h + (1 - i)t-1 Ki1 (5)
h=0
total stock at the beginning of the period t. Then, the weight Wi,teof the ith asset at
where Ki1 denotes the initial stock for the asset i. Let us denote Kt = Pm i=1Kit the
time t in aggregated depreciation rate is defined as:
"X(1
t-2
Wit = - i)h Ii,t-1-h
h=0 + (1 - i)t-1 Ki1#
×"X m t-2 m
- i)h Ii,t-1-h + (6)
i=1 h=0
X(1 X
i=1 (1 - i)t-1 Ki1#-1
Naturally, in this expression the initial conditions {Ki1}m are unknown. Let us
i=1
denote i the ratios of initial value Ki1 to the initial total stock K1 and the ratio of
The weights Wit can be rewritten as:
total public investments to the total public capital stock at the first period, = Ie1/K1.
e
Wit (,) = "X(1 e
t-2 - i)h Ii,t-1 i
-h +
h=0
×"X m t-2 m
- i)h Ii,t-1-h + (7)
i=1 h=0
X(1 µ ¶ÃIe1!X
(1 - i)t-1 Ie1#
i=1 i (1 - i)t-1#-1
13
with = (1,..m). The ratio of public investment to public capital stock is cali-
brated to the average value observed14 for the United States over the period 1950-1996,
i.e. 0,0573. A sensitivity analysis to the value of this parameter will be proposed
with five values of the parameter = {0.01;0.04;0.05;0.06;0.10}. Like the previous
approximation, the public investments Iit are taken from the data of Calderon, Easterly
and Serven (2004) with m = 6 components of investments. The constant depreciation
rates i are taken from Table 2. Finally, the ratios of the initial values of stocks i
are calibrated15 to the mean of the corresponding ratios of investments over the period
1980-1998 for the considered LA countries.
c
The sequence of the corrected depreciation rates t = Pm i=1 i Wit (,) based on
and , the decrease of the depreciation rates is confirmed. However, this decrease
this approximation is reported on Figure 6. For theebenchmark values of parameter
is less important than the decrease observed with the previous approximation. The
depreciation rate ranges from 2.639% in 1998 to 2.707% in 1980. Logically, the profile
of depreciation is smoother than the previous profile computed with ratios of invest-
ments. Consequently, whatever the value of the smoothing parameter, the HP trend
components are similar to the series of estimated depreciations rates. The results are
similar when we consider different values of the ratio of public investments to public
capital stock at the first period. A sensitivity analysis of our results to the values of
= {0.01;0.04;0.05;0.06;0.10} is reportedeon the Figure 7. Given these results, we
c
propose to use our estimated series t and t as benchmark depreciation rates for the
developing countries of our sample.eMore precisely, we consider the smoothed values
of these depreciation rates obtained by a local regression (LOESS) over time (see Ap-
pendix A.1). We prefer this smoothing method to the HP filter, since it is well known
that the HP filter has not good properties at the extreme points of the sample. For the
period before 1980, we assume that the depreciation is constant: its value is fixed to
the first estimated value for 1980, i.e. 2,71% for the corrected depreciation rate t and
2, 77% for the non corrected depreciation rate t. The corresponding depreciation time
profiles are displayed on Figure 8 and the values are reported in Table A3 (Appendix
e ec
B.2). In a next section, we will analyze the sensitivity of our net capital stock estimates
to the depreciation pattern used. Moreover, apart from these quantitative results, we
are convinced that the use of a slight decreasing (or at the most constant) depreciation
time profile for developing countries is justified for the arguments mentioned above.
Given the decomposition of investments observed by Calderon, Easterly and Serven
(2004) for LA countries over the period 1980-1998, we think that the increasing de-
preciation time profile used for OECD countries could not be used as a benchmark for
developing countries.
14A similar value of 0.0666 is founded if we consider that in average the ratio of public capital stocks
to GDP is near to 0.60 and the ratio of public investments to GDP is near to 0.04, as it is observed for
OECD countries
15These values, expressed as a percentage, are the followings. Road: 1 = 14.70, Railways: 2 =
9.60, Electricity: 3 = 46.27, Gas: 4 = 0.53, Water: 5 = 12.97, Telecoms: 6 = 15.89. A analysis of
sensitivity was done with i = 1/6 for all assets results to the same qualitative depreciation profile for
all the considered values of . The maximum value of the depreciation rate was then 2.68% and the
minimum achieved in 1998 was around 2.56%.
14
2.3 Initial Stocks
The second input required in the application of the perpetual inventory method is the
initial public capital stock. When no information about the initial value of the capital
stock is available, one of the possible solutions consists in assuming a null initial stock.
However, in order to limit the consequences of this assumption, we will consider the
same assumptions as those used by Jacob, Sharma and Grawbowski (1997) or Kamps
(2004) for OECD countries. For each country of our panel, an artificial investment
series is constructed since 1880 by assuming that investment increased by 4 percent per
year and finally reaching its observed level at the first date of our sample. This series is
used as input to estimate the initial capital stock for each country. The stock in 1880
is set equal to zero and the depreciation rate used corresponds to the first value of our
estimated series t, i.e. 2.71%. We will analyze in the next section the sensitivity of
our net capital stocks to the choice of initial stock, but we need to remind here that
this influence decreases over time. Indeed, for a time-varying depreciation rate t, the
ec
net public capital stock at time t, denoted Kt, is defined as:
t-1
Kt+1 = X Y
t-1
(1 - t-z) K1 (8)
h=0 "hz=0
-1 (1 - t-z)# It-h + z=0
Y
where Ki1 denotes the initial stock and where t < 1. Consequently, an error on the
value of the initial net capital stock has therefore only a temporary effect.
To sum up, in order to estimate our benchmark net capital stocks by the perpet-
ual inventory method, we made two main assumptions. The first concerns the initial
capital stock. This initial value of the net capital stock is constructed from an artifi-
cial investment series which starts in 1880. The second main assumption concerns the
depreciation pattern. As it was generally the case, we choose a geometric depreciation
pattern. But, we also use a time varying depreciation rate profile, with a slight decrease
after the beginning of the 1980's. This decrease reflects the changes in the composition
of the public investments observed for the LA countries and considered as a benchmark
for our sample of developing countries.
2.4 Net Capital Stock Estimates for Developing Countries
Figure 9 displays the evolution of the ratio of the public net capital stock to GDP for the
26 countries of our panel. The capital stock estimates are expressed in national currency
and valued with national constant prices16. Logically, we found similar conclusions as
those raised for the public investments in the previous section. In order to compare
our benchmark results with those of Kamps (2004a) for OECD countries, the values
of the public capital GDP ratio for four reference years (1975, 1980, 1990 and 1995)
are also displayed in Table 3. The main difference is that we observe an increase of
the average net capital stock ratio for our panel of developing countries, while Kamps
observes a decline of the corresponding ratio for OECD countries. In our sample,
the average ratio increases from 58.6% in 1980 to 63.1% in 1990. When we consider
16The base year is not the same for all countries and corresponds to the base year used for the GDP
deflator. The GDP used is alos expressed in constant prices, WDI code: NY.GDP.MKTP.KN.
15
a balanced panel of 19 countries17, the corresponding values are 59.2% and 64.3%.
During the same period, the estimated ratios for 22 OECD countries decreased from
57.8% to 55.3% (Kamps 2004, Table 2, page 28). At least two factors could explain
this evolution: the first is the depreciation pattern18, the second and most important
is the delay in the decline of public investments in developing countries. As it was
previously mentioned, in our sample the average ratio of public investments to GDP
has declined only after the middle of the 1980s, while in the most of OECD countries
this evolution has occurred in the 1970s. Consequently, the corresponding decrease in
the stock ratios is logically delayed in our sample. More precisely, we do no observe a
decrease as in the OECD countries, but only a drastic decrease in the growth of the
capital to GDP ratio: this ratio increased from 55.5% to 58.6% from 1975 to 1980,
but only from 63.1% to 64.2% from 1990 to 1995. If we consider the balanced panel
of 19 countries, we observe a decrease as severe as that reported by Kamps, since the
ratio declined from 64.3% in 1990 to 61.8% in 1995. In other words, the PIM approach
allows us to date the deceleration of the growth of the public capital stock, on average,
at the beginning of the 1990s, i.e. at least two decades after the deceleration observed
in the United States (Fernald, 1999). Moreover, at the end of the 1990s, the average
capital to GDP ratio in developing countries remained similar, or even superior, to
that observed in OECD countries. However, given the drawbacks of the use of the
PIM for the case of developing countries this conclusion cannot be interpreted as the
diagnostic that, relatively to their GDP, many developing countries have no shortages
of public investments or capital stocks. Our estimates only allow us to point out the
deceleration of the public capital stock growth in 1990s, at least one decade after the
same phenomenon was observed in the OECD countries. This evolution can be viewed
as a consequence of the fiscal adjustments (but this is not the only factor) implemented
in the 1980s. Except if we assume that the efficiency of public investments in creating
public capital dramatically changed over the period 1975 -1995, these observations
based on the PIM remain valid if we consider the true part of public investments
really productive in creating public capital. Nothing allows us to think that any global
improvement in the efficiency of public investments would compensate the decrease of
the ratio of public capital stocks to GDP during the 1990s. Finally, as pointed out by
Kamps (2004a), the main advantage of these benchmark estimates is that they rely on
the same methodology and on homogenous investment data across countries.
These global evolutions hide heterogeneous trends among the countries of our sam-
ple. Between 1975 and 1980, public capital expressed as part of GDP has increased
in 14 countries out of 24. Between 1990 and 1995, this ratio increased only for 10
countries. However, despite these evolutions, there exists a relative stability of the
group of countries with the highest ratios of public capital. If we compare the ranking
according this ratio in 1975 and in 1995, 9 countries are always among the 10 highest
ratios (Table 3). These countries are Botswana, Cameroon, Dominican Republic, Fiji,
17Excluded countries are: Botswana, Burkina Faso, Cameroon, Fiji, Ghana, Mauritius and Zim-
babwe.
18For a given GDP, the decrease of the depreciation rate induces an increase of the capital stock to
GDP ratio. However, the depreciation pattern does not explain a substantial part of the evolution of
our capital stock estimates, as we will see in the section devoted to the robustness of our estimates.
16
Indonesia, Kenya, Morocco, Panama and Tunisia. Only Colombia has been excluded
from this group between 1975 and 1995 and has been replaced by the Venezuela. And,
since the Colombia witnessed an infrastructure investment expansion during the late
1990s, this evolution would not be observed if the reference was taking at the early of
2000s. Thus, these benchmark estimates based on the PIM show a relative stability of
the ranking of the countries according to their ratio of public capital stocks to GDP.
Naturally, the robustness of theses estimates must be evaluated.
2.5 Robustness of the Public Capital Stock Estimates
The drawbacks of the perpetual inventory method are well known. The largest one
is that the use of the PIM requires a variety of assumptions concerning depreciation
patterns and the treatment of an initial stock. That is why we now propose to evaluate
the robustness of our public capital stock estimates to these assumptions as it was
done by Kamps (2004a). For this purpose we will consider four reference countries:
Pakistan, Peru, Philippines and Tunisia. For each assumption used in this section, we
will estimates new net capital stocks series for these four countries. These series will
be compared with our benchmark PIM estimates from the point of view of the levels
but also from the point of view of growth rates.
Concerning the depreciation patterns, two assumptions are made: the first is the
geometric pattern and the second assumption is the use of a time varying depreciation
rate. Firstly, let us consider the influence of time profile of the depreciation rates on
our estimates. As it was shown in the previous section, our aggregated depreciation
rates are estimated using the composition of public investments in LA as a benchmark
for all the developing countries of our sample. Using this method, we found a slight
decreasing time profile reflecting the decline of the part of public investments in assets
having short life services. The influence of these assumptions on our final estimates
must be evaluated. For that, for each reference country two other capital stocks are
estimated with two alternative assumptions on the depreciation time profile. For the
first estimate we consider a constant annual depreciation rate of 4.0% over the period
1970-20002. If we admit that on average the declining-balance rate for equipment is 1.65
and 0.91 for non residential structures, it corresponds to a service lifetime of 41.25 years
and respectively 22.75 years. This values of the depreciation rate is largely superior to
the values considered in our benchmark estimates which range from 2.64% to 2.71%.
For the second estimate, we assume that the depreciation time profile for developing
countries is exactly the same as that estimated by Kamps (2004a) for OECD country.
The author assumes that the depreciation rate gradually increased from 2.50% to 4%
over the period 1960-2001 according to the following formula:
t = ( 2.50%h¡0.025¢41it-2001+41 t < 1960
1 (9)
0.0250 0.04 t 1960
Figure 10 displays the estimates of net public capital stock, expressed in local cur-
rency unit at constant prices, for the alternative assumptions on the depreciation time
profile. Logically, the net stocks evaluated with the constant depreciation rate of 4.00%
17
are always inferior to our benchmark estimates, but both dynamics are quite similar as
we can observe on Figure 11 where the corresponding growth rates are displayed. This
result is logical since our benchmark depreciation rates do not vary before 1980 and
are only slightly decreasing after this date. Then, the corresponding dynamics of our
estimates is quite similar to that obtained with a constant depreciation rate. The stocks
evaluated with the depreciation rates used by Kamps (2004a) are also quite similar to
our benchmark estimates at the early of the 1970s. Indeed, given the formula used by
Kamps, the aggregated depreciation rate for OECD countries is equal to 2.80% in 1970,
while our benchmark estimated depreciation rate at the same date is 2.71%. But in
this case, the difference between both depreciation rates has increased over time. That
is why the difference between the corresponding estimated stocks has also increased
over time. In 2000, the difference between both estimated stocks reached 9.31% of the
benchmark stock for Pakistan, 10.65% for Peru, 10.19% for Philippines and 9.85% for
Tunisia. Given the potential sources of errors due to the perpetual inventory method
in the valuation of the estimates of capital stocks, such a difference can be considered
as negligible. Secondly, as mentioned above, the use of a geometric pattern is very
general in the literature devoted to the net capital stock estimates, and not only for
public capital stocks. Besides, Hulten and Wyckoff (1981) show that if different capital
goods are lumped together, the geometric depreciation rate gives a reasonable statis-
tical approximation of the true depreciation rate. For these reasons, we do not report
the sensitivity analysis of our estimates to other methods of depreciation (straight-
line, hyperbolic etc.). Kamps reports the results of this type of sensitivity analysis for
his net public capital stock estimates for OECD countries. He shows that hyperbolic
and geometric depreciation result in quite similar capital stock estimates whereas the
estimates based on straight-line depreciation differ considerably in levels and also in
dynamics.
Finally, we propose an evaluation of the robustness of our estimates to the choice
of the initial condition. Naturally, the estimation of the initial capital stocks, which
consist in accumulated past investments, presents substantial problems, since no in-
formation is available about public investments over long period of time, even for the
OECD countries. In our study, the benchmark estimates are computed with an initial
stock defined as the accumulation of a pseudo public investment series. This approach
requires several assumptions: (1) an averaged growth rate for the pseudo investment
series (2) an initial condition for the stock or equivalently a starting date for the pseudo
series and (3) a depreciation rate over this period. As it was explained in the previous
section, we consider an averaged growth rate of investments equal to 4.00% for all the
countries of our sample, a depreciation rate equal to 2.71% and a starting date in 1880,
i.e. twenty years after the initial date considered by Kamps for OECD countries. In
order to determine the sensitivity of our estimates to these assumptions, we consider
two alternative sets of assumptions. In the first set, we assume that the pseudo series
starts in 1920. Consequently, this assumption implies a lower initial stock for all the
countries of the panel. In the second set of assumptions, the starting date remains
unchanged, but we consider a heterogeneous growth rate of investments to compute
the pseudo investment series. For each country, the growth rate used is set equal to the
historical growth observed over the period 1970-2000. Figures 12 and 13 display the
18
corresponding estimates of the net capital stocks and the growth rates of net capital
stocks. For three reference countries (Pakistan, Philippines and Tunisia) the use of the
national growth rate of public investment to build the pseudo series of past investments
leads to an initial capital stock higher than the benchmark ones. In these cases, the
difference between the two initial stocks expressed as percentage of the benchmark ini-
tial stock ranges from 20.9% for Philippines to -29.62% for Tunisia. However these
important differences rapidly decrease with time. At the end of the sample, the dif-
ference between the estimates of capital stocks is 1.17% for Pakistan, 0.70% for Peru,
2.39% for Philippines and -2.41% for Tunisia. Besides, the dynamics of growth rates of
capital stocks are quite similar under various assumptions concerning the time profile
of depreciation rates.
These observations show that, at least for these reference countries, the dynamics of
our benchmark PIM estimates of net public capital stocks are relatively robust to some
changes in the time depreciation profile or in the value of the initial stocks. Moreover,
it is obvious that if the public capital stocks in developing countries were overvalued, it
would be due rather to a mis-measurement of the part of monetary investments which
were really transformed in productive capital rather than due to a mis-measurement of
the depreciation rates or of the initial stocks.
3 Public Investment and Efficiency
The main drawbacks of the PIM method are well known. The two largest are the
fact that this estimation method requires a variety of assumptions concerning service
lives and depreciation patterns and an evaluation of initial stock. Firstly, we have
proposed a sensitivity analysis of our estimates to these assumptions, and secondly
these drawbacks are not specific to the developing countries. For both reasons, the net
capital stocks estimated by the PIM can be used to compare the productivity of public
capital in developing countries and in OECD countries. However, there is another
problem considered as more specific19 to developing countries: it is the efficiency of
public investments in creating capital.
Indeed, Summers and Heston (1991), Pritchett (1996) or Canning (1998) state that
the same investment flows in different countries may have very different effectiveness in
actually producing capital, due to the differences in the efficiency of the public sector
and differences in the price of capital. If the investment project is carried out by the
public sector, actual and economic costs (defined as the minimum of possible costs
given available technology) may deviate. So, monetary investment in infrastructure
may be a very poor proxy of the amount of public capital actually produced. For
instance, Pritchett suggests that in a typical developing country less than 50 cents of
capital were created for each public dollar invested. He concludes that "one of the deep
difficulties of development may well be that even when public capital is productive it
19We assert that this issue is specific to developing countries since only studies devoted to these
countries evoked this problem. To the best of our knowledge, in the huge litterature devoted to the
public capital in the United States, there is no study that points out this fact. But, this is not a proof
of the specificity that the inefficiency of public investments is specific in developing countries.
19
may be difficult to create this capital in the public sector" (Pritchett, 1996, page 1).
Given this potential inefficiency of public investments, the use of monetary investment
flows may introduce systematic errors in stock estimates. In particular, it implies that
public capital stock series constructed with the PIM will tend to be overvalued.
In this section, we firstly intend to evaluate this inefficiency using an original
methodology and secondly to estimate net stocks only based on the productive compo-
nent of investments.
3.1 Public Investment Efficiency: a Non Parametric Approach
The idea that a dollar's worth of public investment spending often does not create a
dollar's worth of capital can be expressed as follows:
Kt+1 = (1 - )Kt + f (It) (10)
where Kt denotes the real public capital stock at time t and I denotes the corresponding
gross public investment. The function f (It) represents the efficiency of public invest-
ments to generate new capital. We assume that efficiency is constant over time, i.e.
the function f (.) does not vary with time. Since the relative efficiency may vary with
the amount of investments, such an assumption is not very restrictive. The perpetual
inventory method assumes that this function is always defined as the identity function:
f (It) = It (11)
On the contrary if we assume that, according to Pritchett (1996), in a typical developing
countries only a certain part of investments is used to create the public capital, the
function f (.) satisfies the following constraints:
0 f (It) < It (12)
f (0) = 0 (13)
The fact that f (It)is strictly inferior to It reflects the inefficiency of public investments
in creating capital. The more different from zero is the distance |f (It) - It |, the less
appropriate to estimate public capital stocks is the PIM method. If this efficiency
function were known, it would be possible to estimate the "productive" component of
investments at each date and consequently to estimate the genuine net public capital
stocks by the PIM, as presented in the previous section. It is important to note that it
is not the PIM itself which is a problem for the developing countries. It is the fact that
the PIM is based on the total monetary flows of gross investments which is problematic.
In this context, the main issue is that it is very difficult to find a functional form for
this function f (.). The simplest way would be to assume that productive investments
are strictly proportional to the total investments. If, as it stated by Pritchett, less than
50 cents of capital were created for each public dollar invested, we would have:
f (It) = It (14)
20
with < 1/2. The main drawback of this specification, is that we assume that the
relative efficiency of public investments is constant in the amount of the investments.
It implies that a huge spending going to white elephant has relatively the same efficiency
(or inefficiency) as a little project of infrastructure for which the control can be stricter
and more direct. Since no natural specification of the efficiency function f (.) can
be justified a priori, another solution consists in estimating this function by a non
parametric method for a country or for a sub-group of developing countries and in
taking this estimate as a benchmark for all the countries of our sample. More precisely,
the aim of our non parametric approach is not to evaluate the relative efficiency of public
investments in the 26 countries of our panel, but to estimate the efficiency function for
a sub-group of countries and to consider this estimated functional form (and not the
corresponding efficiency) as benchmark for all the countries. For this purpose, three
ingredients are required: (1) a series a public investments (2) the depreciation rates
(or its time profile) and (3) a series of public capital net stocks effectively available in
the reference countries. The last element is the most difficult to find. Indeed, in order
to precisely estimate the public capital stock in developing countries, the efficiency
function is required, and in order to estimate this function it is necessary to have the
series of public capital.
In this study, we propose to estimate the efficiency function of the public invest-
ments using the physical measure of infrastructure as a proxy20 of the public capital
stock effectively used in the developing countries. Recently, in order to study the
link between the public infrastructure spending and the physical infrastructure stocks,
Calderon, Easterly and Serven (2004) used the same approximation21 for the Latin
American countries. They used an ARDL specification with the sectoral monetary
investments flows (for instance, the investments in the road sector) and the physical
measures of the corresponding infrastructure, expressed as growth rate (for instance,
the number of kilometers of paved road). Their regressions in panel models with or
without fixed effects point out the statistical link between the past investment fluctu-
ations and infrastructure level. "The conclusion from these empirical experiments is
that infrastructure investment is a robust predictor of subsequent changes in the physical
infrastructure stock across countries and over time". (Calderon, Easterly and Serven,
page 62).
Here, our approach is different. Firstly, we do not intend to estimate a long term
multiplier between the past investments and the variation of the physical infrastructure
stocks in an ARDL specification. We intend to estimate the functional form f (.),
which may be not linear, between the current monetary investments and the monetary
variation of the current stock i.e.. Kt+1-(1 - )Kt. Secondly, our study is only devoted
to the link between the public investments and the public capital. On the contrary
Calderon, Easterly and Serven analyze simultaneously the total private investments,
20This idea which consists to estimate the efficiency function by comparing the variations of the
physical measures of the capital to the monetary investments flows has been evoked by Pritchett
(footnote 31, page 36, Pritchett 1996), when he considered the first data collected by Canning and Fay
(1995) on the physical measures of the transport infrastructure.
21Calderon, Easterly and Serven (2004), tables 2.6 and 2.7, starting from the page 61.
21
the private investments and the public investments. Consequently, in order to estimate
the link between the public investments and the public capital using physical measures,
it is required to consider countries (1) for which there are available data about sectoral
public investments over a long enough period of time, and (2) for which the private
investments in the reference sectors are negligible comparing to the public investments.
To the best of our knowledge, only the Calderon, Easterly and Serven's database about
nine countries of the Latin America allow an enough detailed decomposition of the
public investments for a long period of time that allows us to establish a correspondence
with the physical measures of the Caning's database. That is why we choose two
reference LA countries from our panel, Colombia and Mexico. For these two countries
we compare the investments flows to the physical measure variations over the period
1981-1995. We use a decomposition of investments in four sectors for which there
exist corresponding physical measures in the Caning's database. These sectors are the
following: electricity, telecommunication, roads and railways.
One of the problems of this decomposition is that it considers sectors for which
private investments may be important. As in the majority of Latin America countries,
in these two reference countries, the proportion private investments versus public of
the investments in infrastructure deeply changed during the period 1981-1995. In order
to avoid this problem, we consider only a period of time for which the part of private
investments in the total investments is insignificant. More precisely, for each sector,
the sample used for estimation starts in 1981 and ends at the dates reported in the
Table 4. These ending dates have been chosen such that during the considered period
of time, the amount of the private investments in the total investments of the specified
sectors never exceeds 15%. So during these period, we can use the infrastructure as a
proxy of the public capital for the considered sector. Generally, these dates correspond
to the reform dates pointed out by Calderon, Easterly and Serven (2004).
The second problem concerning this sectoral decomposition is the correspondence
between the investments categories and the infrastructure physical measures proposed
by Canning. To measure the stock of infrastructure in the electricity sector we consider
the electricity-generating capacity expressed in million of kilowatts. For the telecom-
munication sector, two types of measures can be used: the number of telephone lines
or the number of telephone main lines. We decide to use the second measure, which
was notably considered in the ARDL specification estimated by Calderon, Easterly and
Serven (2004). Two measures are also possible for the roads sector: the number of road
kilometers or the number of paved road kilometers. For this sector, we decide to use
the measure which allows us to cover better the estimation period for each country.
Thus, for Mexico we use the number of road kilometers while for Colombia we use the
number of paved road kilometers. Finally, to measure the investments in the railways
sector, the number of the rail kilometers was used.
Then, the problem consists in estimating the functional form of the public invest-
ments efficiency function. This function relates on the one hand the monetary flows of
investments expressed in million US dollars and on the other hand the public capital
stocks measured of in the same monetary unit. But, we have only physical measures of
these stocks. Our methodology is then the following. Let us assume that at a certain
22
date t, for a sector j = 1,..N, the capital stocks expressed in monetary units is defined
by the following relation:
Kjt = vjt Xjt (15)
where Xjt denotes the physical measure of the capital in to the sector j and vjt rep-
resents the monetary value of one physical unit of fixed capital. For instance, if we
consider the road sector, Xjt represents the total number of road kilometers effectively
created at the date t and vjt represent the monetary value of one kilometer of road
calculated at the date t, expressed in the same monetary unit as the corresponding
investments. The parameter j represents the depreciation rate of the assets of the jth
sector. For the four reference sectors, we consider the depreciation rate computed in
the previous section (see Tables 2 and 4). We note Kj,t+1 the net increase of the capital
stock: Kj,t+1 = Kj,t+1 - (1 -ej)Kjt
e (16)
We assume that the efficiency function of the public investments is specific to each
sector: fj (.) denotes represents the efficiency function associated to the jth sector. Our
objective is to estimate the functions fj (.) defined as:
where Ijt represents the public investments in the sector j and j, with 0 jt
Kj,t+1 = fj jtIjt¢
e ¡ j = 1,..N (17)
1, denotes the specific part of these sectoral investments which actually correspond
to the assets considered in the Canning's database. For instance, if we consider the
electricity sector, we can state that a part of the investments in this sector is allocated
for something else than the increase of the electricity generating capacity (security
investments, investments made to preserve the natural environment, for instance). This
part of public investments does not correspond to unproductive investments. The
parameters j only measures the inadequacy between our sectoral decomposition of
investments and the physical asset considered in the Canning database. If the function
fj (.) is homogenous of degree , then it can be rewritten in the following way:
vej,t+1Xj,t+1 - (1 - jv)vejt X = fj (Ijt)
j,t j,t (18)
j,t
vej,t = (19)
Except the function fj (.), only the valuations vj,t and the proportions j are un-
known. In order to evaluate them, one of the possible solutions is to use the results
of previous studies devoted to the monetary cost of one road kilometer, of one rail
kilometer, etc. But, there are no available data concerning the long term evolution of
these costs. Canning and Fay (1995) or Canning and Bennathan (2000) provide some
evaluations of these costs for developing countries, but their data are available only
for one baseline year. Moreover, even if we have these information, it is impossible to
identify the parameters vej,t if the proportions j are unknown.
For these reasons, we consider the following methodology. We estimate a sequence of
i.e. to the PIM for which one invested dollar increases the capital of exactly one dollar.
values of vej,t in order to get a situation as close as possible to the full efficiency situation,
23
Thus, we will consider the valuations of stocks which are the more favorable to the
public investments efficiency. Consequently, we mainly take the risk to over-evaluate
this efficiency. More precisely, we know that, if PIM is exact, the function fj (.) is
defined by fj (z) = z. Then, the corresponding sequence of values vej,t is defined by:(20)
We assume that vej,t has a geometric evolution:
Kj,t+1 = vej,t+1Xj,t+1 - (1 - j)vej,tXj,t = Ijt
e
vej,t = v (1 + )t (21)
This hypothesis allows us to take into account the inflation of the costs associated
to the construction of one physical infrastructure unit. So, we try to determine the
parameters (v,), which, in the presence of the historical observations Xjt and Ijt, will
following program:
give us the valuation dynamics vej,t compatible with the PIM. For that, we solve the
1
(v, ) = ArgMin
b b (22)
{v,}R+2 T Xhv
T
t=1 (1 + )t+1 Xj,t+1 - (1 - j) v (1 + )t Xj,t - Ijti2
under the constraints:
v (1 + )t [(1 + ) Xj,t+1 - (1 - j) Xj,t] Ijt t = 1,..,T (23)
These T constraints impose that, for all the considered dates, the monetary in-
creases of stocks, taking into account the depreciation, cannot be more important than
the investments. We exclude the case when the investments are "more" efficient than
those considered in the PIM, which means the case when one invested dollar produces a
ate a sequence of estimated values for vej,t. These coefficients allow us to approximate
the increase of stocks according the formula:
capital of more than one dollar. Using the estimated parameters vb and , we can recre-
b
the example of theeFigure 14 and, after this step, we can estimate the efficiency curve.
T corresponding values observed for Ijt, we can report all the couples ofe points as on
Under our assumption, this stock variation is the closest tobthe variation implied
Kj,t+1 = vb (1 +
e b b)t+1Xj,t+1 -(1 -j)vb (1+ )tXj,t (24)
by the PIM, i.e. Kj,t+1 = Ijt. Using the sequence of T realizations of Kj,t+1 and the
(25)
Kj,t+1 = Ijt, the less appropriate is the estimation of the capital stocks by the PIM
using the total flows of investments. On the other hand, if the estimated function
fbj (Ijt) coincides with the straight line of 450 slope, then the PIM is a good method to
e For a specified level of investments Ijt, the more the value of fbj (Ijt) is far from
Kj,t+1 = fbj (Ijt)
e
approximate the stocks. So, our approach allows to evaluate the form of the efficiency
function and, also, to determine if the relative efficiency (i.e. the ratio of the productive
investments in the total investments) is constant or decreases with the investments level.
24
For each reference sector, the efficiency function is estimated using a local LOESS
regression (Cleveland and Devlin, 1988). The principle of this regression is that a lo-
cal polynomial is estimated for every reference point, using the points situated in the
neighborhood of this reference point. The dimension of these neighborhoods is deter-
mined by a smoothing parameter which is defined by the rapport between the number
of points included in the neighborhood and the total number of observations. If the
smoothing parameter is equal to 1, then all the points of the sample are used to compute
the polynomial parameters and, consequently, we determine a linear adjustment line.
The less the value of this parameter, the more volatile the estimated function22. In
this study, the smoothing parameter was chosen according to a modified AIC criterion
firstly proposed by Hurvich and Simonoff (1998).
In order to assess the quality of our approach, we propose to estimate the efficiency
function of the public investments in road and highways for the United States over the
period 1950-1992. We use the series of public investments (Federal, State and Local) in
road and highways, valued at historical costs expressed in million of US dollars. These
series are taken from the BEA23. For the corresponding physical measures, we consider
the total road kilometers (Canning, 1998). Figure 15 displays the estimated efficiency
function and the corresponding 95% confidence interval. The values of the smoothing
parameter and other statistics of the regression are reported in Table 5. We can observe
that the estimated function is relatively close to the straight line of 450 slope. For a
low level of investments, the estimated efficiency function is statistically not different
to the identity function. Consequently, for the United States, our approach does not
show an important discrepancy between investments and the value of the capital stocks.
The PIM based on total investments provides a good proxy of the public capital stocks
effectively available.
When the same methodology is applied for the case of our two reference developing
countries, the results are very different. Figure 16 displays the estimated efficiency
functions for the sector of electricity, road and telecoms in Colombia. Undoubtedly,
these estimated functions are largely below the straight line of 450 slope. It appears that
the sector where the public investments are the more efficient is the telecommunication
sector. Hoverer, these comparative results must be very carefully used. Given the
data availability and the necessity to end the sample before the dates of reform, our
sectoral samples are very reduced: 1980-1993 for the power sector, 1980-1994 for the
telecommunication sector and 1980-1991 for roads. The short length of our samples
implies that the estimates of the sectoral efficiency functions are relatively imprecise. In
order to obtain more precise estimates, we propose an estimate of the global efficiency
we estimate the global efficiency function f (.) by a LOESS regression.
function³based on the data of the three sectors. More precisely, we report all the
couples Kj,t+1,Ijt´ obtained for the sectors j = 1,2 and 3. Given these 37 couples,
e
Kj,t+1 = f (Ijt)
Fixed and Reproducible tangible Wealth, 1999, Code: EYBGAIT08.
e j = 1,2,3 (26)
22See Yatchew (2003) for more details about non parametric estimations.
23
25
Figure 17 displays the estimated efficiency function for Colombia and Figure 18 displays
the same function for Mexico. Both estimated functions are strikingly similar. They
show that the "productive" component of public investments is largely overvalued when
the PIM is used. Two results are particularly interesting here. Firstly, the estimated
function is near a straight line. The optimal smoothing parameter according to the
modified AIC criteria is equal to 1 for both countries (see Table 5). This conclusion
is robust to the choice of another information criterion as the general cross validation
(GCV) function. This result implies that the estimated function can be approximated
by a simple linear functional form, i.e. f (It) = It where , with 0 < < 1, denotes
an efficiency parameter according to the expression proposed by Pritchett (1996). In
other words, the relative efficiency, defined as the ratio of the "productive" investments
to the total amount of investments, is constant. Secondly, the coefficient of the linear
regression of Kj,t+1 to Ijt is equal to 0.38 in the case of Colombia and 0.40 in the case
of Mexico. According to this evaluation, one peso of public investments creates around
0.40 pesos of public capital in our reference sectors. We need to emphasize that this
e
evaluation is based on artificial data, i.e. Kj,t+1, built so as to minimize the distance
between it and the PIM reference. Our conclusions based on these non parametric
estimates are similar to that of Pritchett who found that "a variety of calculations
e
suggest that in a typical developing country less than 50 cents of capital were created
for each public dollar invested".
3.2 Net Capital Stock Estimates based on the Efficiency Function
We now propose to estimate new series of net capital stocks only based on the productive
component of public investments. For that, we assume that the results obtained for our
two reference countries in the previous section can be generalized to all the developing
countries of our panel. More precisely, we only assume that the linear form of the
efficiency function can be used as a benchmark for all our countries. This assumption
does not imply that the public investments in the 26 developing countries have the same
efficiency in creating capital. It only implies that the relative efficiency is constant for
all the countries. If we note Kit the net stock for the ith country, i = 1,..N and Iit the
corresponding public investment, we assume that:
Ki,t+1 = (1 - t)Kit + iIit (27)
where parameters i satisfy 0 i < 1. The depreciation rates t are identical to
those estimated in the first section (see Table A3, Appendix B.2). As in the case of
our benchmark PIM estimates, the initial stocks are derived from cumulated artificial
investment series. For each country, the investment series starts in 1880, increases by
4 percent a year and finally reaches its observed level at the first date of our sample.
Logically, this observed level is defined by the product iIi1. Consequently, the choice
of the efficiency parameter i affects the initial stock too.
The national efficiency parameters cannot be estimated for all the countries of our
panel. As it was mentioned in the previous section, a non parametric or even a para-
metric estimate of the efficiency function requires long series of sectoral investments.
26
To the best of our knowledge, such information is not available for almost all the coun-
tries of the panel. Moreover, it would be naive to believe that it is possible to evaluate
the relative efficiency of the public sector in the developing countries using only such
an aggregated approach. Many other factors have to be taken into account. For these
reasons, we prefer to propose a sensitivity analysis of the estimates of the productivity
of public capital stocks based on various values of the efficiency parameter. We will
consider 3 cases: a case of low efficiency with i = 0.2, a case of medium efficiency
i = 0.4 and a case of high efficiency i = 0.6. For each country, these three cases
are considered and three corresponding estimates of the net public capital stocks are
computed. Figure 19 displays the evolution of the corresponding ratios of the public
net capital stocks to GDP for the 26 countries of our panel.
4 Conclusion
In this paper, we provide various estimates of the government net capital stocks for
26 developing countries over the period 1970-2001. The use of the public capital stock
concept instead of the use of the concept of infrastructure allows comparing our es-
timates to those generally done for OECD countries and in particular for the United
States. Besides, our series of public capital stocks are homogeneous and comparable:
they are estimated according to the same methodology and they are based on the same
definition of public investments. To our knowledge, this is the first attempt to propose
such comparable series of public capital for developing countries. These internationally
comparable annual estimated stocks would constitute a complementary solution to the
use of public investment flows and to the use of physical measures of infrastructure when
one comes to evaluate the productivity of the public capital formation in developing
countries. It may not be the best solution and each approach has its drawbacks and its
advantages. The use of investment flows implies that the effects of public investment on
growth or private factors productivity are independent of the level of the correspond-
ing stocks. On the contrary, the use of stocks instead of investments flows necessarily
increases the potential measurement errors. Finally, the use of physical measures has
the advantage to precisely measure the stock of infrastructure really available in the
economy. However, it does not allow taking into account the quality of infrastructure.
Besides the list of productive infrastructures considered is largely dependent to the
data availability and may exclude some important components of public investments
(educational building etc.) considered as productive in other studies and covered by
the notion of public capital stocks.
If the usefulness of public capital stock series for developing countries is obvious,
as it is the case for OECD countries (Kamps, 2004), the methodology used to estimate
these series must be adapted. It is recognized that in a typical developing country, an
important part of public investment is inefficient in creating capital. Consequently, the
perpetual inventory method, when it is based on the totality of monetary investment
flows, may overvalue the public stocks. In this paper, we propose to assess this ineffi-
ciency of public investment in creating public capital by direct comparison of physical
measures of infrastructure available to the monetary investment flows. With a non
27
parametric approach, we show that in two LA countries the relative inefficiency, i.e.
the ratio of "productive" investments to the total investments, is constant. It would
be naive to hope to estimate the relative efficiency of public investments for the 26
countries of our panel given the data availability. However, our partial conclusion on
the functional form of the efficiency of public investment in LA countries may be used
for all other countries in order to make a sensitivity analysis of our stock estimates.
More precisely, we propose various estimates of stocks based on various assumptions
of the efficiency parameter. Given the lack of precision of the estimates of the relative
efficiency of public investments, one must be very careful with our corresponding es-
timated levels of stocks. However, these series can be used as inputs in order to test
the sensibility of various measures or policy analysis (on the impact of public capital
stocks on growth, poverty indices, etc.) to the efficiency of public investment in creating
capital.
28
A References
Aschauer D.A. (1989a), "Is Public Expenditure Productive?", Journal of Monetary
Economics 23, pp. 177-200.
Aschauer D.A. (1989b), "Does Public Capital Crowd Out Private Capital?", Journal
of Monetary Economics 24(2), pp. 171-188.
Backus, D.K. and Kehoe, P. (1992), "International Evidence on the Historical
Properties of Business Cycles", American Economic Review, 82, 4, pp. 864-888.
Berndt E.R. and Hansson B. (1992), "Measuring the Contribution of Public In-
frastructure Capital in Sweden", Scandinavian Journal of Economics, 94, pp. 151-168.
Bureau of Economic Analysis (1999), Fixed Reproducible Tangible Wealth in the
United Sates, 1925-1994, Washington, U.S. Department of Commerce.
Bureau of Economic Analysis (2003), Fixed Assets and Consumer Durable Goods
in the United Sates, 1925-1997, Washington, U.S. Department of Commerce.
Cain L.P. (1997), "Historical Perspective on Infrastructure and US Economic Devel-
opment", Regional Science and Urban Economics, 27, pp. 117-138
Calderon, C., Easterly, W. and Serven, L. (2004), "Latin America's Infrastruc-
ture in the Era of Macroeconomic Crises", in The Limits of Stabilization, Easterly W.
and Serven L. eds., The World Bank.
Canning D. (1998), "A Database of World Infrastructure Stocks, 1950-1995", The
World Bank Economic Review, 12, 529-547.
Canning D. (1999), "Infrastructure's Contribution to Aggregate Output", World Bank
Policy Research Working Paper, 2246, Washington D.C.
Canning, D. and Bennathan, E. (2000), "The Social Rate of Return on Infrastruc-
ture Investments", Mimeo, World Bank.
Canning, D. and Pedroni, P (1999), "Infrastructure and Long-Run Economic
Growth", CAER II Discussion Paper 57, Harvard Institute for International Devel-
opment.
Cleveland, W. and Devlin, S. (1988), "Locally Weighted Regression: An Approach
to Regression Analysis by Local Fitting", Journal of American Statistical Association,
83, pp. 596-610.
Declercq C. (1994), "Evaluation du Capital Public Français : 1950-1991. Méthodolo-
gie et Construction de la Série", Document de travail ERASME.
Devarajan S., Swaroop V. and Zou H.F. (1996), "The Composition of Public
Expenditure and Economic Growth", Journal of Monetary Economics, 37(1), pp. 313-
344.
Easterly W. and Serven L. (2004), The Limits of Stabilization, The World Bank.
Evans P. and Karras G. (1994), "Is Government Capital Productive? Evidence
from a Panel of Seven Countries", Journal of Macroeconomics, 16(2), pp. 271-279.
29
Fernald J.G. (1999), "Roads to Prosperity? Assessing the Link Between Public
Capital and Productivity", The American Economic Review, 89(3), pp. 619-638
Ford R. and Poret P. (1991), "Infrastructure and Private-Sector Productivity",
OECD Economic Studies, vol. 0, 17, pp. 63-89.
Garcia-MilA T., McGuire T.J. and Porter R.H. (1996), "The Effect of Public
Capital in State Level Production Functions Reconsidered", The Review of Economics
and Statistics, 78, pp. 177-179.
Gramlich E.M. (1994), "Infrastructure Investment: a Review Essay", Journal of
Economic Literature, 32, pp. 1176-1196.
Hodrick, R.J. and Prescott C.E. (1980), "Postwar U.S. Business Cycles: an Em-
pirical Investigation", Discussion Paper 451, Carnegie Mellon University.
Holtz-Eakin D. (1994), "Public-Sector Capital and the Productivity Puzzle", The
Review of Economics and Statistics, 76, pp. 12-21.
Holzl W. and Leish R. (2004), "Estimates of Capital Stocks and Capital Productivty
in Austrian Manufacturing Industries, 1978-1994", Working Paper, Vienna University
of Economics and Business Administration.
Hulten C.R and Peterson G.E. (1984), "The Public Capital Stock: Needs, Trends,
and Performance", The American Economic Review, 74(2), pp. 166-73.
Hulten C.R. and Schwab R.M. (1997), "A Fiscal Federalism Approach to In-
frastructure Policy", Regional Science and Urban Economics, 27, pp. 139-159
Hulten, C.R. and Wykoff, F.C. (1979), "Tax and Economic Depreciation of Ma-
chinery and Equipment: A Theoretical Appraisal", in Economic Depreciation of the
U.S. Capital Stock: A First Step, Washington,DC: U.S. Department of the Treasury,
Office of Tax Analysis.
Hulten, C.R. and Wykoff, F.C. (1981), "The Estimation of Economic Depreciation
Using Vintage Asset Prices", Journal of Econometrics, 15, pp. 367-396.
Hurlin C. (2000), La Contribution Productive des Infrastructures Publiques. Analyses
Positives et Normatives, Phd Dissertation, University of Paris I Panthéon Sorbonne.
Hurvich, C. M., and Simonoff, J. S. (1998), "Smoothing Parameter Selection in
Nonparametric Regression Using an Improved Akaike Information Criterion" Journal
of the Royal Statistical Society B, 60, pp. 271 -293.
Jacob, V., Sharma, S.C. and Grabowski, R., (1997), "Capital Stock Estimates
for Major Sectors and Disaggregated Manufacturing in Selected OECD Countries",
Applied Economics, 29, pp. 563-579.
Kamps C. (2004a), "New Estimates of Government Net Capital Stocks for 22 OECD
Countries, 1960-2001", IMF Working Paper 04/07.
Kamps C. (2004b), The Dynamic Macroeconomic Effects of Public Capital: Theory
and Evidence for OECD Countries, Springer.
Lynde C. and Richmond J. (1992), "The Role of Public Capital in Production",
The Review of Economics and Statistics, 74, pp. 37-44
30
Maddison A. (1995), "Standardized Estimates of Fixed Capital Stock: A Six Country
Comparison", in Explaining the Economic Performance of Nations, ed. Maddison,
Edwaerd Elgar.
Munnell A.H. (1990a), "Why has Productivity Declined? Productivity and Public
Investment", New England Economic Review, F.R.B. of Boston, pp. 3-22.
Munnell A.H. (1990b), "How Does Public Infrastructure Affect Regional Economic
Performance?", New England Economic Review, 11-32.
Morrisson C.J. and Schwartz A.E. (1996a), "State Infrastructure and Productive
Performance", The American Economic Review, 86, pp. 1095-1111.
Nadiri M.I. and Mamuneas T.P. (1994), "The Effects of Public Infrastructure and
R&D Capital on the Cost Structure and Performance of US Manufacturing Industries",
Review of Economics and Statistics, 76, pp. 22-37.
OECD (2001a), Measuring Capital: OECD Manual, OECD, Paris
OECD (2001b), Measuring Productivity: OECD Manual, OECD, Paris
Otto G.D. and Voss G.M. (1998), "Is Public Capital Provision Efficient?", Journal
of Monetary Economics, 42, pp. 47-66.
Pritchett, L. (1996), "Mind your P's and Q's. The cost of Public Investment is Not
the Value of Public Capital", Policy Research Working Paper 1660, The World Bank.
Ram R. and Ramsey D. (1989), "Government Capital and Private Output in the
United States", Economic Letters, 13, pp. 223-226.
Ravn M.O. and Ulhig H. (2002), "On Adjusting the HP-Filter for the Frequency of
Observations", Review of Economics and Statistics, 84, 2, pp. 371-376.
Romp, W. and De Haan, J. (2005), "Public Capital and Economic Growth: A
critical Survey", Mimeo University of Groningen.
Shah A. (1992), "Dynamics of Public Infrastructure: Industrial Productivity and
Profitability", The Review of Economics and Statistics, 74, pp. 28-36.
Sturm J.E. (1998), Public Capital Expenditure in OECD Countries: the Causes and
Impact of the Decline in Public Capital Spending, Edward Elgar.
Sturm J.E. and De Haan J. (1995), "Is Public Expenditure Really Productive?",
Economic Modelling, 12, pp. 60-72.
Sturm J.E. and De Haan J. (1998), "Public Capital Spending in the Netherlands:
Developments and Explanations", Applied Economics Letters, 5, 5-10
Sturm J.E., Kuper G.H. and De Haan J. (1996), "Modelling Government Invest-
ment and Economic Growth on a Macro Level: A Review", CCSO Series, 29.
Yatchew A. (2003), SemiParametric Regression or the Applied Econometrician, Cam-
bridge University Press.
World Bank (1994), Annual Report, Washington DC.
World Development Indicators (2004), World Bank, Washington DC.
31
Table 1. BEA Depreciation Rates by Assets
Categories Type BEA assets Depr. Rate Service Life
Road Structures Highways and Street 0.0202 45
Railways Equipment Railroad equipment 0.0589 28
Structures Railroad replacement track 0.0275 38
Structures Other railroad structures 0.0166 54
Electricity Equipment Electrical distribution 0.0500 33
Equipment Other electrical equipement 0.1834 9
Structures Electric light and power 0.0211 40
Gas Structures Gas 0.0237 40
Water Structures Water systems 0.0152 60
Structures Sewer systems 0.0152 60
Telecoms Equipment Other equipment 0.1375 12
Structures Telecommunications 0.0237 40
Notes: The service lifetime is expressed in years. Structures and equipments correspond to non
residential structures and equipments . Bureau of Economic Analysis (2003), table C, page M-31.
Table 2. Depreciation Rates by Categories
Categories Equipment Structures Depr. Rate
Road 0.0202 0.0202
Railways 0.0589 0.0275 0.0328
Electricity 0.050 0.0211 0.0260
Gas 0.0237 0.0237
Water 0.0152 0.0152
Telecoms 0.1375 0.0237 0.0429
Note: For each asset, the depreciation rate is defined as a
weighted average of the corresponding rates used for equipments
and structures by the BEA (Bureau of Economic Analysis, 2003,
table C, page M-31).
32
Table 3. Government Net Capital Stocks in 26 Developing Countries
(As percentage of GDP, Constant Price)
1975 1980 1990 1995
Ratio Rank Ratio Rank Ratio Rank Ratio Rank
Botswana 62.6 9 66.5 11 54.7 17 67.2 9
Burkina Faso 14.3 24 16.3 26 23.7 25
Cameroon 67.5 8 63.4 13 103.4 3 114.7 2
Colombia 69.7 6 61.5 14 54.2 18 51.2 17
Costa Rica 59.6 13 59.2 15 63.7 9 52.4 15
Dominican Republic 94.9 1 90.5 3 94.7 5 96.2 4
Fiji 74.7 5 80.1 7 93.0 6 86.6 6
Ghana 59.6 12 70.3 8 57.8 15
India 20.1 23 21.3 25 22.3 26 22.2 24
Indonesia 80.4 3 82.0 4 96.1 4 88.7 5
Kenya 69.4 7 64.5 12 63.7 10 68.8 8
Malaysia 52.6 16 48.8 17 61.0 12 53.3 14
Mauritius 69.1 10 54.8 16 55.4 12
Mexico 47.9 18 44.8 19 58.0 14 57.9 11
Morocco 79.5 4 117.6 1 119.3 1 132.2 1
Pakistan 51.1 17 46.7 18 40.1 19 43.6 18
Panama 87.4 2 94.6 2 86.7 8 65.7 10
Paraguay 28.1 19 24.6 23 26.8 24 28.8 23
Peru 27.1 20 37.5 20 62.6 11 52.0 16
Philippines 24.5 21 25.2 22 36.0 21 42.1 19
Thailand 61.0 11 51.5 16 37.1 20 36.5 21
Tunisia 62.3 10 80.7 6 115.3 2 113.1 3
Turkey 56.9 15 69.4 9 59.2 13 55.3 13
Uruguay 21.0 22 24.0 24 33.2 22 31.8 22
Venezuela, RB 59.5 14 81.1 5 92.0 7 82.1 7
Zimbabwe 32.1 21 32.3 23 41.8 20
Average 55.5 58.6 63.1 64.2
(22.4) (25.7) (28.6) (28.7)
Average (19 Countries) 55.4 59.2 64.3 61.8
(22.7) (27.1) (29.1) (29.1)
Note: The standard deviations of the 26 (or 19 for the balanced panel) individual ratios are in parenthesis.
33
Table 4. The Ending Dates of the Sample
and the Depreciation Rates
Electricity Telecom Roads Rails
Depreciation Rate
Rate 2.60% 4.29% 2.02% 3.28%
Ending Dates of Samples
Colombia 1993 1994 1994 (1990)
Mexico 1998 (1995) 1990 1989
Note: For the case of the road sector in Colombia, the available
data end in 1990. For the case of Mexico, in the electricity sector, the
Canning's data end in 1995. Data concerning the public investments in
Colombia for the railways sector are not available. The data concerning
the road infrastructures for the Mexico before the data of reform (1989)
are also not available.
Table 5. Loess Regressions
Fit Summary Colombia Mexico United States
Number of Observations 37 35 42
Smoothing Parameter 1 1 0.63
AICC 11.95 12.49 17.46
Degree of Local Polynomials 1 1 1
Points in Local Neighborhood 37 35 26
Residual Standard Error 221.70 290.82 3468.58
Notes. The AICC corresponds to the corrected AIC criterion proposed by
Hurvich and Simonoff (1998). The optimal value of the smoothing parameter
is chosen to optimize this criterion.
34
Figure 1. Ratios of Government Capital Expenditure to GDP, 1973-1997
19 Developing Countries, Average and Standard Deviation (In Percentage)
6.5 4.5
6.0 4.0
5.5 3.5
5.0 3.0
4.5 2.5
4.0 2.0
3.5 1.5
74 76 78 80 82 84 86 88 90 92 94 96
Average (Left Axis) Deviation (Right Axis)
Source: The series are issued from the World Development Indicators (2004)
35
Figure 2. Ratios of Government Capital Expenditure to GDP
in 26 Developing Countries (As a percentage of GDP)
14 7 10
6
12 8
5
10 4 6
8 3 4
2
6 2
1
4 0 0
70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00 02 70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00 02 70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00 02
Bostwana Burkina Faso Cameroon
7 6 10
6 5
8
5 4
6
4 3
4
3 2
2 1 2
70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00 02 70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00 02 70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00 02
Colombia Costa Rica Dominican Republic
8 8 2.4
7 2.2
6
6 2.0
5 4 1.8
4 1.6
2
3 1.4
2 0 1.2
70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00 02 70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00 02 70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00 02
Fiji Ghana India
14 7 16
6 14
12
12
5
10 10
4
8 8
3
6
6
2 4
4 1 2
70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00 02 70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00 02 70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00 02
Indonesia Kenya Malaysia
Source: World Development Indicators (2004)
36
Figure 2. (continued) Ratios of Government Capital Expenditure to GDP
in 26 Developing Countries (As a percentage of GDP).
5.5 6 20
5.0
5
16
4.5
4
4.0 12
3
3.5
8
2
3.0
2.5 1 4
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Mauritius Mexico Morocco
5 10 5
8 4
4
6 3
3
4 2
2
2 1
1 0 0
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Pakistan Panama Paraguay
5 5 14
12
4 4
10
8
3 3
6
4
2 2
2
1 1 0
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Peru Philippines Thailand
14 7 2.8
6
12 2.4
5
10 2.0
4
8 1.6
3
6 1.2
2
4 1 0.8
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Tunisia Turkey Uruguay
10 4.0
3.5
8
3.0
6
2.5
4
2.0
2
1.5
0 1.0
70 75 80 85 90 95 00 70 75 80 85 90 95 00
Venezuela, RI Zimbabwe
Source: World Development Indicators (2004)
37
Figure 3. Composition of the Total Government Net Capital Stocks.
United States, 1950-1996. (As percentage of Total Stock).
24 32 32
22 31
30
30
20
29 28
18
28 26
16
27
24
14 26
12 25 22
50 55 60 65 70 75 80 85 90 95 50 55 60 65 70 75 80 85 90 95 50 55 60 65 70 75 80 85 90 95
Equipment Building Highways
8.0 6.5 4.0
6.0
7.5
3.8
5.5
7.0
5.0 3.6
6.5
4.5
3.4
6.0
4.0
5.5 3.5 3.2
50 55 60 65 70 75 80 85 90 95 50 55 60 65 70 75 80 85 90 95 50 55 60 65 70 75 80 85 90 95
Conservation Sewer Water
Sources: Series are issued from Fixed Reproducible Tangible Wealth, U.S Bureau of Economic Analysis (1999).
Figure 4. The Structure of the Total Public Investments in Nine LA Countries
1980-1998. (As Percentage of Total Investments)
70 2.5 25
60 2.0 20
50 1.5 15
40 1.0 10
30 0.5 5
20 0.0 0
80 82 84 86 88 90 92 94 96 98 80 82 84 86 88 90 92 94 96 98 80 82 84 86 88 90 92 94 96 98
Electricity Gas Railways
30 28 40
25 24
30
20 20
20
15 16
10
10 12
5 8 0
80 82 84 86 88 90 92 94 96 98 80 82 84 86 88 90 92 94 96 98 80 82 84 86 88 90 92 94 96 98
Road Telecoms Water
Source: Series are issued from Calderon, Easterly and Serven (2004).
38
Figure 5. Depreciation Rates t and HP Filtered Depreciation Rates. 1980-1998
0.029 e
0.028
0.027
0.026
0.025
0.024
0.023
80 82 84 86 88 90 92 94 96 98
Depreciation Rates HpTrend_625
HpTrend_100
Notes: The series labelled "Depreciation" corresponds to the estimated depreciation rates. The series labelled
"HpTrend_100" corresponds to the trend component of these depreciation rates issued from an HP filter with a
smoothing parameter equal to 100. The series labelled "HpTrend_625" is the same trend component but with
a smoothing parameter equal to 6.25 (Ravn and Uhlig, 2002).
Figure 6. Depreciation Rates t and HP Filtered Depreciation Rates. 1980-1998
0.0272 ec
0.0270
0.0268
0.0266
0.0264
0.0262
80 82 84 86 88 90 92 94 96 98
Depreciation Rates HpTrend_625
HpTrend_100
Notes: The series labelled "Depreciation" corresponds to the estimated corrected depreciation rates. The series
labelled "HpTrend_100" correspond to the trend component of these depreciation rates issued from an HP filter
with a smoothing parameter equal to 100. The series labelled "HpTrend_625" is the same trend component but
with a smoothing parameter equal to 6.25 (Ravn and Uhlig, 2002).
39
Figure 7. Depreciation Rates t. Sensitivity Analysis to the Parameter .
0.0272 ec
0.0270
0.0268
0.0266
0.0264
0.0262
0.0260
80 82 84 86 88 90 92 94 96 98
Alpha = 0.01 Alpha = 0.05 Alpha = 0.10
Alpha = 0.04 Alpha = 0.06
Note. The series labelled "Alpha_0.01" corresponds to the estimated values of the corrected depreciation rates
based on a calibrated value of = 0.01. The same label is reproduced for the values 0.01, 0.04, 0.05, 0.06
and 0.10.
Figure 8. Estimated Depreciation Rates for Developing Countries
(LOESS Smoothing Method). 1960-1998
0.0280
0.0275
0.0270
0.0265
0.0260
0.0255
0.0250
60 65 70 75 80 85 90 95
Corrected Depr. rates Not Corrected Depr. rates
40
Figure 9. Real Government Net Capital Stocks in 26 Developing Countries
1970-2002 (As a percentage of Real GDP)
90 140
80
80 120
70 60 100
60
40 80
50
60
20
40
40
30 0
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Botswana Burkina Faso Cameroon
90 90 120
80 80
100
70 70
80
60 60
50 50 60
40 40
40
30 30
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Colombia Costa Rica Dominican Republic
110 90 90
100 80
80
90 70
70
80 60
70 60 50
60 40
50
50 30
40
40 20
30 30 10
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Fiji Ghana India
120 90 90
80 80
100
70 70
80
60 60
60 50 50
40 40
40
30 30
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Indonesia Kenya Malaysia
Source: Author's estimates
41
Figure 9. (continued) Real Government Net Capital Stocks in
26 Developing Countries. 1970-2002 (As a percentage of Real GDP)
90 90 140
80 80 120
70 70 100
60 60
80
50 50
60
40 40
40
30 30
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Mauritius Mexico Morocco
90 100 90
80 90 80
80 70
70
70 60
60
60 50
50
50 40
40 40 30
30 30 20
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Pakistan Panama Paraguay
90 90 90
80 80 80
70 70
70
60 60
60
50 50
50
40 40
30 30 40
20 20 30
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Peru Philippines Thailande
140 90 90
80
120 80
70
100 70
60
60 50
80
40
50
60 30
40
20
40
30 10
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Tunisia Turkey Uruguay
100 90
90 80
80 70
70 60
60 50
50 40
40 30
30 20
70 75 80 85 90 95 00 70 75 80 85 90 95 00
Venezuela, RI Zimbabwe
Source: Author's estimates
42
Figure 10. Government Net Capital Stocks in Billions LCU (Constant Prices)
for Alternative Depreciation Profiles
Pakistan Peru
350 70
300 60
250 50
200 40
150 30
100 20
50 10
70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00 02 70 75 80 85 90 95 00
Corrected Depr. Rates OECD Depr. Rates Corrected Depr. Rates OECD Depr. Rates
Constant Depr. Rates 4% Constant Depr. Rates 4%
Philippines Tunisia
400 20
16
300
12
200
8
100
4
0 0
70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00 02 70 75 80 85 90 95 00
Corrected Depr. Rates OECD Depr. Rates Corrected Depr. Rates OECD Depr. Rates
Constant Depr. Rates 4% Constant Depr. Rates 4%
43
Figure 11. Growth Rates of Government Real Net Capital Stocks for Alternative
Depreciation Profiles (in Percentage)
Pakistan Peru
10 20
8
15
6
10
4
5
2
0
0
-2 -5
70 75 80 85 90 95 00 70 75 80 85 90 95 00
Corrected Depr. Rates OECD Depr. Rates Corrected Depr. Rates OECD Depr. Rates
Constant Depr. Rates 4% Constant Depr. Rates 4%
Philippines Tunisia
16 16
14 14
12
12
10
10
8
8
6
6
4
2 4
0 2
70 75 80 85 90 95 00 70 75 80 85 90 95 00
Corrected Depr. Rates OECD Depr. Rates Corrected Depr. Rates OECD Depr. Rates
Constant Depr. Rates 4% Constant Depr. Rates 4%
44
Figure 12. Government Net Capital Stocks in Billions LCU (Constant Prices)
for Alternative Assumptions on the Initial Stock
Pakistan Peru
350 70
300 60
250 50
200 40
150 30
100 20
50 10
70 75 80 85 90 95 00 70 75 80 85 90 95 00
Benchmark Estimate National Investment G.R. Benchmark Estimate National Investment G.R.
Assumption 4 p.c. since 1920 Assumption 4 p.c. since 1920
Philippines Tunisia
500 20
400 16
300 12
200 8
100 4
0 0
70 75 80 85 90 95 00 70 75 80 85 90 95 00
Benchmark Estimate National Investment G.R. Benchmark Estimate National Investment G.R.
Assumption 4 p.c. since 1920 Assumption 4 p.c. since 1920
45
Figure 13. Growth Rates of Government Real Net Capital Stocks
for Alternative Assumptions on the Initial Stock (In Percentage)
Pakistan Peru
10 20
8 15
6 10
4 5
2 0
0 -5
70 75 80 85 90 95 00 70 75 80 85 90 95 00
Benchmark Estimate National Investment G.R. Benchmark Estimate National Investment G.R.
Assumption 4 p.c. since 1920 Assumption 4 p.c. since 1920
Philippines Tunisia
14 20
12
16
10
8 12
6 8
4
4
2
0 0
70 75 80 85 90 95 00 70 75 80 85 90 95 00
Benchmark Estimate National Investment G.R. Benchmark Estimate National Investment G.R.
Assumption 4 p.c. since 1920 Assumption 4 p.c. since 1920
46
Figure 14. Estimated Efficiency Function
K(t+1)
PIM Case : F(I) = I
·
·
·
·
· ·
· ·
·
Estimated Efficiency Function
·
·
·
· I(t)
Figure 15. Non-Parametric Estimated Efficiency Function of Public Investments
in Streets and Highways . United States, 1951-1992 (US$ million, Historical Cost)
Notes: For each estimated point, the corresponding 95% confidence limits are represented by a sign "+"
47
Figure 16. Non-Parametric Estimated Efficiency Functions of Sectoral
Public Investments. Colombia, 1980-1994 (US$ million, current prices)
Electricity Telecoms
Roads
48
Figure 17. Non-Parametric Estimated Efficiency Function of Total
Public Investments. Colombia, 1980-1994 (US$ million, current prices)
Notes: For each estimated point, the corresponding 95% confidence limits are represented by a sign "+"
Figure 18. Non-Parametric Estimated Efficiency Function of Total
Public Investments. Mexico, 1980-1994 (US$ million, current prices)
Notes: For each estimated point, the corresponding 95% confidence limits are represented by a sign "+"
49
Figure 19. Real Government Net Capital Stocks in 26 Developing Countries
1970-2002 (As a percentage of Real GDP)
Botswana Burkina Faso Cameroon
50 20 80
40 15 60
30 10 40
20 5 20
10 0 0
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Alpha = 0.2 Alpha = 0.4 Alpha = 0.6 Alpha = 0.2 Alpha = 0.4 Alpha = 0.6 Alpha = 0.2 Alpha = 0.4 Alpha = 0.6
Colombia Costa Rica Dominican Republic
50 50 70
60
40 40
50
30 30
40
20 20
30
10 10
20
0 0 10
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Alpha = 0.2 Alpha = 0.4 Alpha = 0.6 Alpha = 0.2 Alpha = 0.4 Alpha = 0.6 Alpha = 0.2 Alpha = 0.4 Alpha = 0.6
Fiji Ghana India
70 50 16
60 14
40
12
50
30 10
40
20 8
30
6
10
20 4
10 0 2
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Alpha = 0.2 Alpha = 0.4 Alpha = 0.6 Alpha = 0.2 Alpha = 0.4 Alpha = 0.6 Alpha = 0.2 Alpha = 0.4 Alpha = 0.6
Indonesia Kenya M alaysia
70 50 50
60
40
40
50
30
40 30
20
30
20
20 10
10 10 0
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Alpha = 0.2 Alpha = 0.4 Alpha = 0.6 Alpha = 0.2 Alpha = 0.4 Alpha = 0.6 Alpha = 0.2 Alpha = 0.4 Alpha = 0.6
Source: Author's estimates
50
Figure 19. (continued) Real Government Net Capital Stocks in
26 Developing Countries. 1970-2002 (As a percentage of Real GDP)
M auritius Mexico M orocco
50 40 100
35
40 80
30
30 25 60
20 20
40
15
10 20
10
0 5 0
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00 02
Alpha = 0.2 Alpha = 0.4 Alpha = 0.6 Alpha = 0.2 Alpha = 0.4 Alpha = 0.6 Alpha = 0.2 Alpha = 0.4 Alpha = 0.6
Pakistan Panama Paraguay
35 70 30
30 60 25
25 50 20
20 40 15
15 30 10
10 20 5
5 10 0
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Alpha = 0.2 Alpha = 0.4 Alpha = 0.6 Alpha = 0.2 Alpha = 0.4 Alpha = 0.6 Alpha = 0.2 Alpha = 0.4 Alpha = 0.6
Peru Philippines Thailand
40 30 50
25
40
30
20
30
20 15
20
10
10
10
5
0 0 0
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Alpha = 0.2 Alpha = 0.4 Alpha = 0.6 Alpha = 0.2 Alpha = 0.4 Alpha = 0.6 Alpha = 0.2 Alpha = 0.4 Alpha = 0.6
Tunisia Turkey Uruguay
80 50 25
40 20
60
30 15
40
20 10
20
10 5
0 0 0
70 75 80 85 90 95 00 70 75 80 85 90 95 00 70 75 80 85 90 95 00
Alpha = 0.2 Alpha = 0.4 Alpha = 0.6 Alpha = 0.2 Alpha = 0.4 Alpha = 0.6 Alpha = 0.2 Alpha = 0.4 Alpha = 0.6
Source: Author's estimates
51
Figure 19. (continued) Real Government Net Capital Stocks in
26 Developing Countries. 1970-2002 (As a percentage of Real GDP)
Venezuela, RI Zimbabwe
70 30
60
25
50
20
40
15
30
10
20
10 5
70 75 80 85 90 95 00 70 75 80 85 90 95 00
Alpha = 0.2 Alpha = 0.4 Alpha = 0.6 Alpha = 0.2 Alpha = 0.4 Alpha = 0.6
Source: Author's estimates
52
B Appendix
B.1 Loess Regressions of Depreciation Rates
Table A1. Loess Regression of t over Time
Fit Summary e
Number of Observations 19
Smoothing Parameter 0.921
AICC -13.49
Degree of Local Polynomials 1
Points in Local Neighborhood 17
Residual Sum of Squares 5.70e-6
Residual Standard Error 0.00059
Notes. The AICC corresponds to the corrected
AIC criterion proposed by Hurvich, Simonoff, et Tsai
(1998).Here, the optimal value of the smoothing parame-
ter is chosen to optimise this criterion.
Figure 20: Loess Regression of Depreciation Rates over Time. (95% Confidence
Limites are represented by symbols + and -) e
t
53
Table A2. Loess Regression of t over Time
Fit Summary ec
Number of Observations 19
Smoothing Parameter 0.605
AICC -13.49
Degree of Local Polynomials 1
Points in Local Neighborhood 11
Residual Sum of Squares 1.63e-8
Residual Standard Error 3.35e-5
Notes. The AICC corresponds to the corrected
AIC criterion proposed by Hurvich, Simonoff, et Tsai
(1998).Here, the optimal value of the smoothing parame-
ter is chosen to optimise this criterion.
Figure 21: Loess Regression of Depreciation Rates over Time. (95% Confidence
Limites are represented by symbols + and -) ec
t
54
B.2 Depreciation Rates
Table A6. Depreciation Rates
ec
t e#
t
t 1980 2.707 2.771
1981 2.704 2.772
1982 2.701 2.771
1983 2.699 2.771
1984 2.696 2.769
1985 2.694 2.767
1986 2.692 2.764
1987 2.691 2.761
1988 2.690 2.756
1989 2.689 2.742
1990 2.687 2.727
1991 2.685 2.705
1992 2.680 2.679
1993 2.675 2.652
1994 2.668 2.624
1995 2.661 2.595
1996 2.654 2.566
1997 2.647 2.537
t 1998 2.640 2.508
55