WPS4589 Policy ReseaRch WoRking PaPeR 4589 Infrastructure and Economic Growth in East Asia Stéphane Straub Charles Vellutini Michael Warlters The World Bank East Asia and Pacific Sustainable Department Policy Unit April 2008 Policy ReseaRch WoRking PaPeR 4589 Abstract This paper examines whether infrastructure investment significant effect for all infrastructure variables in the has contributed to East Asia's economic growth using context of a production function study. This leads us to both a growth accounting framework and cross-country conclude that results from studies using macro-level data regressions. For most of the variables used, both the should be considered with extreme caution. The Authors growth accounting exercise and cross-country regressions suggest that infrastructure investment may have had the fail to find a significant link between infrastructure, primary function of relieving constraints and bottlenecks productivity and growth. These conclusions contrast as they arose, as opposed to directly encouraging growth. strongly with previous studies finding positive and This paper--a product of the Operations and Policy Unit, East Asia and Pacific Sustainable Department--is part of a larger effort in the department to examine the relationship between infrastructure and economic growth. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at stephane. struab@ed.ac.uk, charles.vellutini@ecopa.fr, or mwarlters@worldbank.org. 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 views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Infrastructure and Economic Growth in East Asia Stéphane Straub, Charles Vellutini, and Michael Warlters University of Edinburgh. Email: stephane.Straub@ed.ac.uk Toulouse School of Economics (Arqade) and ECOPA. Email: charles.vellutini@ecopa.fr World Bank 1. Introduction Policy-makers in developing East Asia see infrastructure investment as an essential determinant of growth.1 The two fastest-growing economies in the region, China and Vietnam, are investing around 10 percent of GDP in infrastructure, and even at that rate they are struggling to keep pace with demand for electricity and telephones, and to install major transport networks. Hopes for a significant contribution to growth in the Greater Mekong countries ­ Laos, Cambodia, Thailand, Vietnam, Myanmar, and China ­ are centered on plans for greater integration of transport and energy markets. Since its election in late 2004, the new Indonesian government has made infrastructure a national priority, seeking to restore investment to its pre-crisis level of 5-6 percent of GDP. The current emphasis on infrastructure draws its inspiration from East Asia's economic history, including the experience of countries such as Japan, South Korea, Malaysia and Taiwan, China, which also made large investments in infrastructure. East Asia's accumulation of infrastructure stocks has outpaced infrastructure investment in other regions (Table 1). And East Asia's economic growth has outpaced the growth of other world regions. Between 1975 and 2005, East Asia's GDP increased ten-fold; South Asia's GDP increased five-fold; and all other regions' economies grew by factors of between two and three.2 For most policy-makers this is no coincidence. Table 1: Growth of GDP and Infrastructure Stocks 1995 levels as multiples of 1975 levels GDP Electricity Roads Telecoms East Asia 4.8 5.9 2.9 15.5 South Asia 2.6 4.4 2.5 8.2 Middle East & North Africa 1.8 6.1 2.1 7.2 Latin America & Caribbean 1.8 3.0 1.9 5.1 OECD 1.8 1.6 1.4 2.2 Pacific 1.7 2.0 4.3 Sub-Saharan Africa 1.4 2.6 1.7 3.9 Eastern Europe 1.0 1.6 1.2 6.9 GDP ­ PPP constant 2000 international $; Electricity - MW of generating capacity; Roads ­ km of paved road; Telecoms ­ number of main lines. See Annex 1 for construction. Sources: World Development Indicators and Canning (1998) 1See ADB, IBRD, WB, JICA, (2005). 2Difference in GDP (PPP) in constant 2000 dollars between 1975 and 2005. 2 But academics aren't so sure. Perhaps it is East Asia's growth success that has driven the high rate of infrastructure investment, rather than the other way around. In the neoclassical growth model, exogenous shocks, such as new technology, increase the rate of return to capital, inducing investment. Investment increases the stock of capital, thereby reducing the rate of return to capital and restoring equilibrium at the initial capital-labor ratio and a higher level of output.3 Within this framework, if infrastructure is merely another form of capital with decreasing returns, infrastructure investment does not "cause" long-term growth, it is an inevitable consequence of growth, but the sources of growth must be found elsewhere. Decreasing returns to infrastructure investment can certainly be observed. For example, electricity supply capacity that exceeds demand growth provides a poor return on investment, as several countries found in the aftermath of the 1997 financial crisis when economies and electricity demand contracted. And most of East Asia's infrastructure investment has occurred as a reaction to emerging constraints. So there are certainly arguments that infrastructure has the same properties as assumed in the neoclassical framework for other forms of capital. But the neoclassical growth model also assumes that investment responds automatically to changes in rates of return. In fact, most infrastructure services are not provided in freely functioning markets. Government regulation, market power and externalities mean that infrastructure services are rarely provided at prices that represent the cost of inputs or their marginal social value. And infrastructure investments are dominated by government decision-making (e.g. public investment) and regulatory constraints (e.g. spatial planning, environmental considerations, etc). If the link between high rates of return and investment is blocked, the economy will not grow in accordance with the neoclassical model's predictions. Complete non-responsiveness of infrastructure investment could be a partial explanation for differences in observed long-run growth rates across countries.4 Mere differences in the speed with which infrastructure investment responds to infrastructure constraints would only affect the speed with which economies return to the long-run equilibrium growth path following a shock, and would not determine countries' long-run growth rates. But for 3See Barro and Sala-i-Martin (2005). 4It is assumed that infrastructure services are strongly complementary with modern technologies ­ that is other forms of capital investment cannot substitute for infrastructure services. 3 practical policy purposes, such "transitory" growth rates are just as important as long-run growth. In a developing economy with chronic under-supply of infrastructure, transitory growth could conceivably last for decades. Following this line of argument, infrastructure policies might play a role in explaining East Asia's relative growth success if East Asia is more effective than other regions in relieving infrastructure constraints as they emerge. A small piece of evidence to this effect may be seen in the results of enterprise surveys (Table 2), which indicate that new connections are provided to firms more quickly and that service interruptions are lest costly in East Asia than in most developing regions. Table 2: Impact of Infrastructure Shortages on Firms Region Electricity Value lost to power Water connection Mainline telephone connection delay outages (% of sales) delay (days) connection delay (days) (days) East Asia & Pacific 21 2.6 18 16 Europe & Central Asia 15 3.0 9 16 Latin America & Caribbean 34 4.1 35 36 Middle East & North Africa 62 4.3 44 49 South Asia 49 7.4 29 50 Sub-Saharan Africa 38 5.9 42 54 OECD 10 2.3 -- 9 Source: The data are derived from World Bank Investment Climate Assessments, and reported at www.enterprisesurveys.org (last visited January 10, 2008) However, this could again reflect causality running the other way round, as economies with stronger growth have readily available resources to address such bottlenecks as they become apparent. Beyond the neoclassical growth framework, endogenous growth theory envisages instances where an aggregate economy may exhibit increasing returns to scale, notwithstanding the presence of diminishing or constant returns to individual factors.5 If infrastructure stocks play a role in the realization of these economies of scale, infrastructure policy has a role in determining long-run growth. An important feature East Asia's infrastructure history has been the construction of major transport links between cities. Korea's Seoul-Pusan highway built in the 1960s, Malaysia's road network built in the 1970s and 1980s, China's rail network and more recent expressways 5See Aghion and Howitt (1998) and Barro and Sala-i-Martin (2005) for a review of endogenous growth models. 4 development, and Vietnam's Hanoi-Ho Chi Minh City and Hanoi-Haiphong highways have all enlarged and integrated domestic markets, as well as providing the logistical connections for access to ports and international markets. Further investment in these transport networks may not give the same boost to productivity, but it is possible that the larger markets they create facilitates the exploitation of economies of scale within firms, the production of more specialized goods and services, and better and more specialized skills matches between employers and workers. That is, notwithstanding the presence of diminishing returns to infrastructure investment, the creation of infrastructure networks could contribute to the rate of innovation and technological advance in the economy, and thereby lift the long-term growth rate. An alternative possible source of ongoing growth may lie in knowledge externalities. Cities play an important role in facilitating the exchange of ideas and innovation, and hence advancing the technological frontier. To the extent that infrastructure services affect the efficiency of cities and the effectiveness with which knowledge is shared, infrastructure services may influence the rate of productivity growth.6 Moreover, this raises the question of whether infrastructure investment should be directed in priority to large urban areas or to lagging regions. It has been hypothesized (Williamson, 1965), that poor countries would go first through a process of concentration, industrialization and regional divergence, in which infrastructure investment is if anything following development, but that, as congestion in cities becomes too important, a reversed process of deconcentration and regional convergence occurs, which could be sustained by regional infrastructure investment. If these linkages are important, understanding the dynamic of cities should play a particularly important role in analyzing the sources of East Asia's growth. Overall, however, the evidence on the link between urbanization, infrastructure and growth is still very limited. East Asia is one of the least urbanized regions in the world. But its rate of urbanization is one of the fastest and the East Asian mega-cities are comparably large and more densely populated. Average urban densities in East Asia range from 10,000 to around 15,000 persons per sq km ­ about double the urban densities of Latin America; triple those of Europe; and ten times those of US cities. On the Williamson's hypothesis, some corroborating evidence has been found for Korea (see Henderson, Shalizi and Venables, 2001), but more work is still due to guide policies.7 6See Henderson (2005) on urbanization and growth. 7See Straub (2008) for a detailed discussion of economic geography and urbanization issues in the context of infrastructure policy. 5 While not a channel that has been greatly explored in modern growth theory, it is plausible that growth is enhanced in countries with lower poverty, all else equal. Poverty reduction could serve to increase market size (e.g. greater disposable income), enhance labor productivity (e.g. health improvements), and enhance innovation through improved human capital (e.g. less poor populations might invest more in education; there may be less scope for innovation in an agrarian society, etc.). If such linkages are important, ensuring that all sections of the population are provided with infrastructure could indirectly boost growth by reducing poverty. It is notable that East Asia has been more successful in providing rural access to all-weather roads than other developing regions. Access to roads has been shown in numerous studies to have a significant effect on rural poverty (Jacoby, 2000; Gibson and Rozelle, 2003). Table 3: Proportion of rural population living within two kilometers of an all-weather road Sub-Saharan Africa 30 Middle East & North Africa 34 Latin America & Caribbean 38 South Asia 58 Europe & Central Asia 75 East Asia 94 Source: Roberts et al (2006). Theoretical speculation on the relationship between infrastructure and growth should be tested against empirical observations. Examining 80 econometric specifications from 30 studies using macro-level data, Straub (2007) reports a significant positive effect of infrastructure on output or growth in 56 percent of specifications, no significant effect in 38 percent, and a significant negative effect in 6 percent. Among the studies that do find positive effects there is wide variation in their estimated magnitude. There are several possible reasons for the variation in empirical results. It seems quite likely that the effects of infrastructure investment do, indeed, vary from location to location, and across different stages of economic development. A further source of variation is the theoretical framework used. Straub (2007) observes that a positive effect of infrastructure on growth is more likely to be detected in studies based on a production function than studies using cross-country regressions. The empirical literature frequently fails to set out the theoretical issues that are being tested so that results may not be strictly comparable, a number of methodological problems are either not considered or cannot be addressed with macro- 6 level data, and above all, aggregate data are simply not adequate to address the important policy issues. To illustrate this, our paper examines whether infrastructure investment has, indeed, contributed to East Asia's economic growth using both a growth accounting framework and cross-country regressions. Our results are then contrasted with the results of Seethepalli, Bramati, and Veredas (2007), who use a production function specification to examine the impact of infrastructure on East Asia's growth. With all three methodologies focused on the same region and the same time-frame, any significant findings that recur across methodologies would shed light on whether infrastructure investment has indeed been a cause of economic growth in East Asia. Two main conclusions emerge. First of all, for most of the variables used both the growth accounting exercise and cross-country regressions fail to find a significant link between infrastructure, productivity and growth. When they do, they produce rather contradictory conclusions, as growth accounting indicates no contribution of infrastructure to productivity in the richer countries (South Korea and Singapore), and some contribution in the relatively poorest countries (of telecommunications in Indonesia and Philippines, and of roads in Thailand), while cross-country growth regressions tend to indicate that the effects are generally negative for low-income countries and positive only for the high-income ones. Second, these conclusions contrast strongly with those of Seethepalli, Bramati and Veredas (2007), who find positive and significant effects for all infrastructure variables in the context of a production function study. This leads us to conclude that results from studies using macro-level data should be considered with extreme caution. Given that macroeconomic data give only limited support to the notion that infrastructure investment has driven growth in East Asia, we conclude by speculating on other aspects, in particular the idea that infrastructure investment may have had the primary function of relieving constraints and bottlenecks as they arose. The structure of the paper is as follows. Section 2 presents the growth accounting exercise. Section 3 turns to cross-country growth regressions. Finally, Section 4 discusses the results, compares them to other related studies and concludes. 7 2. Growth accounting 2.1. Methodology Standard growth accounting The formal framework of growth accounting is the production function (1) Y = A.F(K, L) , where Y is aggregate GDP, A is the time-varying total factor productivity (TFP) and K and L are respectively (total) capital and labor. Taking logs and differentiating with respect to time yields (2) Y& A& FL.L L& FK .K K& = + + . Y A F L F K Assuming that marginal factor productivities equal factor prices, we get the standard formula for growth accounting, where the growth of TFP is computed as the residual between the growth of GDP and the growth of factors: (3) A& Y& = -SL L&-SK K& . A Y L K In this equation SL and SK are therefore the respective observed shares of income. Importantly, (3) is typically not implemented through econometric estimation but rather through direct calculation: all the variables on the right-hand side are observed. As reported in Barro and Sala-i-Martin, (3) has been used in many country-specific studies with the objective of calculating TFP growth. 8 Growth accounting with infrastructure Assume, as in Hulten et al. (2005) that infrastructure (denoted X in the following equations) influences output through two channels. First, it impacts TFP through (4) A = A(X ) = A.X ~ where A is the « true » TFP and is the elasticity of A with respect to X. Here, infrastructure ~ raises output without any payments by firms for infrastructure services. This channel captures the externality aspect of infrastructure. Second, infrastructure can enter the production function as an additional production factor: (5) Y = A.X.F(K~, L, X ) . ~ where K is the stock of non-infrastructure capital. ~ The presence of infrastructure as one more factor reflects its market-mediated impact, whereby firms pay for infrastructure services. This leads to: Y& A ~& K~& (6) = + + SX X& X&+ SL L&+ SK~ Y A~ X X L K~ where S X is the share of GDP that accrues to market-mediated infrastructure and SK~ the share of revenue that accrues to non-infrastructure capital. A few remarks are in order. First, , the elasticity of TFP with respect to infrastructure, is not observable as it captures the externality dimension of infrastructure: there are no payments involved, and therefore no income and price data can be used. Second, (6) shows that should data on SX be available, that relationship would enable us to disentangle the market-mediated influence of infrastructure from its externality incidence. However, even though in principle the market-mediated part of infrastructure could be tracked by the corresponding payments 9 and prices, in practice data on infrastructure prices are not available in a consistent way for the countries under analysis. In addition, available data on capital do not distinguish between different types of capital, including infrastructure. Instead of having data on K , we have data ~ on K. Because of this, it is clearer to rewrite to model so as to fit the available data, as: (5') Y = A.X.F(K, L) ~ which leads to Y& A ~& (6') = + + SL X& L& + SK K& + . Y A ~ X L K Finally, the trick is to substitute (3) into (6'), so that we get (appending an error term): A& A ~& (7) = + + . X& A A ~ X The left-hand side of (7) is TFP growth as computed (not estimated) in the standard growth accounting approach. An alternative route to a full estimation of (6') is thus to estimate the A& reduced form (7) using the (year by year) results of (3) in terms of TFP growth rates ( ), A which is convenient as these are available from standard growth accounting exercises for a number of countries. Either (7) or (6') provide an estimation of , the pure externality effect of infrastructure, as opposed to the full elasticity of output with respect to infrastructure. For example, if an estimation of (7) produces a value for not significantly different from zero, it suggests that infrastructure has no externality role in that particular economy. However, because K includes X, it does not imply that infrastructure is not productive: it is just not more productive than other types of capital. A ~& Finally note that (7) or (6') also provide a basis for estimating , the "true" TFP growth. A ~ 10 2.2. Data and estimation There are two main options for estimating (7). One is based on regional panel data, while the other one is a country-per-country approach based on time series data. The panel estimation technique rests on the assumption that a common production function exists for the Asian countries under analysis, with individual country effects to be controlled for. While this approach has been extensively used with state / provincial panel data for India (Hulten et al. (2005)), Italy (La Ferrara and Marcellino (2000)) and the US (Holtz-Eakin (1994)), the above assumption is dubious when applied to a set of countries as diverse as those in our sample. We report below tentative panel estimations that confirm such cross- country heterogeneity. We therefore give priority to individual country estimations, which more realistically do not assume that there is a common underlying technology for all countries. This has been the approach used by most non-infrastructure growth accounting studies. 8 Concerning any possible simultaneity in the estimations, note again that the left hand-side quantity in (8) ­ TFP growth ­ is computed directly from data, not estimated. In particular, we do not have to worry about the typical simultaneity problem in regression-based growth accounting studies, namely reverse causation from the growth rates of GDP to K. The only possible remaining source of simultaneity would be an influence of TFP growth on investment in infrastructure, X& . Possible causes of simultaneity include endogenous X responses of infrastructure policies to TFP growth, making it necessary to test the presence of reverse causation in the data. Country-specific estimations, as opposed to panel estimations, call for longer time series in order to produce efficient estimators. Two sets of long time series can be considered. First, physical indicators of infrastructure stocks have been used in the literature. Canning (1999) uses indicators of telephones lines availability, electricity generating power and length of 8See for example Barro and Sala-i-Martin (2004). 11 paved roads and railways to estimate an aggregate production function. This dataset includes time series of usable length for key infrastructures (excluding water) for all countries included in our exercise. Second, it is in theory possible to build time series of infrastructure stocks based on investment data together with the perpetual inventory method ­ just as time series of K are normally constructed. Unfortunately, in practice financial data on infrastructure (in monetary terms or as percentage of GDP) are scarce for the sample countries. Also, some authors (see Pritchett, 1996) have warned against the poor quality of financial indicators of public investment. For these reasons, we concentrate on physical indicators of infrastructure. With respect to explanatory variables, we have used data from the Canning database in five countries: Indonesia, Philippines, Thailand, South Korea and Singapore. The following series of physical infrastructure indicators from the Canning (1998) database, as extended with the World Bank's infrastructure database, have been used: - Number of telephones and telephone main lines; - Electricity generating capacity; - Total roads (railways and paved roads). Finally, the TFP growth rates calculated by the Asian Productivity Organization (APO, 2004) for the five countries under analysis have been used as the dependent variable, as in (7). The APO has calculated (not estimated) TFP growth rates following the standard methodology that is, following equation (3) and, in addition, taking into account changes in labor quality. 2.3. Results The main results from individual growth accounting regressions are reported in Table 4. First, in South Korea and Singapore, which are the two most developed economies in our sample (Figure 1), we cannot reject the hypothesis that the coefficients on the three infrastructure variables are zero. Again, recall that the interpretation for this result is not that infrastructure is not productive but rather that there is no evidence from this exercise that it is more productive than other types of capital. 12 Second, in Indonesia, Thailand and Philippines we report preliminary evidence that some infrastructure variables are significantly more ­ or less ­ productive than other types of capital. In Indonesia, the number of telephones has a positive coefficient of 0.12, significant at the 90% level. This suggests a productivity level above that of the rest of capital, specifically an externality effect expressed as an output elasticity of 0.12. However, in the same country electricity generating capacity appears to be less productive, at the 95% level of significance. With a R2 of 0.59, it is interesting to note that the growth of the two significant infrastructure indicators seem to explain a large share of the standard TFP growth. Since the electricity generating capacity variable carries a negative coefficient, it implies that the bulk of TFP growth has rested on the increase in the number of telephones. The estimate of the "true" TFP growth (after accounting for infrastructure growth) is only 0.0430% per year. In the Philippines, the telephone variable also has a positive coefficient, significant at the 90% level, again supporting externalities from this variable. The road variable is significant in only one country, Thailand, at the 95% level. But with a R2 of 0.49, this variable alone explains a lot of the standard TFP growth. The estimate of the true TFP growth is a negative -0.3964% per year ­ suggesting that roads have been a primary driving force of productivity growth. With 11 observations only in Thailand, however, caution is warranted in interpreting this result. One possible interpretation for the presence of two groups, with the most developed countries (South Korea and Taiwan, China) exhibiting no specific impact of infrastructure, is that infrastructure is not a binding constraint in these countries because it has been tailored to the needs of the economy, whereas it is in developing countries such as Philippines, Thailand and Indonesia, infrastructure has yet to catch up with the economy's needs and could still be a bottleneck. The negative impact of electricity generating capacity in Indonesia could possibly be interpreted in this context as the result of the instability of infrastructure needs in a rapidly changing economy. However, this interpretation, which is impossible to test with a sample of 5 countries, is clearly not consistent with the results reported in the next section on growth regressions, 13 which are based on a broader sample of countries. Tables 9 to 12 show that the interaction terms between the Low-Income dummy and infrastructure variables often carry a significantly negative coefficient. This suggests that the explanation for the mixed outcome from our growth accounting regressions could be related to factors which are orthogonal to GDP, for example if the productivity impact of the infrastructure stock is conditional on complementary factors such as the quality of regulation and governance in the sector. Figure 1 GDP per capita (US$) 20000 18000 16000 Taiwan 14000 12000 Korea, Republic of 10000 Thailand 8000 6000 Indonesia 4000 Philippines 2000 0 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 The simple OLS specification based on equation (7) has been tested in several directions. First, the regressions have been tested for the presence of endogeneity. For each country, Hausman tests using various lags of the explanatory variables as instruments have been performed, all rejecting endogeneity. However, for several countries autocorrelation for some of the exogenous variables is rejected, making the latter invalid instruments for Hausman tests. In those specific cases, we follow the literature9 in using population and population density (both contemporary and lagged) as instruments for Hausman test, which also leads to the rejection of endogeneity. Second, time dummies were tentatively introduced as explanatory variables in each of the regressions above. The objective of this introduction was to test for possible time-varying effects on TFP growth, for example the role of the 1997 crisis in Asia. We do not report the 9See Straub (2007). 14 results of these estimations as the time variable is never a significant determinant of TFP growth. Thirdly, the individual country regressions used above have the obvious shortcoming that they cannot account for cross-country variations. Could not pooled data reveal cross-country regularities masked by individual estimations? With the important caveat noted above ­ a common technology in all five country is a strong assumption ­ pooled/panel estimations were performed under various specifications. Table 5 reports the outcome of these estimations, with none of the infrastructure significantly different from zero and very low R2. Next, we turn to the results from cross-country regressions. 3. Growth regression This section applies growth regression techniques to the study of the link between infrastructure and growth in the case of East Asian countries. 3.1 Standard framework Standard cross-country regressions in general start from a specification that intends to explain real per capita GDP growth by the initial level of real per capita GDP and explanatory factors such as physical investment, human capital (for example proxied by enrollment in different education levels) and additional factors that vary across studies. Indeed, approximately 60 different variables have been used in this abundant literature (Romp and de Haan, 2005), of which varying subsets have been deemed "robust" by different authors.10 Adding infrastructure capital to this framework yields the following reduced form equation to be estimated: (9) gi =yi0 + Ki + Zi +i I 10See Levine and Renelt (1992), Sala-i-Martin (1997), and Temple (2000) for a discussion. 15 where gi is the growth rate of real per capita GDP for country i, yi0 is initial income (possibly in log form), KIi is a measure of infrastructure capital, and Zi is a vector of covariates as mentioned above. 3.2. Data We opt for physical infrastructure indicators. Three specific reasons support this choice.11 1. As mentioned above, public investment data are subject to a lot of problems, which make them unlikely to capture infrastructure stock or availability properly. 2. Physical indicators allow for a longer time frame and a higher number of countries. 3. They will allow for direct comparisons with the results from the growth accounting exercise. Physical indicators for three different sectors (telecom, energy and transport) are taken from Canning's database, covering the 1971-1995 period. Specifically, we use the following series: · Main telephone lines per 1,000 people. This series is extended up to 2002/2003 using Estache and Goicoechea (2005). · Electricity generating capacity in million kilowatt per 1,000 people. · Rail route length in km per 1,000 people. · Paved road length in km per 1,000 people. Additionally, we perform some tests with alternative variables: Telephone mainlines (per 1,000 people) from the World Bank World Development Indicator (WDI), fixed line and mobile phone subscribers (per 1,000 people) also from WDI, to capture the rise in mobile connections in the second half of the 1990s, roads total network and percentage of paved roads from WDI, which is used here as a quality proxy. We introduce additional proxies for 11See Straub (2007) for a more general discussion of public investment versus physical infrastructure indicators. 16 the quality of the other services under study, namely telephone faults (per 100 mainlines) and electric power transmission and distribution losses in % of output, both from WDI. Other general data include (from WDI, unless mentioned otherwise) measures of GDP per capita, gross fixed capital formation, primary and secondary school enrollment (from Barro and Lee, 2000), primary and secondary schooling expenditures, government stability and Corruption (from Political Risk Service, International Country Risk Guide), life expectancy, M2/GDP (as a measure of financial development), imports/GDP and inflation. 3.3. Sample We rely on a sample of 93 developing or emerging countries. Of these 40 are classified by the World Bank as low income, 25 as lower middle income, 19 as upper middle income and 9 as high income. Note that this last category includes Hong Kong, Korea, Singapore and a number of oil producing countries. Overall, 16 East Asian and Pacific countries are included: China, Fiji, Hong Kong, Indonesia, Korea, Lao PDR, Malaysia, Mongolia, Myanmar, Papua New Guinea, Philippines, Singapore, Thailand, Tonga, Vanuatu, Vietnam. 3.4. Techniques In what follows we present two types of estimations. First, we perform simple cross country estimations based on the collapsed data set for 1971-1995 or 1984-1995 alternatively, using the rate of growth of GDP per capita as dependent variable and standard controls (initial level of GDP, investment, proxies for human capital). In each case, we instrument potentially endogenous infrastructure indicators and perform related tests. We also test specifications with different set of regional dummies (specific East Asian dummy, income groups), and the alternative infrastructure indicators mentioned above. Then, we present panel regressions on 5-year subperiod averages with the same dependent variable. This frequency should result in enough variations in infrastructure indicators to allow the use of fixed effects. Following best practice in this type of exercise, we compare fixed vs. random effects and perform instrumental estimations. Finally, we use Arellano-Bond dynamic panel techniques. 17 3.5. Interpretation Table 6 presents the results from cross country regressions with the 1971-1995 averages. Overall, only the number of telephone lines per hab. is significant, with a positive sign of 0.022. This implies that an increase in 100 lines per hab., from the average level over the period of Venezuela (63) to that of Korea (163) would add 2.2 points to the average growth rate of per capita GDP. All other infrastructure variables are insignificant and the paved roads length one is of the wrong sign. In columns 5 and 6, we add measures of quality of infrastructure, namely the number of telephone faults and electricity losses. These measures are not significant and the main indicators' coefficients are unchanged. When considering instead the 1984-1995 period, in Table 7, which in particular enables us to introduce indices of government stability and corruption as additional control variables, we get even less conclusive results. The number of phone lines is now only significant when quality is controlled for and its coefficient is about half of the 1971-1995 one, while the paved roads variable is now negative and significant. In Tables 8 and 9, we address the fact that infrastructure stocks may be determined simultaneously with output. Following previous contributions in the literature, we use beginning of the period (1971) values of the indicators themselves, as well as 1971 values of the level of population, population density and the share of agriculture in GDP. Overall, the results from IV estimations are similar to simple OLS. The coefficient for the number of phone lines is now larger, between 2.8 (1971-1995) and 4.7 (1984-1995). Note however that a Wu-Hausman test does not reject exogeneity in all but one of the 12 estimations. Next, we test regional effects by interacting the infrastructure indicators with regional dummies. Tables 9 to 13 present the results for telecom, energy, railroad and roads respectively. In each case, we first use an East Asian dummy, then dummies for low and middle income countries respectively. As for telecom, the group of East Asian countries does not display any significantly different behavior (the coefficient is negative but not significant), while income classification indicates that telecom impact is significantly lower in low income countries, a result that may appeal to 18 Röller and Waverman's (1999) conclusions on network externalities in telecom kicking in at near universal coverage level. In Table 10, we observe that the impact of energy is positive and significant for the subgroup of East Asian countries, suggesting that the development of the electric network may have been an important contributor to growth of per capita output during the period. To compare again the same countries as before, the difference between the period average electricity generating capacity of Korea (0.667 million kw per 1,000 hab.) and that of Venezuela (0.376) implies an additional 1.1% per capita GDP growth. As for the level of development, the impact of electricity generation appears lower in low and middle income countries. In Table 10, the impact of the railroad network is positive and weakly significant for East Asian countries, and it is again lower for low and middle income countries (actually slightly negative for low income ones). Finally, a similar pattern is repeated in Table 13 with respect to paved roads. Note finally, that in all cases instrumental estimations fail to yield significant results, and that the Wu-Hausman test fails to reject exogeneity in all but one of the 8 specifications tested. Overall, this exercise seems to provide two main insights. First, East Asian countries display positive and significant returns from infrastructure across most dimensions. Second, a pattern emerges that indicates low or possibly negative returns for low income countries, slightly higher returns for middle income ones and strongly positive returns for the richer countries in the sample. The type of data we use does not allow for a very detailed analysis of this result. One possibility is simply a network effect type of explanation, although it is not clear how this applies to roads for example. Alternatively, it may be the case that richer countries also display more favorable conditions along other dimensions (better incentive structure, more efficient political interactions) that provide the required conditions for a favorable effect of infrastructure investment. Finally, we use alternative infrastructure indicators in Table 14. Using the number of fixed and mobile phone lines, we inquire whether the very quick surge in mobile telephony in the 1990s had a special effect on growth above the effect of traditional main lines, as suggested by Waverman, Meschi and Fuss (2005). In column 1, we use 1984-2003 averages, and introduce both the number of fixed line 1984-1995 and the number of fixed plus mobile lines 19 1996-2003. Mobile lines appear to have a significant and positive effect on GDP per capita growth and render the effect of fixed lines negative. This result loses significance when instruments are used, but again exogeneity is not rejected at usual levels. In column 3, we combine the total length of the road network and the percentage of paved roads, which results in only the second indicator being significant. This indicates that it is the quality of the road network that mostly provides growth dividends. In column 4, an indicator of the number of vehicles per kilometer of road is added to the specification. This variable now shows up positive and significant at the 5% level, while road length and proportion of paved road fail to be significant. If anything, this seems to indicate that, because it is usage of infrastructure that ultimately drives aggregate growth benefit, a proxy for the average use of roads capture the benefits from the extension and the quality of the network. Again, IV estimations yield no clear results and endogeneity is rejected. Next, we perform panel estimations using 5 year averages of the different indicators. The results from fixed effects vs. random effects estimations are shown in Table 15, and a Hausman test is performed to decide which estimation technique is more suited. In all cases, a full set of time dummies is included. The main conclusions are that none of the infrastructure indicators introduced individually is significant, except negative and significant signs for electricity in the random effect specification and for paved roads in the fixed effects one respectively. Fixed effect estimations are supported by the test in 2 out of 4 cases (telecom and roads). In columns 9 and 10, we introduce all four indicators together. The number of phone lines is positive and significant, while electricity and roads remain negative and significant. In this case, the Hausman test favors fixed effect estimation. Again, the interpretation of the signs of the coefficients, and especially the negative ones, is made difficult by the nature of the data. Several lines may be relevant, among which an "optimal stock" type of argument (returns may become negative in case of over accumulation), or arguments about investment decisions being politically driven and therefore departing significantly from efficiency. In Table 16, we address the issue of endogeneity. Instruments are now the lagged value of infrastructure indicators, as well as the ones used previously (1971 values of the share of agriculture in GDP, population, and population density). Following the outcome of the test in 20 column 9 and 10 of Table 15, we chose a fixed effect specification and introduce a full set of time period dummies. Overall, few results are again significant, with only electricity generating capacity being positive and significant. This holds true when all indicators are introduced together. Note that exogeneity is now rejected for the individual estimations with the number of phone lines and the electricity generating capacity.12 4. Conclusion Our results on growth accounting are mixed: in Indonesia and the Philippines telecommunications investment has generated externalities and has contributed to growth more than other types of capital. Roads have positively influenced TFP growth in only one country, Thailand. In South Korea and Singapore, however, two countries which have markedly higher GDP than the other countries in the sample, no significant effect of infrastructure on TFP growth has been detected. Our cross country growth regressions provide relatively fragile results on the impact of infrastructure in per capita GDP growth, a conclusion that contrasts with previous studies that found robust results (Easterly and Servén, 1993; Calderón and Servén, 2004 among others). The number of phone lines appears positively related to growth in the cross country exercise, and some regional patterns emerge, showing above average effects for East Asia and high income countries. However, most results appear not to be robust when using panel techniques or when controlling for an endogenous response of infrastructure to growth. Our growth accounting estimates indicate that infrastructure has contributed to TFP growth in poorer countries, while having no significant effect in other, richer economies. A possible explanation would be that poor countries have less developed infrastructure networks, and experience a one-off productivity dividend as they develop those networks. But in our cross- country growth regressions, which draw on data extending beyond East Asia, the interaction terms between the Low-Income dummy and infrastructure variables (Table 10 to Table 13) carry significantly negative coefficients. This suggests that the explanation for the mixed 12Finally, we perform Arellano-Bond IV estimations similar to the one implemented in Calderón and Servén (2004). Two types of instruments are used: internal ones, constituted by the lagged values of the differenced explanatory variables including infrastructure indicators, and the external ones used above, namely 1971 values of the share of agriculture in GDP, population, and population density. Only electricity generating capacity is significant, and its coefficient is negative. Results are not shown here to save space. 21 outcome from our growth accounting regressions could be related to factors which are orthogonal to GDP, such as government policies and the quality of regulation and governance. The two results could be reconciled with a growth story in which infrastructure constraints, if left unaddressed by governments, can slow the transition towards the long-run growth path, but do not ultimately affect the long-run rate of growth. Governments of poor countries in East Asia may do a better job of addressing these constraints than governments of poor countries elsewhere. It is interesting to compare our results with those of Seethepalli et al (2007). As in Estache et al. (2005), these authors compare a "benchmark" (without infrastructure) production function estimated at the steady state with the same specification including infrastructure variables (these include physical indicators for telecom, electricity, roads, sanitation and water).13 Using data for East Asian countries14, they find that virtually all dimensions of infrastructure positively influence GDP per capita when controlling for education and investment. The cross-country regressions have similar controls, while our growth accounting estimations both investment and changes in the quality of labor are captured in the APO calculations of TFP growth rates, making comparison meaningful. Our growth accounting results in the Philippines, Indonesia and Thailand tends to support their result that telecom and road infrastructure enhances productivity conditional on investment and education, but not so for electricity. But our results from South Korea and Singapore, where no infrastructure impact has been found, suggest that individual countries could be at variance with the cross-country results of Seethepalli et al (2007). And our growth regressions provide much weaker results than those obtained by Seethepalli et al. One of the reasons might be the fact that they do not control for the potential endogeneity of infrastructure stocks. While they argue that the use of stocks rather than flows mitigates the problem of reverse causation, countries may have unobserved characteristics that lead them both to have higher infrastructure stocks and higher growth. The fact that fixed effects estimations are not carried out reinforces the concern that this may bias the results (Holtz-Eakin, 1994). 13This approach parallels our interpretation of our results in terms of infrastructure productivity greater than or less than that of other types of capital. 14Australia, Cambodia, China, Fiji, Indonesia, South Korea, Laos, Mongolia, Papua New Guinea, Philippines, Singapore, Thailand, Tonga, Vanuatu, Vietnam. 22 A first conclusion is therefore that the results from studies using aggregate data lack robustness, Indeed, as shown above, different techniques (production function, growth regressions, growth accounting) produce very different results, even when looking at similar set of countries. Moreover, similar techniques, when applied to slightly different samples, also fail to produce consistent results. For example, we were unable to reproduce the results from Calderón and Servén (2004) in our sample of 93 developing countries. Keeping these caveats in mind, what are the potential lessons for East Asian economies? Overall, our results give only limited support to the notion that infrastructure investment has driven growth in East Asia. Our results do not seem to be inconsistent with a story in which infrastructure can constrain growth, when that growth potential is generated exogenously, and that East Asian countries have been relatively successful in addressing infrastructure constraints as they arise. But the weakness of our data and results do not permit any definitive conclusions about the theoretical channels by which infrastructure may have influenced growth in East Asia. If indeed East Asia is more effective than other regions at responding to infrastructure constraints it would be useful to understand why. Various arguments could be mounted. For example, East Asia has high levels of savings, and the availability of financing may facilitate more rapid responses. East Asian countries have typically relied on powerful planning agencies, such as Japan's MITI, etc. And to the extent that private investment in infrastructure has played a role in total investment, it is notable that the modalities employed in East Asia have differed from those employed elsewhere: for example, while East Asia focused on attracting investment at the wholesale level and greenfield sites (eg independent power producers), Latin America placed greater emphasis on the concessioning of existing retail systems. Testing such hypotheses is a subject for separate enquiry. 23 Table 4. Results from growth accounting (single-country OLS estimations) Dependent variable : Indonesia Thailand Philippine South Singapore TFP growth rate s Korea Constant 0.000430 -0.003964 -0.032361 -0.011897 0.013270 (0.012378) (0.013298) (0.016032) (0.049837) (0.019744) Number of telephones 0.122859* -0.082006 0.282970* -0.071994 -0.090847 (0.055356) (0.048992) (0.152046) (0.123365) (0.107652) Total roads (railways and -0.119878 0.470495** -0.017326 0.539977 -0.191091 paved roads) (0.129726) (0.196164) (0.221852) (0.534860) (0.300689) Electricity generating -0.047187** 0.027691 0.062823 -0.026010 -0.003853 capacity (0.015937) (0.052692) (0.123118) (0.083775) (0.076603) R2 0.590285 0.486470 0.153937 0.069669 0.092980 Number of observations 14 11 28 21 18 Standard errors in parentheses. ** significant at the 5% level; * significant at the 10% level. 24 Table 5. Results from growth accounting (panel estimations) Dependent variable : TFP Pooled Pooled Fixed Fixed Random Random growth rate regression, regression, effects, effects, effects, effects, unbalanced balanced unbalanced balanced unbalanced balanced sample sample sample sample sample sample Constant -0.008737 -0.012026 -0.003989 -0.013044 (0.007908) (0.010743) (0.010032) (0.010060) Number of telephones 0.043324 0.057099 0.027650 0.030315 0.035615 0.059584 (0.048094) (0.079778) (0.050246) (0.084314) (0.048686) (0.079787) Total roads (railways and 0.154079 0.207674 0.029425 -0.042937 0.085562 0.226184 paved roads) (0.107193) (0.193470) (0.125109) (0.276359) (0.115693) (0.187275) Electricity generating -0.015537 -0.031110 -0.017844 -0.027643 -0.016328 -0.032020 capacity (0.027578) (0.034907) (0.027756) (0.037000) (0.027313) (0.034794) R2 0.046681 0.091113 0.112922 0.167258 0.086519 0.080648 Number of observations 92 40 92 40 92 40 Standard errors in parentheses. ** significant at the 5% level; * significant at the 10% level. 25 Table 6. Cross section 1971-1995, OLS. (1) (2) (3) (4) (5) (6) OLS OLS OLS OLS OLS OLS pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth Constant -2.528 -3.061 -3.657 -3.528 -2.665 -2.486 (0.970)** (1.161)** (1.239)*** (0.996)*** (1.265)** (2.058) GDPpc71 -0.001 -0.000 -0.000 -0.000 -0.001 -0.001 (0.000)*** (0.000) (0.000) (0.000) (0.000)*** (0.000)*** prim_expen_70 -0.053 -0.028 -0.058 -0.041 -0.060 -0.025 (0.034) (0.039) (0.065) (0.036) (0.037) (0.058) second_expen_70 0.002 0.001 -0.000 0.001 0.002 -0.013 (0.001) (0.001) (0.004) (0.002) (0.002) (0.009) Invt_GDP7195 0.227 0.223 0.289 0.270 0.234 0.220 (0.042)*** (0.054)*** (0.054)*** (0.042)*** (0.053)*** (0.074)*** main7195 0.022 0.022 (0.003)*** (0.004)*** egc7195 2.871 4.108 (2.084) (2.577) rail7195 0.127 (1.687) pavroads7195 -0.172 (0.362) tel_faults7195 0.001 (0.003) elec_loss7195 0.030 (0.071) Observations 51 48 41 51 47 33 R-squared 0.68 0.53 0.50 0.56 0.69 0.50 Robust standard errors in parentheses. Significant at 10%; ** significant at 5%; *** significant at 1%. 26 Table 7. Cross section 1984-1995, OLS. (1) (2) (3) (4) (5) (6) OLS OLS OLS OLS OLS OLS pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth Constant -5.691 -6.385 -6.289 -8.614 -4.570 -4.183 (1.813)*** (1.744)*** (1.919)*** (1.483)*** (2.326)* (3.437) GDPpc71 -0.000 0.000 -0.000 0.000 -0.000 0.000 (0.000) (0.000) (0.000) (0.000)** (0.000) (0.000) prim_enrol8495 0.007 0.014 0.013 0.020 -0.010 0.018 (0.013) (0.013) (0.013) (0.011)* (0.014) (0.033) Invt_GDP8495 0.261 0.272 0.297 0.281 0.238 0.290 (0.066)*** (0.063)*** (0.065)*** (0.054)*** (0.075)*** (0.086)*** Govstab8495 0.070 0.045 -0.002 0.533 0.344 -0.113 (0.305) (0.317) (0.320) (0.293)* (0.321) (0.417) Corrup8495 0.110 0.202 0.151 0.140 -0.478 0.107 (0.356) (0.326) (0.417) (0.337) (0.459) (0.463) main8495 0.005 0.012 (0.004) (0.005)** egc8495 -0.411 -1.629 (1.343) (1.777) rail8495 1.262 (1.720) roads8495 -1.414 (0.631)** tel_faults8495 0.006 (0.005) elec_loss8495 -0.104 (0.095) Observations 55 54 47 49 48 42 R-squared 0.51 0.48 0.48 0.60 0.55 0.43 Robust standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. 27 Table 8. Cross section 1971-1995, 2SLS. (1) (2) (3) (4) (5) (6) 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth Constant -2.537 -1.983 -2.839 -3.184 -4.132 27.406 (1.109)** (1.245) (1.341)** (1.322)** (3.062) (93.724) GDPpc71 -0.001 0.000 -0.001 -0.000 -0.001 0.003 (0.000)*** (0.000) (0.000)** (0.000)** (0.001) (0.012) prim_expen_70 -0.063 -0.075 -0.047 -0.047 -0.068 0.107 (0.047) (0.055) (0.072) (0.050) (0.051) (0.614) second_expen_70 0.002 0.001 -0.003 0.002 0.004 -0.073 (0.003) (0.003) (0.006) (0.003) (0.004) (0.203) Invt_GDP7195 0.230 0.230 0.248 0.250 0.253 0.245 (0.042)*** (0.053)*** (0.055)*** (0.051)*** (0.048)*** (0.416) main7195 0.028 0.028 (0.013)** (0.014)** egc7195 -6.453 -48.742 (5.223) (164.983) rail7195 1.892 (1.522) pavroads7195 0.464 (0.429) tel_faults7195 0.012 (0.022) elec_loss7195 -1.881 (6.248) Observations 44 44 36 41 41 30 Wu-Hausman F 0.70 0.13 0.11 0.18 0.83 0.28 test, p-value Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. Instruments: 1971 Infrastructure physical stock, 1971 share of agriculture in GDP, 1971 population density. 28 Table 9. Cross section 1984-1995, 2SLS. (1) (2) (3) (4) (5) (6) 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth Constant -3.785 -6.569 -6.276 -10.275 -3.602 3.534 (2.091)* (3.047)** (1.999)*** (2.832)*** (2.562) (17.124) GDPpc71 -0.000 0.000 -0.000 0.000 -0.000 0.003 (0.000) (0.001) (0.000) (0.000) (0.000) (0.004) prim_enrol8495 -0.011 0.030 0.017 0.023 -0.052 0.222 (0.022) (0.045) (0.023) (0.020) (0.032) (0.285) Invt_GDP8495 0.245 0.180 0.248 0.264 0.286 0.122 (0.070)*** (0.083)** (0.083)*** (0.078)*** (0.087)*** (0.307) Govstab8495 0.019 0.511 0.190 0.935 0.387 1.065 (0.395) (0.924) (0.417) (0.527)* (0.497) (2.511) Corrup8495 -0.130 -0.303 0.003 0.065 -1.030 -1.900 (0.368) (1.085) (0.407) (0.359) (0.591)* (3.664) main8495 0.024 0.047 (0.017) (0.021)** egc8495 -5.502 -38.323 (14.823) (51.504) rail8495 1.393 (1.892) roads8495 -1.789 (1.217) tel_faults8495 0.028 (0.022) elec_loss8495 -1.380 (1.764) Observations 48 47 44 40 43 37 Wu-Hausman F 0.29 0.77 0.14 0.29 0.33 0.05 test, p-value Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. Instruments: 1971 Infrastructure physical stock, 1971 share of agriculture in GDP, 1971 population density. 29 Table 10. Cross section, Telecom, regional dummies interactions. (1) (2) (3) (4) OLS OLS 2SLS 2SLS pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth Constant -2.577 -1.769 -0.034 2.113 (0.980)** (0.913)* (3.507) (5.107) GDPpc71 -0.001 -0.001 -0.001 -0.001 (0.000)*** (0.000)*** (0.001) (0.001)* prim_expen_70 -0.055 -0.063 -0.083 -0.084 (0.034) (0.034)* (0.097) (0.090) second_expen_70 0.002 0.003 0.004 0.002 (0.001) (0.001)* (0.006) (0.005) Invt_GDP7195 0.229 0.220 0.065 0.148 (0.043)*** (0.040)*** (0.197) (0.113) main7195 0.029 0.022 0.027 0.000 (0.007)*** (0.003)*** (0.025) (0.000) EA*main -0.006 0.313 (0.005) (0.339) LI*main -0.222 -1.188 (0.066)*** (1.224) MI*main 0.004 0.009 (0.003) (0.030) Observations 51 51 44 44 R-squared 0.68 0.75 Wu-Hausman F test, 0.22 0.47 p-value Robust standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. Regional dummies: EA (East Asia), LI (low income countries), MI (middle income countries). Instruments: see Table 3 and 4. 30 Table 11. Cross section, Energy, regional dummies interactions. (1) (2) (3) (4) OLS OLS 2SLS 2SLS pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth Constant -2.058 -1.085 -0.784 -0.646 (1.159)* (1.150) (1.752) (1.542) GDPpc71 -0.000 -0.000 0.000 -0.000 (0.000) (0.000)** (0.000) (0.001) prim_expen_70 -0.048 -0.079 -0.053 -0.093 (0.038) (0.037)** (0.064) (0.048)* second_expen_70 0.000 0.002 0.001 0.002 (0.002) (0.001) (0.003) (0.003) Invt_GDP7195 0.197 0.188 0.106 0.182 (0.049)*** (0.047)*** (0.128) (0.060)*** egc7195 -0.899 4.450 -3.038 -0.057 (2.244) (1.454)*** (6.514) (6.997) EA*egc 4.625 32.843 (1.361)*** (30.305) LI*egc -18.451 -19.050 (3.343)*** (16.486) MI*egc -3.719 0.000 (1.186)*** (0.000) Observations 48 48 44 44 R-squared 0.60 0.70 Wu-Hausman F test, 0.25 0.87 p-value Robust standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. Regional dummies: EA (East Asia), LI (low income countries), MI (middle income countries). Instruments: see Table 3 and 4. 31 Table 12. Cross section, Rail, regional dummies interactions. (1) (2) (3) (4) OLS OLS 2SLS 2SLS pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth Constant -3.245 -2.387 -0.320 -0.489 (1.257)** (1.138)** (2.900) (1.867) GDPpc71 -0.000 -0.000 -0.000 -0.001 (0.000) (0.000) (0.000) (0.000)*** prim_expen_70 -0.043 -0.046 0.001 -0.049 (0.061) (0.061) (0.121) (0.081) second_expen_70 0.001 0.004 0.000 0.012 (0.004) (0.004) (0.010) (0.010) Invt_GDP7195 0.241 0.224 0.024 0.172 (0.058)*** (0.049)*** (0.194) (0.071)** rail7195 0.661 52.269 4.073 -19.234 (1.771) (16.292)*** (2.968) (9.942)* EA.rail 22.630 101.633 (11.833)* (78.180) LI.rail -57.714 0.000 (16.315)*** (0.000) MI.rail -50.803 23.624 (16.076)*** (10.950)** Observations 41 41 36 36 R-squared 0.55 0.61 Wu-Hausman F test, 0.04 0.12 p-value Robust standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. Regional dummies: EA (East Asia), LI (low income countries), MI (middle income countries). Instruments: see Table 3 and 4. 32 Table 13. Cross section, Roads, regional dummies interactions. (1) (2) (3) (4) OLS OLS 2SLS 2SLS pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth Constant -3.045 -1.686 -2.085 2.604 (1.043)*** (1.011) (2.169) (7.157) GDPpc71 -0.000 -0.000 -0.000 -0.001 (0.000) (0.000) (0.000) (0.001) prim_expen_70 -0.043 -0.062 -0.062 -0.087 (0.036) (0.034)* (0.058) (0.084) second_expen_70 0.002 0.002 0.002 0.001 (0.002) (0.001) (0.003) (0.004) Invt_GDP7195 0.236 0.209 0.187 0.097 (0.048)*** (0.043)*** (0.109)* (0.196) pavroads7195 -0.101 4.250 0.353 -8.633 (0.350) (1.621)** (0.480) (10.902) EA* pavroads7195 1.746 2.914 (0.891)* (4.421) LI* pavroads7195 -6.112 0.000 (1.736)*** (0.000) MI* pavroads7195 -4.037 9.525 (1.461)*** (11.397) Observations 51 51 41 41 R-squared 0.59 0.65 Wu-Hausman F test, p- 0.29 0.32 value Robust standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. Regional dummies: EA (East Asia), LI (low income countries), MI (middle income countries). Instruments: see Table 3 and 4. 33 Table 14. Cross section, Alternative infrastructure indicators. (1) (2) (3) (4) (5) (6) OLS 2SLS OLS OLS 2SLS 2SLS pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth Constant -3.083 -1.456 -3.770 -3.544 -2.077 -9.343 (1.624)* (4.275) (1.549)** (1.627)** (6.697) (11.441) GDPpc71 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 (0.000)*** (0.000) (0.000) (0.000)* (0.000) (0.000) prim_enrol8403 -0.018 -0.026 (0.016) (0.033) second_enrol8403 0.021 -0.012 (0.011)* (0.050) Invt_GDP8403 0.073 -0.015 (0.377) (0.554) Invt_GDP8495 0.204 0.202 0.154 0.321 (0.054)*** (0.072)*** (0.143) (0.339) prim_enrol8495 0.007 0.007 -0.010 0.061 (0.012) (0.013) (0.058) (0.109) Govstab8495 (0.052) (0.157) -0.114 0.086 -0.041 0.351 0.412 0.488 (0.270) (0.304) (0.700) (1.133) Corrup8495 (0.252) (0.573) -0.066 -0.370 -0.489 0.196 -0.102 -0.155 (0.289) (0.347) (0.853) (1.389) telmain_8495 -0.014 -0.080 (0.008)* (0.092) fix+mob_9603 0.008 0.038 (0.003)** (0.039) roads_tot_8495 0.000 0.000 0.000 -0.000 (0.000) (0.000) (0.000) (0.000) pavroads%_8495 0.028 0.007 0.057 -0.076 (0.016)* (0.021) (0.087) (0.175) vehicles8495 0.018 0.063 (0.009)** (0.095) Observations 46 41 53 52 41 40 R-squared 0.64 0.48 0.52 Wu-Hausman F 0.34 0.82 0.42 test, p-value Robust standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. Instruments: see Table 3 and 4. 34 Table 15. Panel 5 years average, Fixed and random effects. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) pcGDPgrowt pcGDPgrowt pcGDPgrowt pcGDPgrowt pcGDPgrowt pcGDPgrowt pcGDPgrowt pcGDPgrowt pcGDPgrowt pcGDPgrowt h h h h h h h h h h Fixed Effects Random Fixed Effects Random Fixed Effects Random Fixed Effects Random Fixed Effects Random Effects Effects Effects Effects Effects Constant -3.624 -3.580 -4.288 -5.847 -2.489 -2.503 -4.853 -8.293 3.691 -3.969 (2.595) (1.093)*** (6.055) (1.477)*** (6.391) (1.833) (6.059) (1.475)*** (6.833) (1.707)** GDPpc -0.000 -0.000 -0.000 -0.000 0.000 -0.000 -0.000 -0.000 -0.000 0.000 (0.000)*** (0.000)** (0.000)*** (0.000) (0.000) (0.000) (0.000) (0.000)*** (0.000) (0.000) Invt/gdp 0.111 0.145 0.093 0.109 0.147 0.196 0.087 0.134 0.113 0.186 (0.030)*** (0.023)*** (0.061) (0.036)*** (0.069)** (0.043)*** (0.053) (0.034)*** (0.075) (0.040)*** lifeexpect 0.091 0.045 0.102 0.099 0.062 0.041 0.120 0.091 -0.006 0.079 (0.042)** (0.020)** (0.105) (0.029)*** (0.120) (0.036) (0.103) (0.029)*** (0.129) (0.032)** m2/gdp -0.054 -0.006 -0.032 0.015 -0.049 -0.002 -0.020 -0.000 -0.006 0.014 (0.014)*** (0.008) (0.028) (0.014) (0.030) (0.016) (0.030) (0.014) (0.032) (0.015) imports/gd 0.030 0.007 0.029 0.003 0.040 -0.012 -0.003 0.014 0.063 -0.028 p (0.019) (0.009) (0.042) (0.013) (0.040) (0.015) (0.035) (0.012) (0.047) (0.017)* inflation -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** main -0.003 0.001 0.023 0.012 (0.003) (0.003) (0.011)** (0.006)* egc -2.427 -1.325 -6.563 -4.407 (1.516) (0.709)* (2.014)*** (1.258)*** rail -3.314 -1.242 -4.596 -0.642 (5.158) (1.093) (5.445) (1.083) pav -2.823 0.116 -4.053 -0.851 (1.084)*** (0.188) (1.776)** (0.448)* Obs 497 497 313 313 237 237 276 276 202 202 R-squared 0.23 0.20 0.27 0.24 0.35 Haus test 51.06*** 12.43 11.48 18.41** 19.79* FE vs RE Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. Full set of period dummies included. 35 Table 16. Panel 5 years average, 2SLS estimations. (1) (2) (3) (4) (5) Fixed Effects, Fixed Effects, Fixed Effects, Fixed Effects, Fixed Effects, 2SLS 2SLS 2SLS 2SLS 2SLS pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth pcGDPgrowth Constant -5.587 -17.221 -7.866 -20.282 -8.576 (2.496)** (6.732)** (7.535) (6.758)*** (11.968) GDPpc -0.000 0.000 -0.000 -0.000 0.004 (0.000)** (0.000) (0.000) (0.000) (0.002)* Invt/gdp 0.143 0.267 0.216 0.043 0.205 (0.031)*** (0.070)*** (0.065)*** (0.054) (0.104)** lifeexpect 0.106 0.239 0.094 0.340 0.071 (0.042)** (0.109)** (0.124) (0.106)*** (0.192) m2/gdp -0.057 -0.040 -0.043 -0.030 -0.081 (0.017)*** (0.033) (0.031) (0.033) (0.053) imports/gdp 0.038 -0.070 0.023 0.045 -0.025 (0.021)* (0.050) (0.039) (0.039) (0.076) inflation -0.001 -0.001 -0.001 -0.001 -0.001 (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)** main -0.002 -0.082 (0.005) (0.052) egc 6.211 38.411 (2.714)** (20.499)* rail 2.650 -11.985 (7.567) (14.405) pav -0.459 -6.015 (1.889) (5.749) Observations 362 218 177 202 148 Haus test endog 60.17*** 38.84*** 0.15 13.12 7.28 p-value Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. Instruments: Lagged infrastructure, agri/gdp71, population density 71, population 71. Full set of period dummies included. 36 References ADB, IBRD, WB, JICA (2005), Connecting East Asia. A New Framework for Infrastructure, The World Bank. Aghion P. and P. Howitt (1998), Endogenous Growth Theory, MIT Press, Cambridge, MA. Asian Productivity Organization, 2004, Total Factor Productivity Growth: Survey and Country Reports. Barro, Robert J. and J.-W. Lee (2000), "International Data on Educational Attainment: Updates and Implications", CID Working Paper No. 42. Barro R.. and X. Sala-i-Martin (2005), Growth Theory, 2nd Ed., MIT Press Calderón, C. and L. Servén (2004), "The Effects of Infrastructure Development on Growth and Income Distribution" Central Bank of Chile, Working Papers N° 270. Canning D. (1998), "A Database of World Infrastructure Stocks, 1950-95", World Bank Policy Research Working Paper No. 1929. Canning, D. (1999), "Infrastructure's Contribution to Aggregate Output", Policy Research Working Paper Series 2246, The World Bank. Easterly W. and L. Servén (2003), The limits of stabilization. Infrastructure, Public Deficits and Growth in Latin America, 1980-2000, Princeton, N.J.: Princeton University Press. Estache A., B. Speciale and D. Veredas (2005), "How much does Infrastructure Matter to Growth in Sub-Saharan Africa", mimeo. Estache A. and A. Goicoechea (2005), "A "research" database on infrastructure economic performance", World Bank Policy Research Working Paper 3643. 37 Gibson J. and S. Rozelle, 2003, "Poverty and Access to Roads in Papua New Guinea", Economic Development and Cultural Change. Henderson V. 2005, Urbanization and Growth, in Handbook of Economic Growth, Philippe Aghion and Steven Durlauf, eds., ELSEVIER, 2005. Henderson J.V., Shalizi Z. and Venables A., 2001, "Geography and Development", Journal of Economic Geography, 1, 81-106. Holtz-Eakin D. (1994), "Public-Sector Capital and the Productivity Puzzle", Review of Economics and Statistics, 76, 12-21. Hulten C.R., Bennathan E. and S. Srinivasan (2005) "Infrastructure, Externalities, and Economic Development: A Study of Indian Manufacturing Industry", mimeo, World Bank. Jacoby H., 2000, "Access to Markets and the Benefits of Rural Roads", The Economic Journal, 110, 713-737. La Ferrara E. and M. Marcelino (2000), "TFP, Costs and Public Infrastructure: An Equivocal Relationship", mimeo, Bocconi University and IGIER. Levine, R. and D. Renelt, (1992) "A sensitivity analysis of cross-country growth regressions", American Economic Review, (82), 942-963. Pritchett L. (1996), "Mind your P's and Q's. The Cost of Public Investment is Not the Value of Public Capital", World Bank Policy Research Working paper 1660. Roberts, P., Shyam KC, and C. Rastogi (2006), Rural Access Index: A Key Development Indicator, World Bank Transport Paper 10. Romp W. and J. de Haan (2005), "Public capital and economic growth: a critical survey", European Investment Bank Papers Volume 10. n°1. Sala-i-Martin, 1997, "I just ran two millions regressions", American Economic Review. 38 Seethepalli, K., Bramati, M.C. and Veredas, D., 2007, How Relevan Is Infrastructure to Growth in East Asiat, mimeo, World Bank. Straub S. (2007), "Infrastructure and Development: A Critical Appraisal of the Macro Level Literature", mimeo, World Bank. Straub S. (2008), "Infrastructure and Growth in Developing Countries: Recent Advances and Research Challenges", World Bank Policy Research Working Paper 4460. Temple J., 2000, Growth regressions and what the textbooks don't tell you, Bulletin of Economic Research. Waverman, L., Meschi, M., and Fuss, M. (2005). "The impact of telecoms on economic growth indeveloping countries". Vodafone Policy Paper Series: Africa: The impact of mobile phones. Number 2,10-23. Williamson Jeffrey, 1965, "Regional Inequality and the Process of National Development: A Description of the Patterns," Economic Development and Cultural Change, vol. XIII, no. 4, Part II (July 1965), Supplement pp. 84. 39 Annex 1 Data Used to Construct Table 1 GDP, PPP (constant 2000 international $millions) 1975 1995 1975 1995 Benin 2764 5405 Fiji 2287 3644 Botswana 1610 9813 Kiribati 639 312 Burkina Faso 4541 9027 Papua New Guinea 6516 11732 Burundi 2716 4527 Solomon Islands 219 902 Cameroon 12873 22120 Pacific 9661 16590 Central African Republic 3391 4070 China 553368 3289651 Chad 4068 5764 Hong Kong, China 34260 147174 Congo, Dem. Rep. 52934 36764 Indonesia 145836 578545 Congo, Rep. 1541 3214 Korea, Rep. 130306 612756 Cote d'Ivoire 16038 22699 Malaysia 36847 155960 Gabon 5605 7232 Philippines 142576 251787 Gambia, The 879 1722 Singapore 14853 69738 Ghana 19302 30880 Thailand 80327 379609 Guinea-Bissau 663 1190 East Asia 1138373 5485220 Kenya 13012 28492 Lesotho 1346 4022 Australia 220530 405317 Madagascar 10203 11314 Austria 121243 198386 Malawi 3038 5542 Belgium 158164 244173 Mali 4607 7099 Canada 403919 690682 Mauritania 2795 4417 Denmark 90191 132958 Niger 5246 6942 Finland 71506 106834 Nigeria 56673 86360 France 825783 1330942 Rwanda 3706 4689 Germany 1146873 1886329 Senegal 7724 11896 Greece 112493 157198 Sierra Leone 2754 2549 Iceland 3512 6318 South Africa 238007 336107 Ireland 30949 69229 Sudan 19806 36507 Italy 780564 1311128 Togo 4179 5889 Japan 1604460 3154122 Zambia 6958 7322 Luxembourg 7021 15457 Zimbabwe 17294 30394 Netherlands 240912 378810 Africa 526273 753968 New Zealand 49479 67479 Norway 67021 128940 Bangladesh 70353 154277 Portugal 79042 151892 India 686903 1812285 Spain 463198 734659 Nepal 11141 25576 Sweden 145256 199406 Pakistan 71015 221173 Switzerland 149436 202905 Seychelles 439 1002 United Kingdom 877471 1353078 Sri Lanka 20570 52119 United States 4276900 7972800 South Asia 860421 2266432 OECD 11925923 20899042 Georgia 19101 7102 Algeria 77437 139093 Hungary 90088 106685 Egypt, Arab Rep. 57323 183865 Latvia 16076 14353 Iran, Islamic Rep. 241211 303055 Eastern Europe 125265 128140 Israel 49149 115993 40 Annex 1 1975 1995 1975 1995 Jordan 4775 17044 Ecuador 20981 37910 Kuwait 29998 39060 El Salvador 19155 24836 Morocco 41924 83587 Guatemala 20962 37241 Oman 6398 26067 Guyana 2634 2550 Saudi Arabia 164947 246567 Haiti 11586 11407 Syrian Arab Republic 18863 47456 Honduras 7433 15887 Tunisia 19582 45666 Jamaica 6964 9322 Turkey 164329 359470 Mexico 379924 690902 United Arab Emirates 22440 54161 Nicaragua 15676 12071 MENA 898376 1661084 Panama 7971 14222 Paraguay 8450 22475 Bahamas, The 1818 4168 Peru 80292 108373 Argentina 289859 392783 St. Vincent and the Belize 347 1066 Grenadines 206 525 Bolivia 12241 16759 Suriname 2297 2228 Brazil 595873 1119487 Swaziland 1496 3700 Chile 36824 114771 Trinidad and Tobago 7342 9159 Colombia 105789 239245 Uruguay 17484 26419 Costa Rica 11651 25234 Venezuela, RB 92975 134806 Dominican Republic 18864 37741 LAC 1777094 3115287 Electricity Generating Capacity ('000 MW) 1975 1995 1975 1995 ANGOLA 523 617 NIGER 20 63 BENIN 15 15 NIGERIA 860 5881 BURUNDI 6 43 REUNION 74 299 CAMEROON 225 627 RWANDA 35 34 CAPE VERDE IS. 6 7 SENEGAL 130 231 CENTRAL AFR.R. 17 43 SEYCHELLES 11 28 CHAD 22 29 SIERRA LEONE 95 126 COMOROS 1 5 SOMALIA 18 70 CONGO 32 118 SUDAN 205 500 DJIBOUTI 24 85 TANZANIA 160 543 ETHIOPIA 320 464 TOGO 24 34 GABON 58 310 UGANDA 163 162 GAMBIA 10 29 ZAIRE 1217 3193 GHANA 896 1187 ZAMBIA 1031 2436 GUINEA 175 176 ZIMBABWE 1192 2148 GUINEA-BISS 8 11 AFRICA 9699 25172 IVORY COAST 360 1173 KENYA 283 809 LIBERIA 300 332 BAHAMAS 255 401 MADAGASCAR 95 220 BARBADOS 67 140 MALAWI 87 185 BELIZE 11 25 MALI 37 87 COSTA RICA 404 1165 MAURITANIA 39 105 DOMINICA 6 8 MAURITIUS 132 364 DOMINICAN REP. 732 1450 MOZAMBIQUE 793 2383 EL SALVADOR 314 751 41 Annex 1 1975 1995 1975 1995 GRENADA 7 9 BANGLADESH 933 3284 GUATEMALA 327 766 INDIA 22249 93755 HAITI 89 153 NEPAL 62 292 HONDURAS 159 305 PAKISTAN 2236 14025 JAMAICA 687 1182 SRI LANKA 381 1555 MEXICO 11559 44257 South Asia 25861 112911 NICARAGUA 252 457 PANAMA 346 957 BULGARIA 7060 12087 PUERTO RICO 3453 4575 ROMANIA 11577 22276 ST.KITTS&NEVIS 13 16 POLAND 20057 29465 ST.LUCIA 14 22 Eastern Europe 38694 63828 ST.VINCENT&GRE 9 14 TRINIDAD&TOBAGO 404 1150 AUSTRALIA 21509 39693 ARGENTINA 9260 19610 AUSTRIA 10016 17440 BOLIVIA 376 805 BELGIUM 9809 14916 BRAZIL 19569 59036 CANADA 61352 113340 CHILE 2620 5854 DENMARK 5958 11144 COLOMBIA 3504 10758 FINLAND 7395 14427 ECUADOR 525 2539 FRANCE 46289 107611 GUYANA 170 114 GREECE 4664 8942 PARAGUAY 191 6533 HUNGARY 4291 7012 PERU 2400 3831 ICELAND 514 1081 SURINAME 301 425 IRELAND 2051 4399 URUGUAY 796 2052 ITALY 39163 65821 VENEZUELA 4570 19975 JAPAN 112285 226966 LAC 63390 189335 LUXEMBOURG 1157 1257 MALTA 110 250 ALGERIA 1107 6007 NETHERLANDS 14931 19012 BAHRAIN 187 1080 NEW ZEALAND 4901 7520 EGYPT 3955 16015 NORWAY 16928 27674 IRAN 4850 26257 PORTUGAL 3227 9378 IRAQ 840 9500 SPAIN 25756 45764 ISRAEL 2251 4480 SWEDEN 23135 33623 JORDAN 92 1126 SWITZERLAND 11846 16657 KUWAIT 1474 6988 U.K. 73923 70213 MOROCCO 958 3795 U.S.A. 527346 764876 OMAN 91 1744 OECD 1028556 1629016 QATAR 204 1365 SAUDI ARABIA 425 20934 FIJI 83 200 SYRIA 684 4330 PAPUA N.GUINEA 255 490 TUNISIA 426 1736 SOLOMON IS. 8 12 TURKEY 4165 20953 TONGA 3 7 UNITED ARAB E. 175 5390 Pacific Islands 349 709 YEMEN 14 810 MENA 21898 132510 CHINA 35000 204100 HONG KONG 2274 10096 INDONESIA 1259 20296 KOREA, REP. 5135 35355 LAOS 55 256 42 Annex 1 MALAYSIA 1227 10600 SINGAPORE 1150 4513 MONGOLIA 266 901 THAILAND 2754 17544 MYANMAR 437 1344 East Asia 52788 312727 PHILIPPINES 3231 7722 Paved road length (average of five years to 1975 and 1995), kilometers 1975 1995 1975 1995 ANGOLA 7292 8995 MEXICO 51278 86988 BENIN 798 1210 NICARAGUA 1422 1605 BOTSWANA 165 3635 PANAMA 2146 3004 BURKINA FASO 473 1768 ARGENTINA 36904 61400 BURUNDI 116 1028 BOLIVIA 1128 1954 CAMEROON 1125 3750 BRAZIL 54418 142919 CENTRAL AFR.R. 193 510 CHILE 8724 11974 CHAD 267 430 COLOMBIA 6664 12778 CONGO 402 1030 ECUADOR 3161 5663 DJIBOUTI 269 363 PARAGUAY 872 2785 GABON 209 680 PERU 5074 7571 GAMBIA 296 590 VENEZUELA 19643 31379 GHANA 6958 8523 LAC 199470 384367 IVORY COAST 1549 4500 KENYA 5767 13078 CHINA 92000 207000 LESOTHO 215 802 HONG KONG 1031 1594 LIBERIA 296 574 INDONESIA 28356 153046 MADAGASCAR 3782 5352 KOREA, REP. 7803 51530 MALAWI 1299 2480 MALAYSIA 15977 42910 MALI 1734 2297 PHILIPPINES 15990 25827 MAURITANIA 613 870 SINGAPORE 1605 2893 MAURITIUS 1613 1730 TAIWAN, CHINA 9415 16987 MOZAMBIQUE 3458 5309 THAILAND 14058 43659 NIGER 1014 4405 East Asia 186235 545444 NIGERIA 16713 31667 RWANDA 84 790 EGYPT 9216 17902 SENEGAL 2439 4300 MOROCCO 21937 29813 SEYCHELLES 85 210 OMAN 332 5598 SIERRA LEONE 1099 1743 SAUDI ARABIA 9950 33820 SOUTH AFRICA 38141 57511 SYRIA 11222 24308 SWAZILAND 192 787 TUNISIA 10087 15310 TANZANIA 3418 3800 TURKEY 22480 49180 TOGO 710 2037 MENA 85224 175931 ZAIRE 2020 2550 ZAMBIA 4062 6575 AUSTRALIA 209978 296532 Africa 108865 185876 AUSTRIA 98919 109500 DENMARK 61086 71059 BARBADOS 1216 1365 FINLAND 28525 47167 COSTA RICA 1688 5604 GREECE 21699 35748 EL SALVADOR 1342 1740 ICELAND 113 2595 GUATEMALA 2554 3237 IRELAND 81761 86787 HONDURAS 1238 2401 ITALY 268500 305443 43 Annex 1 1975 1995 1975 1995 JAPAN 265350 798521 BULGARIA 24133 33900 LUXEMBOURG 4460 5057 CZECHOSLOVAKIA 67794 73496 NETHERLANDS 82501 108142 HUNGARY 40514 53389 NEW ZEALAND 44280 54672 POLAND 163541 207264 SPAIN 132400 157666 Eastern Europe 295982 368048 SWEDEN 50690 120484 U.K. 329469 363707 BANGLADESH 3752 8278 U.S.A. 2854706 3716867 INDIA 411898 1001000 OECD 4534436 6279946 NEPAL 1307 3242 PAKISTAN 19769 58267 South Asia 436727 1070787 Telephone Main Lines 1975 1995 1975 1995 ALGERIA 128900 1176316 TOGO 4596 21715 BENIN 5313 28206 UGANDA 20100 43245 BOTSWANA 5000 59673 ZAIRE 26900 36000 BURKINA FASO 2400 30043 ZAMBIA 28400 76769 BURUNDI 2700 17169 ZIMBABWE 81672 154621 CAPE VERDE IS 1490 21513 Africa 1830771 7082217 CENTRAL AFR.R. 2336 7769 BARBADOS 27000 90132 CHAD 2400 5334 COSTA RICA 81000 557226 COMOROS 450 4510 DOMINICA 2180 17800 CONGO 5600 21410 EL SALVADOR 48100 284777 DJIBOUTI 1551 7556 GRENADA 3050 23200 ETHIOPIA 45250 142452 GUATEMALA 47583 289531 GAMBIA 1470 19202 HONDURAS 17003 160819 GHANA 31259 59978 JAMAICA 49700 291780 GUINEA 6448 10855 MEXICO 1644499 8801030 IVORY COAST 24022 115790 NICARAGUA 21947 96611 KENYA 53000 239639 PUERTO RICO 242900 1195921 LESOTHO 1917 17792 ST.LUCIA 3600 30576 MADAGASCAR 14643 32581 ST.VINCENT & MALAWI 8700 34338 GRE 3170 18236 MALI 3567 17164 TRINIDAD& MAURITANIA 1329 9249 TOBAGO 40800 209310 MAURITIUS 15434 148185 ARGENTINA 1651000 5531700 MOZAMBIQUE 31100 59819 BRAZIL 1800000 12082563 NIGER 3400 13342 CHILE 297000 1884762 REUNION 14900 218723 COLOMBIA 837600 3872847 RWANDA 2300 15000 ECUADOR 165000 748167 SENEGAL 14432 81988 GUYANA 14200 44615 SEYCHELLES 1480 13527 PARAGUAY 29977 166895 SIERRA LEONE 7598 16627 PERU 239000 1109232 SOUTH AFRICA 1156000 3919085 SURINAME 10494 53158 SUDAN 42300 75000 URUGUAY 187000 621996 SWAZILAND 3358 19762 VENEZUELA 501000 2463166 TANZANIA 27056 90270 LAC 7964803 40646050 44 Annex 1 1975 1995 1975 1995 BANGLADESH 54000 286600 NETHERLANDS 3335500 8119999 BHUTAN 570 5243 NEW ZEALAND 987000 1719000 PAKISTAN 208000 2127000 NORWAY 913900 2431272 SRI LANKA 42500 204350 PORTUGAL 778000 3586002 INDIA 1465415 11977999 SPAIN 4697998 15095385 South Asia 1770485 14601192 SWEDEN 4209002 6012999 SWITZERLAND 2470000 4318496 BAHRAIN 14000 140850 U.K. 13229992 29408730 EGYPT 345000 2716212 U.S.A. 80515048 164624240 IRAN 685000 5090363 OECD 180703559 401578885 ISRAEL 597000 2342618 KUWAIT 89000 382287 FIJI 16174 64772 MOROCCO 110000 1158000 TONGA 590 6610 OMAN 3701 169939 VANUATU 1010 4215 QATAR 12300 122701 Pacific Islands 17774 75597 SAUDI ARABIA 138000 1719413 SYRIA 128000 930000 CHINA 1692001 40706032 TUNISIA 65000 521742 HONG KONG 837023 3278287 UNITED ARAB INDONESIA 207500 3290854 E. 25808 672330 KOREA, REP. 1058075 18600216 MENA 2212809 15966455 LAOS 5400 20410 MALAYSIA 169000 3332447 AUSTRALIA 3538998 9199997 MYANMAR 25400 146670 AUSTRIA 1505000 3749087 PHILIPPINES 284000 1409639 BELGIUM 1849960 4632093 SINGAPORE 210390 1429000 CANADA 8277996 17457262 TAIWAN, CHINA 774233 9174816 DENMARK 1706660 3202525 THAILAND 219000 3481996 FINLAND 1353000 2809999 East Asia 5482022 84870367 FRANCE 7098997 32399992 GREECE 1687001 5162774 HUNGARY 508000 1892892 ICELAND 75500 148675 POLAND 1386000 5728497 IRELAND 330000 1310000 ROMANIA 857000 2967957 ITALY 9659995 24854022 TURKEY 680585 13227696 JAPAN 32377012 61105824 Eastern Europe 3431585 23817042 LUXEMBOURG 107000 230512 For each value reported in Table 1, the region's aggregate stock in 1995 is calculated as a multiple of the aggregate stock in 1975. 45