Government_Spending_and_Growth_in_Latin_America_and_the_Caribbean

GOVERNMENT SPENDING AND ECONOMIC GROWTH IN LATIN AMERICA AND THE CARIBBEAN by Rudolph C. Matthias Anthony Birchwood July 2002 INTRODUCTION An important issue in development economics concerns the role governments1 should play in the economy. There are two main views on the role of government. One view is that a smaller role for government will be less distortionary and therefore governments should limit their roles to some core functions.2 This view is associated with the neo-classicals and has gained orthodoxy in recent times. For them the market is the supreme mechanism for the distribution and allocation of resources. Hence, they see government intervention as being distortionary. One of the major concerns of those advocating less government intervention and a smaller role for government is the tendency for ―big‖ governments to run large and persistent fiscal deficits, generate high inflation and accumulate large public debt, which usually generate marginal welfare gains at the cost of creating disincentives for private sector investment. Proponents of public spending, on the other hand, argue that downsizing as an ultimate policy goal makes little sense since governments can play a pivotal role in the economy, through both the size and allocation of their expenditure.3 For example, Musgrave (1997) argues that public finances are needed (a) to provide services where externalities cause market failure, (b) to address issues of distribution, and (c) to share in the conduct of macroeconomic policy. Governments may also play a meaningful role in the economy by spending on maintaining law and order to help provide an environment conducive to private investment, and by spending on education and health to raise the productivity of labour. Hall and Jones (1997, pp. 173) support this latter view of the role of government. They view government activity together with the laws and institutions as an important and integral part of the infrastructure or the economic environment of an economy. They argue that: 1 2 In this study, we define government to include central and local governments. These core functions might include redistributing income, helping to stabilise the economy at full employment levels and addressing issues of market failure and externalities. Here we refer to the amount of government spending as well as the purposes to which the government chooses to allocate its limited resources. 3 1 ―…differences in levels of economic success across countries are driven primarily by institutions and government policies that frame the economic environment in which people produce and transact.‖ This paper examines the impact of government spending on economic growth in 18 Latin American and Caribbean countries over the last three decades, 1970 to 1997. The study attempts to establish, which, if any, components of government spending are positively related to economic growth in these countries and can therefore be encouraged, and which components, from the point of view of social welfare, can be cut with minimal growth costs when fiscal austerity demands. We approach the issue by first looking at the impact of overall government spending on economic growth and then by segregating government spending into four parts to see whether spending on certain social indicators is welfare and growth enhancing. A BRIEF ON GROWTH THEORY AND EVIDENCE The differing views on the role of government in the economy are partly fuelled by the fact that economic theory does not tightly delineate the impact of government spending on economic growth. The orthodox neo-classical theory, for instance, does not explicitly account for the impact of government activity and size on economic growth and development.4 From the preponderance of neo-classical growth models developed over the years, two central themes have emerged from this literature. The first is that in the absence of continuing improvements in technology, per capita growth will eventually cease. Underlying this prediction is the assumption of diminishing returns to capital,5 which includes human capital. The other central theme for these models, the conditional convergence theory, is that the lower the starting level of per capita gross domestic product of a particular country, the higher is its predicted growth rate, holding other factors constant. There is a large body of cross-country evidence, which supports the conditional convergence theory. However, unlike the predictions of the neo-classical theory, most of these studies also provide evidence that countries continue to achieve positive rates of per capita growth over the long term and that these growth rates have no clear tendency to decline. Endogenous growth theories attempt to explain this divergence between the empirical results and the neo-classical theory. The central theme in these competing theories is that growth occurs from the creation of new ideas—technical progress—a crucial missing link in neoclassical theory.6 According to the endogenous growth theories, long-term growth depends not only on technological progress, but also on government policies. On the other hand, structuralists argue that because developing countries are, in most cases, primary exporting countries, which tend to face declining terms of trade, government intervention is needed to change the structure of production in these economies. 7 Policy 4 5 Since the new endogenous growth theory, neo-classical models have been extended to include government. For a discussion of some of these later neo-classical models, see for example, Lucas (1988), Rebelo (1991), Mulligan and Sala-iMartin (1993), Barro and Sala-i-Martin (1995). See for example, Romer, (1990), Aghion and Howitt (1992) and Grossman and Helpman (1991). This view is associated with the early work of Raul Prebisch ( ), and more lately Lance Taylor (1990). 6 7 2 involving the protection of ―infant industry‖ through the erection of trade barriers, the imposition of foreign exchange controls, the maintenance of overvalued exchange rate and the provision of cheap credit and labour are considered important policy measures to promote industrialisation. These theories also view government investment as complementary to private investment. The empirical evidence is not clear-cut, one way or the other as the evidence from cross-country studies of government spending on growth still lacks consensus. Using a sample of 115 advanced and developing countries over the period 1960 to 1980, Ram (1986, pp. 202), for example, argues, ―it is difficult not to conclude that government size has a positive effect on economic performance and growth‖. His study finds that government spending had a positive externality effect on the rest of the economy, and perhaps more interestingly, that government spending in the 1960s had higher relative factor productivity than the rest of the economy. Kormendi and Meguire (1985) find supporting results. In contrast, using a sample of 86 countries, Landau (1983) finds that the larger the size of the government sector in the economy the lower the rates of economic growth. This finding is, for the most part, corroborated by his later study [Landau (1986)], which undertakes a more comprehensive examination of the impact of various components of government spending on economic growth in developing countries. Evidence in support of a negative effect of government expenditures on growth is also provided by Barro (1991), who studied a sample of 98 countries. One of the factors that account for the differences in these empirical results is the way in which government activity is measured. While some studies focus on the overall size of government spending (Landau, 1993), some focus on components of government spending (Barro, 1991; Landau, 1986), while others focus on the change in government spending (Ram, 1986). The complexity of the problem involved in the empirical studies in this area is underscored by the work of Levine and Renelt (1992). This study investigates whether the links between growth and economic, political, and institutional factors are robust to small changes in the information set. They find that apart from investment and measures of international trade, almost all the variables used to explain cross-country growth rates are not robustly related to growth. In contrast, Sala-iMartin (1997) adds to this debate by showing that there is a substantial number of institutional, economic and political variables that are robust. In the next section, we discuss the framework within which we re-examine these issues. GOVERNMENT SPENDING AND GROWTH: THE ANALYTICAL FRAMEWORK The framework we employ for analysing the role of government spending in relation to economic growth is a synthesis of the Barro (1996) and Tanzi and Schuknecht (1997) models. The Barro (1996) framework is an extended version of the neoclassical model to include variables designed to account for differences in institutional factors across countries. The model can be represented as: y = f(y, y*), (1) 3 where y is the growth rate of per capita output, y is the current level of per capita output and y* is the long run or steady-state level of per capita output. The target variable, y* depends on an array of choice and environmental variables. Equation 1 suggests that for any given level of initial per capita income, y, an increase in the steady state income, y*, raises per capita income, y. Within this extended neoclassical framework, the private sector has to make decisions about savings and labour fertility among other things, while the government faces different types of spending and revenue options, market distortions, property rights, rule of law and political freedoms. Terms of trade issues are also likely to be important in the case of open economies. Within this framework, governments have significant potential to influence the long-term (steady state) growth rate of the economy. For example, if the government creates a climate that is conducive to business investment, by say, maintaining the rule of law and reducing corruption and ―red tape‖, the steady-state growth rate, y*, can increase over a transitional interval. As the growth rate rises, diminishing returns pushes it back to its steady-state level determined by the rate of technological progress. However, because the transitional period can be fairly long the growth effects from shifts in government policy can persist for a very long time. The endogenous growth theories show how research and development can lead to technological progress, and how once there is no tendency to run out of ideas, that technical progress can generate positive growth rates over the long-term. However, the rate of growth and underlying amount of inventive activity in any economy tend not to be Pareto optimal because of distortions relating to the creation of new goods and production methods (Barro, 1996). It is for this reason that government policy relating to taxation, maintenance of law and order, infrastructure, protection of intellectual property rights, and regulation of international trade among other things becomes important in helping to achieve the potential long-term rate of growth of the economy. Tanzi and Schuknecht (1997) also shed some light on the role of government policy in economic development. Their analysis attempts to establish whether there is a continuous positive relationship between higher government spending and higher social welfare or whether there are diminishing returns in terms of welfare gains to such spending. In addressing this complex issue, they model social welfare, , as a function of several indicators, , as: f ( 1, 2,..... n) n such that, i 1 f i i Within this framework, governments can influence social welfare through its spending and other policies in that the greater the positive effect of public spending on these indicators, the greater the improvement in social welfare. Changes in a wide range of socio-economic indicators may provide evidence of changes in social welfare. Tanzi and Schuknecht (1997) themselves point out some of the obvious limitations implicit in this model. First, more public spending often generates an opportunity cost in terms of private spending forgone, either because government collects higher taxes to finance its spending or that government borrowing crowds out private investment. Secondly, it is difficult, if not impossible, to consider all the different social indicators on which governments may choose to spend. In 4 essence, the model attempts to assess whether public spending is associated with positive improvements in indicators that are assumed to influence social welfare without considering the opportunity costs, and without applying weights to capture the importance of such indicators. In spite of the obvious limitations of the Tanzi and Schuknecht (1997) model, we can combine this with the Barro (1996) type extended neoclassical model to evaluate the impact of government spending on economic growth by examining spending on various socioeconomic indicators. In this study, we focus on government spending on social indicators such as: (a) defence which captures the government policy on national security and defence against threats from internal and external sources, and the general maintenance of law and order; (b) education and health which together capture government’s contribution to raising labour productivity. MODEL AND DATA DESCRIPTION Using the analytical framework discussed above, we attempt to explain the determinants of growth for 18 Latin America and Caribbean countries, over the period 1970 to 1997. The variables in our model are computed over the following six periods: 1970-74, 1975-79, 1980-84, 1985-89, 1990-94 and 1995-97. Most variables are calculated as averages over each period, while the others are recovered from the first year for each of the six periods. For example, the birth rate is observed only once in five years and hence the value for Birth for 1970-74 is the value in 1970, the only year in this period when the variable was measured. Our dependent variable is the growth rate of gross national product per-capita (GNY), averaged over each period. Apart from the main variables capturing government spending in our model, the other explanatory variables are similar to those used in Barro (1996) and other similar cross-country studies. We include the initial level of real per-capita GDP (GDP70) calculated as per capita GDP using the purchasing power parity method averaged over the first five-year period, 197074, and denominated in US dollars, (GDP70). The coefficient’s magnitude indicates the rate at which the economy approaches its long-term position when the other variables are held constant. Apart from the initial level of income, there are three other social variables included in our model. The first is the crude birth rate per 1,000 people (Birth), used as a crude proxy for the growth rate of the labour force, which should correlate positively with growth. The second is life expectancy at birth (Life) expressed in number of years, to proxy for health and more generally, quality of life and is expected to correlate positively with growth. The final social indicator is the dependency ratio (DEPEN), calculated as the age dependency of young and old to the working age population. High dependence ratios imply that there is a sizeable proportion of the population that does not work and hence does not contribute to output growth. We therefore expect a negative relationship between dependence and per capita growth rates. Besides government spending, there are three economic variables in the model. We include the annual rate of inflation (INFL) calculated as the percentage change in the GDP deflator. This variable should capture the negative effects on growth of expansionary monetary policies and 5 external price shocks. The investment to GDP ratio (GDI)8 is included to account for the positive effects of savings and investment on output growth. And the ratio of the value of export of goods and services to GDP (EXPORT) is included to account for the positive impact of international trade on the growth rate of the open economies in our sample. Our model can therefore be written in a general form as: y = f(social indicators, economic indicators, indicators of government spending) We estimate two regression equations, each of which contains the four social and three economic variables just defined. In addition to these seven variables, we include in the first equation general government consumption to GDP (GOVC), which is our main focus in this equation. General government consumption includes all current expenditures for the purchase of goods and services by all levels of government and capital expenditure on national defence and security but excludes spending by most government enterprises. Landau (1983, 1986) and Barro (1996) use a similar measure of government to account for the effects of government size on growth. In the second equation, we also include the seven social and economic variables as well as the four components of government spending: defence (DEFEXP), education (EDUEXP), health (HEALTHEXP) and other government expenditure (OTHGOVEXP). Other government expenditure is calculated as total government expenditure minus amounts allocated to defence, education and health. Each of these four variables is scaled by total government expenditure. We include both linear and squared terms for the first three government-spending variables. The squared terms are supposed to account for non-linear effects. The three categories defence, education and health are the components of public spending, that we believe are associated with positive improvements in social welfare. If government outlays on defence, education and health are also growth enhancing, we expect them to correlate positively with growth. Having removed spending on these social indicators, we do not expect other government outlays to enhance productivity and therefore we expect this component to correlate negatively with economic growth. In the next section, we present the empirical evidence. First, we present some evidence on the size of government across the globe, looking at size by region and countries, and by income groupings. In the next section, we present some evidence to show the relationship between government spending and changes in several indicators of social welfare in Latin America and the Caribbean. In the final section, we present the econometric results from our two models that relate changes in growth rates to the social, economic and government spending indicators. 8 Investment here is gross domestic investment. 6 EMPIRICAL EVIDENCE Government size across the globe Before we focus on the Latin America and Caribbean area, it is useful to put into context government size in this region by showing how it compares with that of the other regions of the world.9 This information is presented in Table 1, which shows average government consumption to GDP over six time periods, for the World, and deviations from the world’s average for various regions and countries. The results suggest that government consumption to GDP averages 14.9 per cent around the world and appears to have risen slightly from 13.7 per cent of GDP in the early 1970s to about 15.0 per cent in the late 1990s. On average, government size is bigger in the richer regions and countries of the world. For instance, the size of governments in South Asia (-4.9%), East Asia and Pacific (-3.6%) and Latin America and The Caribbean (-3.7%) is below world average, whereas Canada (7.1%), United Kingdom (5.9%) and the United States (2.4%) are above the world’s average. Canada has the biggest government in the sample; compared to the size of government in Latin America and the Caribbean, Canada’s is twice as large. Table 1: Government consumption to GDP averaged over five-year periods for regions This table shows government consumption to GDP ratios for the World, various regions and countries. The world column gives the government consumption to GDP, while the observations for regions and countries are deviations from the world average. The Grand average is the average over the six time periods. Latin America East Asia & South Asia Sub-Saharan & Caribbean Pacific Africa Middle East & North 3.3 Africa Canada United Kingdom United States World 1970-74 -3.3 -4.7 1975-79 -4.2 -3.9 1980-84 -5.1 -1.7 1985-89 -4.5 -3.1 1990-94 -2.6 -4.0 1995-97 -1.6 -4.5 Grand Average -3.7 -3.6 Source: World Bank Development Indicators -4.6 -5.5 -5.7 -3.6 -4.8 -5.1 -4.9 -1.0 -0.1 -0.2 1.5 2.3 1.6 0.6 6.3 6.3 7.7 4.2 2.7 5.4 6.9 7.4 7.0 6.6 7.9 6.0 7.1 5.1 6.5 6.8 5.4 5.9 5.9 5.9 4.0 2.6 2.2 2.7 1.5 0.6 2.4 13.7 14.6 15.3 15.1 15.7 15.2 14.9 A similar picture emerges when we look at government consumption by income groups. Table 2 is similar to Table 1 in that it shows the average government consumption to GDP for the average income group and deviations from this average for each income group. These results suggest a positive relationship between government consumption and wealth. Countries in the low to upper middle-income groups have lower government consumption to GDP ratios than the overall average and lower than those in the high-income groups. The results in this table lends support to those in Table 1, which show that government size in Canada, the United Kingdom and the United States is bigger than that in Latin American and The Caribbean and other poorer regions of the world. The difference between government spending to GDP of poor (low income) and rich (high-income OECD) is about 5.5 percentage points, in favour of rich countries. 9 These regions are based on World Bank’s classification. 7 Table 2: Government consumption to GDP averaged over five-year periods by income groups Low income Low & middle income Lower middle income -1.8 -0.8 0.2 -0.9 -1.5 -1.7 -1.0 Middle income -1.2 -0.8 -0.8 -0.9 -0.3 0.0 -0.7 Upper middle income -0.7 -0.8 -1.6 -1.0 0.8 1.4 -0.4 High income: non-OECD 4.2 3.0 3.2 2.6 2.5 1.8 2.9 High income: OECD 2.6 2.7 2.6 2.3 1.5 1.9 2.3 Average for all income groups 11.7 12.6 13.3 13.4 14.5 13.7 13.210 1970-74 -1.9 -1.2 1975-79 -2.4 -1.0 1980-84 -2.6 -1.0 1985-89 -1.1 -1.0 1990-94 -2.5 -0.5 1995-97 -3.1 -0.4 Grand Average -2.2 -0.9 Source: World Bank Development Indicators The size of government in the richest countries of the world suggests that public expenditure increases with the level of development. This positive relationship between public spending and development—measured by income levels—suggests that public spending is a superior good. A recent study by ECLAC (1998)11 notes that the high level of public spending in rich countries is strongly associated with the development of social security, which is reflected by the increasing proportion of the public spending allocated to transfers and subsidies, and health. In contrast, Agénor and Montiel (1996) presents evidence to show that compared to their industrial counterparts, developing countries devote a much larger fraction of their spending directly on production of goods and services. Government spending and social welfare in Latin America and the Caribbean We can examine the efficiency and quality12 of government spending by looking at selected indicators for 18 countries in Latin America and the Caribbean over the period 1970 to 1997. Table 3 reports a number of indicators, which suggest that the countries are a diverse group. For instance, the average per-capita income in the 1970-74 period ranges from US$ 1,400 to US$ 4,100. Over the period, the growth rates in these countries range from –1.2 per cent to 3.1 per cent. Moreover, there are large variations across countries in terms of government size and spending on defence, education, and health. 10 The average for all income groups is different from the average for the world probably because some of the countries included in the world average are not included in the income groups shown in this table. ECLAC is an acronym for the Economic Commission for Latin America and the Caribbean. When we talk about the quality of government expenditure, we consider things such as the productivity gained by spending on management that is efficient and results oriented, spending on projects that are sustainable, and decentralising the management of government spending and projects. 11 12 8 Table 3: Country averages for variables used in regression equations for 18 Latin American and Caribbean countries over five-year periods between 1970 and 1997. GNY is the growth rate of gross national product (GNP) per capita. BIRTH is the crude birth rate per 1,000 people; DEPEN is the dependency ratio, calculated as the age dependency of young and old to the working age population. Life is the life expectancy at birth expressed in number of years. GDP70 is the per capita GDP average over 1970-74 using the purchasing power parity method and expressed in US dollars. INFL is the annual inflation rate calculated as the percentage change in the GDP deflator. GDI is gross domestic investment to GDP. EXPORT is the ratio of the value of export of goods and services to GDP. GOVC is government consumption divided by GDP. General government consumption includes all current expenditures for purchases of goods and services by all levels of government, excluding most government enterprises. It also includes capital expenditure on national defence and security. DEFEXP is defence expenditure calculated as the percentage of total government expenditure on defence. EDUEXP is education expenditure to total government expenditure; HEALTHEXP is expenditure on health as proportion of total government expenditure. OTHGOVEXP is other government expenditure, the sum of total government expenditure minus expenditure on defence, education and health to total government expenditure. GNY BIRTH DEPEN LIFE GDP70 INFL GDI EXPORT GOVC DEFEXP EDUEXP HEALTHEXP OTHGOVEXP Argentina 1.0 0.62 22.4 70.3 4,146 305.7 21.3 8.1 6.6 7.3 8.5 4.3 79.9 Barbados 1.2 0.65 16.8 73.1 3,928 7.1 20.0 57.2 16.9 1.8 21.4 12.6 64.2 Belize 2.7 0.99 36.3 70.2 1,226 4.9 24.2 54.0 17.4 4.7 16.4 9.1 69.7 Brazil 2.4 0.71 27.7 63.6 2,508 394.9 21.5 8.6 13.1 5.1 4.5 6.7 83.6 Chile 2.8 0.65 23.2 70.4 2,666 77.7 20.6 24.9 12.3 10.1 13.5 8.8 67.5 Colombia 2.1 0.77 29.5 66.5 2,114 22.3 19.4 15.5 11.0 8.6 21.4 6.9 63.0 Costa Rica 1.4 0.76 28.4 73.1 2,636 19.5 25.1 35.2 15.8 2.7 22.8 16.7 59.0 Dom. Rep. 3.1 0.80 31.7 65.8 1,544 14.7 22.9 29.0 6.9 7.0 12.1 10.3 70.7 Ecuador 2.2 0.83 33.2 64.9 1,946 27.3 21.4 26.0 11.6 14.1 25.9 8.2 51.8 El Salvador 0.4 0.89 34.5 61.9 1,392 11.4 16.4 24.8 11.7 11.9 18.4 8.5 61.2 Guatemala 1.0 0.94 40.9 58.8 1,800 12.2 15.0 18.3 6.8 13.1 15.9 8.8 62.2 Mexico 1.5 0.87 32.5 67.9 3,098 35.3 22.1 15.8 9.2 6.8 17.3 2.6 73.3 Nicaragua -1.2 0.98 41.5 61.1 1,662 973.6 20.6 26.5 19.1 15.4 14.4 10.4 59.8 Paraguay 2.3 0.89 35.5 67.5 1,382 16.9 22.9 19.4 7.5 13.0 12.2 3.9 71.0 Peru 0.8 0.80 33.2 62.2 2,132 372.0 22.2 14.8 10.1 14.3 18.1 5.2 62.4 T&T 3.0 0.69 24.2 69.3 2,830 9.2 22.5 44.3 14.9 1.7 12.5 7.1 78.7 Uruguay 1.7 0.60 18.9 71.3 3,112 55.4 15.9 20.0 13.8 9.3 10.2 4.1 76.4 Venezuela -0.2 0.78 30.9 69.4 4,776 29.2 24.3 26.9 10.0 6.5 17.1 7.1 69.5 Sources: World Bank Development Indicators on CDROM. Government spending on defence, education and health came from Government Finance Statistics Yearbook, IMF, various years. Table 4 and Figures 1 and 2 shed further light on the indicators over time for our sample countries. As the table shows, when we take cross-sectional averages, the components of government spending on defence, education and health grows relatively slowly over the years. As an example, spending on education has been quite stable, deviating from the mean of 15.3 percent (expressed as a percentage of total expenditure) by only 1.4 per cent over time. Table 4: Averages over five-year periods for variables used in regressions for 18 Latin American and Caribbean countries GNY BIRTH DEPEN LIFE INFL GDI EXPORT GOVC DEFEX EDUEXP HEALTHEXP OTHGOVEXP 1970-74 3.8 0.90 35.2 61.8 29.3 20.9 21.9 11.0 P 8.3 16.6 7.2 68.0 1975-79 2.4 0.86 33.3 64.3 38.5 23.4 24.0 11.3 9.0 16.4 6.0 68.6 1980-84 -1.9 0.81 31.3 66.2 44.1 21.4 24.7 13.5 8.8 15.5 8.8 66.8 1985-89 0.7 0.77 29.0 68.5 354.0 19.7 26.0 12.8 8.6 13.8 6.2 71.5 1990-94 1.8 0.71 27.1 69.8 314.6 19.9 26.8 11.3 11.1 14.1 9.0 67.5 1995-97 2.5 0.67 24.3 71.4 15.9 20.6 28.1 10.9 5.9 15.2 10.5 69.5 Grand 1.57 0.79 30.01 67.04 132.73 20.96 25.26 11.82 8.74 15.28 7.87 68.57 Average Source: World Bank Development Indicators on CDROM and Government Finance Statistics Yearbooks. 9 Figure 1 suggests that government consumption and investment have also been stable over the years, with government consumption averaging around 12 per cent of GDP and investment 21 per cent of GDP (Table 4). However, the figure also exhibits a weak negative relationship between these two indicators: In periods when investment spending is comparatively high (197579 and 1995-97), government consumption is comparatively low. When the latter is relatively high in 1985-89, the former is comparatively low. This could be interpreted in several ways. One possible explanation is that government spending might be playing the role of automatic stabiliser in the recession period in the 1980’s (see Figure 1) when government consumption appears comparatively high. It might also be, as ECLAC (1998) argues that public spending in the region emphasizes size rather than the quality of spending, hence public spending crowds out private investment rather than create conditions for higher private spending. Figure 1: Investment and Government S pending 25.0 Per cent of GDP 20.0 15.0 10.0 5.0 0.0 1970-74 1975-79 1980-84 GDI 1985-89 GOVC 1990-94 1995-97 The evidence nonetheless indicates some welfare gains from government spending over the years. Figure 2 shows that over the last two decades, there were improvements in life expectancy (or quality of life) and lower rates of fertility, which Barro and Lee (1994) argue is a measure of prosperity and a sign of higher levels of primary education, especially among females. To get some idea of whether there are significant welfare gains associated with higher income (wealth) and government spending in the Latin American and Caribbean countries in our study, we conduct difference of means tests, the results of which are reported in Table 5. The approach taken in testing the significance in the differences of means is to create two groups, based on the lower and upper extremities in certain parameters, to see whether there are significant differences with respect to various indicators between the two groups. In the first panel of the table, our two groups were segmented on initial per-capita income levels (GDP70). Comparing the groups of high and low per-capita income, the results suggest that the average dependency ratio and the crude birth rate are both lower and life expectancy is significantly higher for countries in the high per-capita income group compared to those in the low-income group. These results are consistent with the significantly higher levels of public spending on education, but probably not defence, by the higher income group than the lower income group. Several of the other indictors for the high-income group appear to be better, although the difference is not statistically significant. 10 Table 5: Difference of means test based on GDP70, GNY and GOVC for Latin American and Caribbean GNY BIRT DEPE LIFE GDP7 INFL GDI EXPOR GOV DEFEX EDUE HEALTH OTHGOVE Mean, Low 1.49 0.87 35.1 64.2 1,689 161.7 20.4 TT 24.3 11.1 11.2 XP 16.8 H N 0 C P EXP 8.2 XP 63.8 Mean, High 1.65 0.70 25.0 69.8 3,300 103.8 21.5 26.2 12.5 6.4 13.8 7.6 73.2 GDP75 GDP75 Significance * * * * * ** * level Mean, Low 0.65 0.81 31.2 66.4 2,841 196.2 20.8 24.7 11.8 9.0 16.8 8.4 66.6 growth Mean, High 2.49 0.76 28.8 67.7 2,148 69.2 21.1 25.8 11.9 8.5 13.6 7.3 70.6 growth Significance * * ** *** level Mean, Low 1.54 0.81 32.2 65.9 2,549 92.8 21.3 19.3 9.7 15.7 6.4 68.1 8.9 GOVC Mean, High 1.60 0.76 27.8 68.2 2,440 172.6 20.6 31.5 14.9 7.8 14.9 9.2 69.0 GOVC Significance * ** * * * level Source: World Bank Development Indicators on CDROM and Government Finance Statistics Yearbooks. In the second panel of Table 5, we segment the countries based on per-capita growth rates. These results show that countries with high growth rates over the period 1970 to 1997 also have lower income levels in 1970-74, supporting the neo-classical convergence hypothesis. Surprisingly, however, high growth countries spent significantly less on education than low growth countries, but are not any worse off on many of the indicators of social development. What these two sets of findings suggests is that, although wealth (measured by per-capita income) helps predict social development levels, the level of social development in each country changes very slowly. It does not vary significantly with contemporaneous growth rates. In the third segment of Table 5, we show two groups based on government consumption spending. These results suggest that countries with high government consumption have higher life expectancy and lower dependency ratios. Countries with high government spending also spent significantly more on health care, which might explain their better health status. It is not clear to us, however, why countries with higher government consumption also have higher export ratios. One possible explanation is that a sizeable proportion of government consumption in these countries could have been used to provide greater incentives and subsides to the private 11 sector to encourage exports. This might also explain their higher per capita growth rates, although not statistically significant, given the positive impact of higher exports on growth. In summary, this section presents evidence to suggest that in Latin America and the Caribbean, the richer countries and those with higher public spending tend to have higher rates of social development, measured by higher life expectancy, lower birth rates and dependency ratios, and high rates of per-capita and export growth over the period 1970 to 1997. Regression results In this section, we evaluate the results from our two regression models. Before doing so, it is useful to first examine the simple correlation between the dependent and explanatory variables used in our models. These correlation coefficients are presented in Table 6, which shows some fairly strong correlation between per-capita growth and investment, inflation and government consumption respectively. However, there does not appear to be any strong correlation between output growth and any of the four components of government spending, even though the relationship between growth and total government consumption seems relatively strong (0.24). One can also see this in the scatter plots presented in Figure 3. The graphs in this figure show the relatively weak negative relationships between output growth and government consumption and output growth and government spending on defence, education and health, respectively. Table 6: Correlation coefficients for variables used in regression equations for 18 countries, 1970-1997. GNY BIRTH DEPEN LIFE GDP75 INFL GDI EXPORT GOVC DEFEXP EDUEXP HEALTHEXP GNY BIRTH DEPEN LIFE GDP75 INFL GDI EXPORT GOVC DEFEXP EDUEXP HEALTHEX P OTHGOVEX P 1.00 0.01 -0.06 0.06 -0.08 -0.29 0.24 0.08 -0.29 -0.07 -0.04 -0.04 0.10 1.00 0.90 -0.68 -0.51 0.04 0.05 -0.05 -0.01 0.29 0.29 -0.02 -0.38 1.00 -0.80 -0.56 0.05 0.08 -0.23 -0.11 0.37 0.23 0.00 -0.39 1.00 0.43 -0.08 0.17 0.41 0.14 -0.37 -0.06 0.17 0.24 1.00 -0.04 0.11 0.01 -0.01 -0.34 -0.11 -0.14 0.34 1.00 -0.03 1.00 -0.20 0.19 0.40 0.08 0.18 -0.23 -0.31 0.08 -0.04 0.09 0.10 0.09 1.00 0.33 -0.36 0.38 0.46 -0.23 1.00 -0.05 0.22 0.28 -0.20 1.00 0.03 -0.13 -0.53 1.00 0.24 -0.76 1.00 -0.51 The regression results using these variables in our two models are presented in Tables 7 and 8. The results are derived from panel data analysis for 18 Latin America and Caribbean countries using data from six five-year periods between 1970 and 1997.13 Our dataset is unbalanced, with at most 103 observations on each variable in the model. Both tables present six different sets of coefficient estimates for each model. The first two columns of the tables present ordinary least squares (OLS) estimates in levels and first differences. Modelling the problem in this way ignores differences in the intercept over time and between countries. In effect, we are running ordinary least squares on the pooled sample, with a common intercept and slope. 13 The last period uses data from 1995 to 1997. 12 Figure 3: Scatter plots of the growth rate of GNP per capita on government consumption to GDP, government spending on defence, education and health for five-year periods 1970 to 1997. In each graph, the per capita GNP (GNY) is plotted on the y axis. GNY is plotted against government consumption (GOVC), defence expenditure (DEFEXP), education expenditure (EDUEXP), and health expenditure (HEALTHEXP) in the four graphs. 10 5 0 -5 5 10 15 GNY 10 5 0 -5 20 GOVC 25 30 0 5 10 GNY 15 DEFEXP 20 25 10 5 0 -5 5 10 15 GNY 10 5 0 -5 20 EDUEXP 25 30 35 5 10 GNY 15 20 25 HEALTHEXP The next two columns in the tables report fixed effects estimates. The first fixed effects estimator uses the least squares dummy variable approach (LSDV),14 while the second is generated by a within effects transformation.15 The fixed effects approach used here allows us to take advantage of the country specific effects, or to put another way, they allow us to control for the cross-sectional dimensions in the data. Another way of accounting for country effects is to generate random effects estimators (GLS), reported in the final two columns of the tables. The random effects estimators are weighted least squares estimators. The first uses weights from within and between transformations, while the other uses weights from the OLS residuals.16 We note that the fixed effects, but not the random effects estimators remain unbiased even if our assumption that the country effects are uncorrelated with the error term does not hold. This assumption is critical; if there is any systematic correlation between the error term and effects included in the model, the estimator cannot determine how much of the change in the dependent variable associated with an increase in the included effects to assign to the coefficient versus 14 The LSDV approach involves introducing dummies to take account of the country effects. This means that, compared to the other regression estimates, we are loosing at least 17 degrees of freedom; one less than the 18 countries in our study. The within transformation is another fixed effects estimator. In generating estimates using this approach, the country effects are ―swept‖ out of the model in the transformation process. This within estimator therefore does not estimate country effects. The fixed effects (LSDV and Within) and random effects (GLS) estimators differ in terms of their efficiency and consistency. To estimate these effects, we have to make one of two assumptions about the distribution of the cross sectional units. The fixed effects estimators are based on the assumption that these cross-sectional effects have no distribution, in which case we treat them as fixed quantities to be estimated. Inferences from this model are therefore restricted to the behaviour of the particular countries, over the specific time period observed. That is, the model is analysed conditional on the effects that are present in the sample. In contrast, the random effects estimator assumes that the country effects are random and the model is to be analysed unconditionally. 15 16 13 how much to attribute to the unknown country effects. For this reason, while Tables 7 and 8 present results of six estimators, we focus on the fixed effects results from the LSDV approach.17 Table 7 reports results from the first model, which includes the seven social and economic indicators, and the government consumption variable, on which we focus. Most of the seven social and economic variables have the correct signs, based on the theory and earlier empirical findings, but are not statistically significant at conventional levels. Investment is found to be positively and significantly related to growth, a finding consistent with other empirical work in this area.18 We also find, as expected, that inflation, dependency ratio, and the initial level of income correlate negatively with output growth. In addition, exports, life expectancy and birth rate correlate positively with growth. However, in our preferred model (LSDV), only the coefficients on investment and birth rate are statistically significant at conventional levels. Our main interest in estimating these equations is, of course, to see the impact of government spending on economic growth. The results from this model indicate that government spending correlates negatively and significantly with growth.19 This is the case not only in our chosen model, but also in all of the other five equations, which indicates the robustness of the results. Our results are also consistent with those of Barro (1996), Landau (1983, 1986), and Summers and Heston (1988), among others, who find an inverse relationship between government size (measured by government consumption to GDP) and economic growth. Table 8 presents estimates for model 2 for the seven social and economic variables and the four components of government spending. Again, our primary focus is on the LSDV results. These results show that investment and the birth rate remain robustly positively related to growth, while inflation remains negative, but now becomes significant. Apart from the initial level of income, none of the other variables enter the model significantly. Indeed, in most cases, their standard error is larger than the estimate, which leaves little room for confidence in the estimate. Interestingly, however, the initial level of wealth becomes positive and significant in this model. This is inconsistent with our explanation of convergence from the results in Table 7. 17 The results from the fixed effects estimator using the within transformation are similar to those of the LSDV, but because there is little variation over time for some of the variables of interest, they are swept out of the model and hence do not have an estimate. That is why we ignore the results of the within transformation. See Levine and Renelt (1997) for a review of empirical findings in studies of determinants of cross-sectional growth. It was pointed out to us that if we were to divide government consumption into current and capital spending, we might have observed a positive relationship between growth and capital expenditure and a negative relationship between growth and current expenditure. This suggests that government capital spending to enhance the basic infrastructure in the country should correlate positively with growth. This is interesting and debatable, but is not the central issue in this paper. We intend to address this issue in our next paper on the subject. 18 19 14 Table 7: Model 1 panel data results regressing the per-capita growth rate of GNP on seven socio-economic variables and government consumption for Latin America and the Caribbean, 1970 to 1997. The dependent variable in each equation is the growth rate of gross national product (GNP) per capita. The sample is a panel dataset for 18 Latin American and Caribbean countries over the period 1970 to 1997. The dependent and independent variables in the regressions are computed over the five-year periods, beginning in 1970-74, 1975-79…1995-97. There are, at most, 6 observations on each variable. The variables are as defined in Table 3. Country effects are estimated for some equations but not reported. The asterisks represent significance levels for the t-test. They are to be interpreted as * significant at the 0.01 level, ** significant at the 0.05 level, and *** significant at the 0.10 level. T-statistics are in ellipses. Standard errors are robust standard errors. OLS Coefficie nt10.9624 Birth Rate (2.600) Dependency Ratio -0.3688 (-3.46) Life Expectancy -0.1199 (-1.16) GDP per capita, 1970-74 -0.0008 (-3.09) Inflation -0.0012 (-3.32) Gross Domestic Invest. 0.2790 (2.82) Exports -0.0013 (-0.06) Government -0.2035 Consumption (-3.36) Constant 10.9337 (1.54) R-squared No of observations F-test 0.288 103 2.4 OLSa LSDV Within Coeffici Coeffici Coeffici ent 13.6471 *** ent 15.4947 ** ent 15.4947 ** (1.750) (2.060) (2.060) -0.3849 -0.1642 -0.1642 (-1.560) (-0.863) (-0.863) 0.1545 0.1896 0.1896 (0.332) (0.690) (0.690) -----0.0051 ----(-0.863) -0.0016 ** -0.0010 -0.0010 (-2.410) (-1.580) (-1.580) 0.3585 * 0.2903 ** 0.2903 ** (2.960) (2.210) (2.210) -0.0149 0.0006 0.0006 (-0.130) (0.008) (0.008) -0.2907 *** -0.3169 ** -0.3169 ** (-1.700) (-2.480) (-2.480) -0.7542 --------(-1.060) 0.258 85 102.0 0.365 103 167.6 0.285 103 164.2 GLSb Coeffici ent 7.5108 *** (1.770) -0.3196 * (-3.610) -0.1078 (-1.610) -0.0009 * (-4.150) -0.0014 ** (-2.080) 0.2498 * (4.110) -0.0049 (-0.252) -0.1412 ** (-2.340) 11.5433 ** (2.320) 0.380 103 55.7 GLSc Coeffici ent 10.9624 ** (2.310) -0.3688 * (-3.430) -0.1199 (-1.190) -0.0008 ** (-2.280) -0.0012 *** (-1.690) 0.2790 * (3.780) -0.0013 (-0.047) -0.2035 * (-2.640) 10.9337 (1.320) 0.288 103 38.1 * * * * * * In Table 7, the following applies: OLS is ordinary least squares estimator, LSDV is least squares dummy variable estimator, GLS is Generalized least squares estimator, Within is the within transformation estimator. Also: a OLS regression on the first difference of each variable; b Generalized least Squares weighted by between and within estimates; c Generalized least squares weighted by OLS residuals. The results for government spending in Table 8 support our decision to model both linear and non-linear effects of government spending on growth. The results suggest that low levels of government spending on defence is negatively correlated with output growth, mainly because of the opportunity cost of such spending. However, higher levels of spending appear to generate a positive impact, but this effect does not seem large enough to offset the negative effects of spending on the military at the lower levels. Spending on health care also seems to generate some negative influence on growth, whether this spending is at a low or high level. On the other hand, government spending on education at low levels seems to have a positive influence on growth, but this effect seems to turn negative at higher levels of spending. Finally, as we expect, other government spending is significantly negatively correlated with per-capita growth. 15 Table 8: Model 2 panel data results regressing the per-capita growth rate of GNP on seven socio-economic and four government spending variables for Latin America and the Caribbean, 1970 to 1997. The dependent variable in each equation is the growth rate of gross national product (GNP) per capita. The sample is a panel dataset for 18 Latin American and Caribbean countries over the period 1970 to 1997. The dependent and independent variables in the regressions are computed over the five-year periods, beginning in 1970-74, 1975-79…1995-97. Hence, there are, at most, 6 observations on each variable. The dependent variables are as defined in Table 3. Country effects are estimated for some equations but not reported. The asterisks represent significance levels for a t-test. They are to be interpreted as * significant at the 0.01 level, ** significant at the 0.05 level, and *** significant at the 0.10 level. Standard errors are robust standard errors. OLS Coefficient 9.522 (1.580) -0.333 ** (-1.890) * -0.040 (-0.332) -0.001 ** (-2.640) -0.003 * (-3.830) 0.237 ** (2.410) -0.070 * (-3.200) -0.148 (-0.839) 0.006 (1.160) 0.239 (1.030) -0.006 (-1.430) 0.375 * (3.580) -0.015 * (-3.500) 0.048 (0.531) OLSa Coefficient 19.340 ** (2.440) -0.236 (-0.958) 0.669 (1.510) LSDV Coefficient 18.119 ** (2.130) 0.009 (0.040) 0.429 (1.350) 0.004 ** (2.330) -0.002 ** (-1.800) * 0.248 ** (2.180) 0.010 (0.142) -1.144 ** (-2.400) 0.018 (1.560) 0.293 (0.796) -0.023 * (-4.930) -0.333 (-0.766) -0.018 ** (-2.350) -0.680 ** (-2.150) Within Coefficient 18.119 ** (2.130) 0.009 0.040) 0.429 1.350) GLSb Coefficient 9.908 ** (1.830) * -0.350 ** (-2.640) -0.036 (-0.335) -0.001 * (-3.040) -0.003 ** (-1.730) * 0.223 ** (2.420) -0.069 ** (-1.760) * -0.171 (-0.712) 0.007 (0.772) 0.205 (0.746) -0.005 (-0.693) 0.365 ** (1.720) * -0.014 (-1.620) 0.061 (0.699) GLSc Coefficient 9.522 (1.640) -0.333 ** (-2.340) -0.040 (-0.315) -0.001 ** (-2.350) -0.003 (-1.570) 0.237 ** (2.500) -0.070 (-1.610) -0.148 (-0.570) 0.006 (0.629) 0.239 (0.799) -0.006 (-0.800) 0.375 (1.610) -0.015 (-1.580) 0.048 (0.457) Birth Rate Dependency Ratio Life Expectancy GDP per capita, 1970-74 Inflation Gross Domestic Invest. Exports Defence Expenditure Defence Expend. Squared Education Expenditure Education Expend. Squared Health Expenditure Health Expend. Squared Other Govt. Expend. Constant -0.003 (-1.810) 0.372 (2.870) -0.020 (-0.178) 0.000 (0.000) 0.013 (1.080) 1.144 (2.210) -0.017 (-2.480) 0.517 (1.460) -0.010 (-1.210) 0.377 (1.130) -1.197 (-1.560) ** * * -0.002 -1.800) 0.248 2.180) 0.010 0.142) -1.436 -3.670) 0.018 1.560) ** * ** * ** ** -0.023 (-4.930) -0.626 (-2.130) -0.018 (-2.350) -0.972 (-4.770) * ** ** * R squared 0.23 0.27 0.44 0.30 0.29 0.23 Number of observations 78 60 78 78 78 78 F-statistics 6,145.0 1,334.0 1,457.0 1,601.0 46.6 34.4 In Table 8, the following applies: OLS is ordinary least squares estimator, LSDV is least squares dummy variable estimator, GLS is Generalized least squares estimator, Within is the within transformation estimator. Also: a OLS regression on the first difference of each variable; b Generalized least Squares weighted by between and within estimates;c Generalized least squares weighted by OLS residuals. While it is clear from model 2 that other government spending has a negative effect on growth, it might not be so clear what impact the other three components of government spending has on growth, given the inclusion of linear and non-linear terms in the model. To get some sense of the quality (net effect on growth) of government spending on defence, education and health, we can evaluate the estimates in model 2 at their sample means. To do this, we take the partial derivative 16 of per-capita growth, y, with respect to government spending on defence (Defexp), education (Eduexp), and health (healthexp). These partial derivatives are as follows: Evaluating defence spending: Evaluating education spending: Evaluating health spending: y/ (Defexp) = y/ (Eduexp) = Defexp Eduexp +2 +2 Defexpsq Defexp Eduexpsq Eduexp (1) (2) (3) y/ (Healthexp) = Healthexp + 2 Healthexpsq Healthexp Evaluating each component of government spending at the sample mean produces the results in Table 9. These results suggest that government spending, in all forms, have a negative influence on growth. As one would expect, spending on education seems to have the smallest negative impact on growth while spending on defence seems to generate the largest opportunity cost. These findings support those who argue that because the public sector does not respond to market signals, its spending can be inefficient and tends to crowd out more efficient private sector activity generating a negative effect on the growth rate. Although as we show above that government spending contributes to social development, such as longer life expectancy and lower fertility rates, because public spending tends to be inefficient, larger governments generate higher opportunity costs to the economy in terms of private sector activity forgone. This seems to have implications for economic efficiency and the rates of growth achieved by the economy. Table 9: Net impact on growth as evaluated at sample mean for components of government spending DEFEXP EDUEXP HEALTHEXP OTHGOVEXP Sample Mean 8.74 15.28 7.87 68.11 Effect on Growth -1.13 -0.41 -0.62 -0.68 CONCLUSION AND POLICY IMPLICATIONS This paper examines the impact of government spending on economic growth in eighteen Latin America and Caribbean countries over the period 1970 to 1997. One of the key questions that this study attempts to address is whether countries with higher government consumption and spending on defence, education and health tend to have higher rates of social development and economic growth or whether there are diminishing returns in terms of growth and other welfare gains to higher government spending. We approach the issue in several ways. First, we examine the relationship between government consumption across income groups and regions across the globe; secondly, we examine the relationships between government spending and several indicators of social development; and finally, we examine the impact of overall government spending on economic growth and then segregate government spending into four parts to see whether certain components of government spending are growth enhancing while others are not. We find that while government spending is associated with improvements in certain social indicators, government spending in all forms has a negative influence on the rates of economic 17 growth. There are several ways of looking at these results. The mainstream might argue that they indicate a need for scaling down the size of government or the extent to which the state taxes and spends. Consistent with this view, one might argue that governments should ensure that current public spending does not burden future generations. Consequently, current outlays should, as far as possible, be paid for currently. To the extent that government spending retards growth, one can only make weak arguments for capital expenditure to be financed by debt, which becomes a tax on future generations. What all this suggests is that increasing the size of government may not necessarily lead to greater growth in Latin America and the Caribbean. On the other hand, we have also shown above that government spending on education and health is associated with longer life expectancy and lower fertility rates when females are more educated. The evidence suggests that there is a role for government in society and this implies a certain quantum of government spending. For instance, governments have a responsibility to promote social equity, which they can do through their spending on social programmes in education, health and in reducing unemployment levels. Human resource development through investments in education can play a meaningful role, not only in raising the productivity of each worker, but also in providing greater skills, which can lead to a higher income. Indeed, social expenditure is a major means by which governments can influence income distribution. On the basis of this, we conclude that there is a social developmental role for government, but also that governments should restrict their spending to the core purposes of facilitating income redistribution, raising the productivity of labour, addressing issues of market failure and other externalities, and providing for law and order in the country. 18 REFERENCES Agénor, P. and Montiel, P.J. (1996), Development Economics, Princeton, New Jersey: Princeton University Press. Aghion, P. and Howitt, P. (1992), ―A model of growth through creative destruction‖, Econometrica, Vol. 60, No. 2, 323-351. Barro, R.J. (1991), ―Economic growth in a cross-section of countries‖, Quarterly Journal of Economics, Vol. 104, 407-44. Barro, R.J. (1996), ―Determinants of Economic Growth: A cross-country empirical study‖, NBER Working Paper 5698. Barro, R.J. and Lee, J. (1994), ―Sources of economic growth‖, Carnegie-Rochester Conference Series on Public Policy (June), 1-46. Economic Commission for Latin America and the Caribbean, (1998), The Fiscal Covenant: Strengths, Weaknesses and Challenges. Santiago, Chile. Grossman, G.M. and Helpman, E. (1991), Innovation and Growth in the Global Economy, Cambridge MA, MIT Press. Hall, R.E and Jones, C.I. (1997), ―Levels of economic activity across countries‖, The American Economic Review, Vol. 87, No. 2, 173-177. Koopmans, T.C. (1965), ―On the concept of optimal economic growth‖, in The Econometric Approach to Developmental Planning, Amsterdam, North Holland. Kormendi, R. and Meguire, P. (1985), ―Macroeconomic determinants of growth: Cross-country evidence‖, Journal of Monetary Economics, Vol. 16, 141-63. Landau, D. (1983), ―Government expenditure and economic growth: A cross-country study‖, Southern Economic Journal, Vol. 49, January, 783-92. Landau, D. (1986), ―Government expenditure and economic growth in less developed countries: An empirical study for 1960-80‖, Economic Development and Cultural Change, Vol. 35, October, 35-75. Levine, R. and Renelt, D. (1992), ―A sensitivity analysis of cross-country growth regressions‖, The American Economic Review, Vol. 82, No. 4, pp. 942-63. Lucas (1988), ―On the mechanics of economic development‖, Journal of Monetary Economics, Vol. 22, No. 1, 3-42. Mulligan, C.B. and Sala-i-Martin, X. (1993), ―Transitional dynamics in Two-sector models of endogenous growth‖, Quarterly Journal of Economics, Vol. 108, No. 3, 737-773. 19 Musgrave, R. (1997), ―Reconsidering the fiscal role of government‖, The American Economic Review, Vol. 87, No. 2, 156-159. Ram, R. (1986), ―Government size and economic growth: A new framework and some evidence from cross-section and time series data‖, The American Economic Review, Vol. 76, No. 1, 191203. Rebelo, S. (1991), ―Long-run policy analysis and long run growth‖, Journal of Political Economy, Vol. 99, No. 3, 500-521. Sala-i-Martin X.X. (1997), ―I just ran two million regressions‖, The American Economic Review, Vol. 87, No. 2, 178-183. Stroup, M.D. and Heckelman, J.C. (2001), ―Size of the military sector and economic growth: A panel data analysis of Africa and Latin America‖, Journal of Applied Economics, Vol. 4, No. 2, 329-60. Romer, P.M. (1990), ―Endogenous technological change‖, Journal of Political Economy, Vol. 98, No. 5, part II, S71-S102. Solow, R. (1956), ―A contribution to the theory of economic growth‖, Quarterly Journal of Economics, Vol. 70, No. 1, 65-94. Summers, R. and Heston, A. (1988), ―A new set of international comparisons of real product and price level estimates for 130 countries, 1950-85‖, Review of the Income and Wealth, Vol. 34, 125. Swan, T.W. (1956), ―Economic growth and capital accumulation‖, Economic Record, Vo. 32, November, 334-361. Tanzi, V. and Schuknecht, L. (1997), ―Reconsidering the fiscal role of government: The international perspective‖, The American Economic Review, Vol. 87, No. 2, 164-168. 20