Change Font: A A A A Contact Us What's New FAQs Subscribe ADB.org home
HomePublicationsGrowth and Adjustment in East Asia and Latin AmericaDeterminants of Economic Growth: A Cross-Country Analysis

Determinants of Economic Growth: A Cross-Country Analysis

In this section, we explore the main factors that influenced the growth of per capita income over the past three decades. The analysis is based on a general framework of cross-country regressions, and places the experience of an individual country in a global context. This approach allows us to understand the specific factors associated with economic growth across countries and the key differences between fast and slow growing economies. Based on this framework, we explore the factors that explain why Latin American countries grew at much lower rates than the best performing economies in East Asia. This exercise provides a basis for understanding the future growth prospects of the East Asian and Latin American countries.

3.1 The Basic Empirical Framework

The basic empirical framework is based on an extended version of the neoclassical growth model, as described by Barro (1991), Mankiw, Romer, and Weil (1992), Barro and Lee (1994), and Barro and Sala-i-Martin (2004). This model predicts a "conditional convergence" of income, implying that a country with a lower initial income relative to its own long-run (or steady-state) potential level of income will grow faster than a higher-income country over time. The basic idea is that the further an economy is located from its steady-state level, the greater will be the gap of reproducible (physical and human) capital stock per worker and technical efficiency from their long-run potential levels. The gap of existing capital and technology from their steady-state levels provides the low-income economy with the chance to catch up rapidly with the highincome country, through high rates of capital accumulation as well as the diffusion of technology from the more technically advanced economies. In a cross-country context, convergence implies that poorer countries grow faster than richer countries when controlling for the variables that influence the steady-state level of per capita income. As a reduced form the model can be represented by

where the dependent variable is the growth rate of per capita income for the period T for country i, log(y0i) is a log value of the initial level of per capita income for country i, and where Zi denotes an array of the variables that influence the country i's steady-state level of per capita income. The conditional convergence implies a negative coefficient for initial income. Note that the variables included in Z can affect either the rate of productivity growth or the rate of capital accumulation. A wide variety of external environment and policy variables affect growth rates by influencing the long-run potential income and the rate of productivity growth. The extended Solow-type neoclassical growth model emphasizes investment rate, population growth, and human capital as important factors that determine the steady-state level of income (see, for example, Mankiw, Romer, and Weil [1992]). Previous empirical research also takes institutions and policy factors to be important determinants of longrun per capita income (Mauro, 1995; Knack and Keefer, 1996; and Barro, 1997). These factors include government consumption, rule of law, inflation, democracy, and trade openness.

The external environment and policy variables that we consider are as follows4:

Investment: In neoclassical growth models, a higher value of the saving rate, both domestic and foreign, raises the steady-state level of output per capita (equation [1]) and thereby increases the growth rate for a given starting value of GDP.

Fertility: The fertility rate has an important influence on population growth, which has a negative effect on the steady-state ratio of capital to workers in the neoclassical growth model. Hence, the model predicts a negative effect of fertility on economic growth. Higher fertility also reflects a greater devotion of resources to childrearing, and for this reason it is better to use fertility rates than population growth.

Human Resources: The various models of new growth theories emphasize human capital as a key factor driving the long-term growth of income. In the framework of the extended neoclassical growth model, for given values of other explanatory variables, a higher human capital stock leads to a higher steady-state per capita income. In the endogenous growth model, human capital generates perpetual growth by either preventing returns to broad capital from falling or by increasing capabilities for the innovation and adaptation of new technologies. The human resource variables include a measure of human capital stock. We use the average years of secondary and higher schooling for males aged 25 and over, available from Barro and Lee (2001). The greater initial educational stock indicates that a more skilled workforce can produce more output from given natural and physical resources. Hence, a country with a greater education stock is in a better condition for future growth. In addition, life expectancy at birth, as a log value at the initial year of the period, is used to measure health attainment, which is seen as another important component of human capital stock. A higher life expectancy tends to indicate healthier, more productive workers.

Institutions and Policy Variables: We consider five institutions and policy variables. The first is government consumption (defined as the average ratio of government consumption in final goods to GDP). The measure of government consumption used here excludes public expenditures on education and defense, because these two categories of government expenditures can be regarded primarily as investment (Barro, 1991 and Barro and Lee, 1994). Higher government consumption leads to lower growth by shifting resources from productive activities and distorting private decisions.

The second institution and policy variable is a measure of the overall maintenance of the rule of law in the economy. An environment that secures property rights and provides a strong legal system is critical for investment and other aspects of economic activities. The best available indicators to measure quality of institutions come from international consulting firms that provide advice to international investors based on information collected by local experts. Knack and Keefer (1996) introduce measures of institutional quality initially devised by Political Risk Services. The measures consist of five indicators including: (a) quality of bureaucracy, (b) corruption in government, (c) rule of law, (d) expropriation risk, and (e) risk of repudiation of contracts by government. Among the various indicators, the measure of the rule of law is considered to have the most explanatory power for economic growth (Barro, 1997). We use this measure of law enforcement, rescaled to a zero-to-one scale, as being the most effective.

The third policy measure is the inflation rate. De Gregorio (1992, 1996), Fischer (1993), and Barro (1997) find that inflation has a significant negative effect on growth. Hence, worsening price stability, caused by macroeconomic mismanagement, seems to lead to a lower steady-state level of per capita output for given values of other explanatory variables.

We also include a measure of "democracy" as another institution variable. This measure is constructed by Barro (1997) based on the measure originally constructed by Gastil. It measures the strength of electoral rights and civil liberties, scaled from zero to one, where one corresponds to the highest level of democracy. The relationship between democracy and economic growth is not clear. For example, a more democratic political regime can be characterized by a redistribution of income from rich to poor. This redistribution may reduce the incentives of people to work and invest, and thus work against economic growth. However, reducing income inequality and having an open political system can reduce the tendency for social unrest and thus contribute positively to overall economic activity.

The last policy variable is a measure of openness. Open economies have greater access to cheap imported intermediate goods, large markets, and advanced technologies. Lee (1993) and Frankel and Romer (1999) find evidence that more open economies tend to grow faster. We measure the extent of each economy's openness by the ratio of exports plus imports to GDP. Openness is well-known to vary by country size—larger countries tend to be less open because a larger internal market can help reduce reliance on international trade. The openness measure used in this analysis filters out the normal relationship (estimated in another regression system) of international openness to the logs of population and area. This filtered variable thus reflects the influences of government policies, such as tariffs and trade restrictions, on international trade (see Barro and Sala-i-Martin, 2004, Ch. 12).

Terms of Trade Shock: The terms of trade shock is considered to be an exogenous factor that affects the growth rate of an economy. Improvement in the terms of trade, measured as the ratio of export to import prices, can make a country produce more and expand its export sector.

Balance-of-Payments Crisis: External imbalances normally affect cyclical fluctuations rather than long-run growth. However, when a significant balance-ofpayment difficulty causes a crisis, it can disrupt the entire economy because the uncertainty it generates discourages investment and other productive activities, while increasing speculative activities. Financial distress may lead to bankruptcies of profitable firms that would otherwise have been viable. Barro (2001) shows that currency crises have a negative influence on economic growth.

We define a balance-of-payments crisis dummy variable for each country during any five-year period as equaling one if a crisis occurred during the period and otherwise to take on the value zero. The definition of balance-of-payments crisis is discussed in section 4.1.

Table 2 [ PDF: 171kb | 1 page ] shows basic statistics for all 85 countries in the sample, for the beginning and latest sub-periods, 1970-75 and 1995-2000. It also compares the statistics between Latin America and East Asia, confirming that over the past three decades, the group of nine East Asian economies was better placed for rapid growth than the Latin American group in terms of most structure and policy environments. A notable exception is that in 1995-2000, a larger fraction of the East Asian region was subject to balance-of-payments crises than the Latin American region. Also, the faster increase in income due to the higher growth performance in the East Asian economies led them to face a less favorable convergence effect in the 1995-2000 period than in their earlier period, and also with respect to the Latin American countries as a whole.

Our regression of specification (1) applies to a panel set of cross-country data over six five-year periods from 1970 to 2000, corresponding to the periods 1970-75, 1975-80, 1980-85, 1985-90, 1990-95, and 1995-2000.5 The dependent variables are the annual growth rates of real GDP per capita over the same six five-year periods. Some previous studies use cross-section data in which each country has only one observation. The approach based on the panel data set seems to consider more information than what is available from time series variations within each country.

One concern in the empirical specification is that any effect from contemporaneous explanatory variables may reflect a reverse causation from GDP growth to the explanatory variable. For example, the relationship between contemporaneous investment and growth may be because high growth causes high saving. This problem, however, can be solved by adopting an instrumental-variables estimation technique. We estimate this system of the six equations using three-stage least squares.6 The instrumental-variable technique controls for the possible simultaneity problem when Zi—the control variables—are endogenously determined. Instruments are mostly lagged values of the independent variables (see the notes to Table3). We use prior colonial status (Spanish or Portuguese colonies and other colonies) as instruments for inflation rate in the instrumental-variable technique as in Barro (1997). In order to control for the possible reverse causation from lower growth to higher frequencies of balance-of-payments crisis, we use the ratio of international reserve to monthly imports at the beginning of each five-year period as an instrument for balance-of-payments crisis.

3.2 Regression Results

Table 3 [ PDF: 132kb | 1 page ] shows the regression results using the basic framework of equation (1) and the explanatory variables just described. The three-stage least squares technique is applied to a data set for 85 countries.

Column 1 of table shows the result of the basic regression without including the balance-of-payments crisis dummy variable. Column 2 includes as an independent variable the balance-of-payments crisis dummy. Although the two columns show a similar pattern of results, substantial differences arise for inflation and schooling variables. The estimated effect of inflation on growth becomes much smaller when we include the balance-of-payments crisis variable. This may reflect the strong positive correlation between inflation and balance-of-payments crisis. On the contrary, the schooling variable becomes more significant in column 2 where the balance-ofpayments crisis variable is added. Since the balance-of-payments crisis variable itself enters very significantly, we focus on the results of column 2.

The results provide strong evidence for conditional convergence: the coefficient on the log value of initial GDP in column 2 is highly significant, and the estimated coefficient is -0.025 (standard error = 0.004). Thus, a poor country with a lower initial income level grows faster, controlling for the variables that influence the steady-state level of income. Specifically, the coefficient implies that a country with half the income level of another will grow 1.73 percentage points (=2.5%*ln(2) ) faster. The investment rate and fertility variables have strong effects on the growth rate. The estimated coefficient on investment rate is positive and statistically significant at the 5% level. The coefficient 0.056 (s.e. = 0.027) implies that a one standard deviation increase, equivalent to 8.3 percentage point in the ratio of investment to GDP in the 1995-99 period, is associated with an increase in the growth rate of about 0.5 percentage points per year. The estimated coefficient on the logarithm of fertility rate is strongly negative, -0.015 (s.e. = 0.006), implying that an increase of 0.51 (the variable's standard deviation) in the fertility rate in 1995 lowers the growth rate by about 0.8 percentage points per year.

The result of column 2 shows that human resource variables have a significantly positive effect on economic growth. The educational attainment variable, which is measured by an average year of secondary and tertiary schooling of the male adult population, has a positive effect on the growth rate: the estimated coefficient, 0.0029 (s.e. = 0.0017), is statistically significant at the 10% critical level. The mean and standard deviations of the schooling variable are 2.4 and 1.5 per year respectively for 1995. Therefore, the coefficient indicates that one standard-deviation increase in the secondary and higher schooling raises the growth rate of per capita income by about 0.4 percentage points per year. The logarithm of life expectancy at age one—a measure of health attainment—is highly significant in the regression: the estimated coefficient 0.065 (s.e. = 0.021) implies that an increase in life expectancy by 0.13 (the standard deviation of the log of life expectancy) in 1990 is estimated to raise the growth rate by about 0.9 percent per year.

We find clear evidence that institution and policy variables play a significant role in determining economic growth. The government consumption variable has a significantly negative impact on growth: an increase in government consumption ratio by one percentage point reduces growth by 0.07 percentage points a year. In the sample, a one standard deviation of 5.5 percentage points over the 1995-99 period decreases the growth rate of per capita income by about 0.4 percentage points per year.

The rule of law index has a strong positive effect on growth, indicating that countries with more effective protection of property and contractual rights tend to have higher growth rates. The estimated coefficient, 0.018 (s.e. = 0.008) implies that a one standard deviation increase of 0.22 in this index (on a scale of 1.0) in the 1995-99 period is associated with an increase in the growth rate of about 0.4 percentage points.

The openness variable appears to be positively associated with the growth rate. The estimated coefficient 0.0086 (s.e. = 0.0046) is significant at the 10% level. An increase of an economy's openness by 0.4 (its standard deviation) over the 1995-99 period is estimated to raise the growth rate by about 0.4 percentage points per year. The regression result confirms the non-linear relationship between democracy and growth, as found by Barro (1997). The coefficients of the indicator of democracy and its square terms are positive and negative respectively, and both are statistically significant. The pattern of coefficient values indicates that growth increases with political freedom when the level of democracy is low, but decreases with it once the society has attained a certain level of political freedom. The estimated coefficients in column 2 imply that the switch occurs at a level of democracy of 0.635. Both the Latin America and East Asian regions on average were below this critical value in the 1970s. However, in the 1995-99 period, Latin America's average level of democracy, 0.732, slightly exceeded this critical level.

Column 2 shows that the effect of inflation on economic growth is negative but statistically insignificant. The estimated coefficient, -0.013 (s.e. = 0.009), implies that an increase in the average rate of inflation by one standard deviation of 9.7 percent over the 1995-99 period lowers the growth rate by 0.1 % per year. Note that the coefficient is less than half the value of column 1, where inflation has a greater impact on growth. As we saw earlier, the problem is the correlation between balance-of-payment crises and inflation.

The regression result shows a less significant effect of the terms of trade change on per capita GDP growth. The estimated coefficient on the growth rate of the terms of trade is 0.035 (s.e.= 0.023), indicating that countries with favorable terms of trade shocks by one standard deviation of 0.039 in the 1995-99 period grew by about 0.1 percentage point per year above other countries.

The balance-of-payments crisis turns out to have a strong, negative effect on economic growth. The estimated coefficient of the balance-of-payments crisis variable is -0.017 (s.e.= 0.005), indicating that a balance-of-payments crisis shock lowers the growth rate by 1.7 percentage points per year.

Column 3 of Table 3 [ PDF: 132kb | 1 page ] adds the lagged effect of a balance-of-payments crisis. The result confirms those of Barro (2001). The retardation of growth by a balance-ofpayments crisis does not persist into the subsequent five-year period. In fact, the effect of a balance-of-payments crisis on economic growth in the following five-year period turns out to be positive but statistically insignificant. Therefore, a balance-of-payments crisis reduces income permanently, but has no permanent effects on growth.

Table 3 [ PDF: 132kb | 1 page ] also shows the result of the regression with the inclusion of regional dummies. Column 4 shows that the Latin American dummy variable has a statistically insignificant coefficient, while the East Asian dummy is marginally significant at the 10 percent level. It is interesting to note that earlier empirical studies found a significant and negative "Latin American dummy" (Barro, 1991), which in the current empirical framework becomes insignificant, indicating that the explanatory variables included in the right-hand side explain most of the poor performance of Latin American economies. However, the point estimates, although small in magnitude and statistically insignificant, still indicate that in addition to the variables included, Latin America has lower growth rates than average, and East Asia has higher growth rates than average. Even controlling for the two regional dummies, the regression shows that most of the explanatory variables are still significant and have estimated coefficients of the same magnitude, compared to those in column 2 of Table 3 [ PDF: 132kb | 1 page ].

3.3 Economic Growth of Latin America in Comparative Perspective

The results of the cross-country regression allow us to analyze the growth performance of the Latin American countries relative to that in other regions. We compare the growth performance of Latin America to the best performance of East Asia. Average per capita growth rates for the nine economies in the East Asia region were 5.6%, 5.1% and 4.3% over each decade in the 1970-2000 period, while those for the 21 Latin American countries were 2.1%, -0.8% and 1.6% respectively.

We use the point estimates of the parameters in the regression (2) of Table 3 [ PDF: 132kb | 1 page ] to carry out a simple "accounting" that breaks down the fitted values of growth rates for each country into the contributions from each of the explanatory variables. Although the residual errors in individual country growth rates are substantial, it is worthwhile to examine the differences in the explanatory variables that generate differences in the fitted growth rates. We then explore the sources of the differences in the fitted growth rates between East Asia and Latin America.

Table 4 [ PDF: 215kb | 1 page ] presents the results. The basic regression can account for a substantial part of the growth differences between the two regions. For the 21 Latin American countries, the predicted growth rate is 3.1 percentage points lower on average than that for East Asia over the whole period from 1970 to 2000. The actual difference was 3.6 percentage points, and therefore we can explain the bulk of the differences. It is interesting to note, however, that the larger difference occurs during the "lost decade" of the eighties. The predicted difference can be broken down separately into the contributions of the 12 explanatory variables.

The relatively higher income level of Latin America compared to that of Asia in 1970 led to lower growth there in the 1970-90 period because of the convergence effect. However, this convergence effect became rather favorable to Latin America after 1980, when the income of East Asia surpassed it. Hence, the net convergence effect becomes negligible over the three decades from 1970 to 2000.

In 1970, Latin America had a slightly higher life expectancy and thus better conditions for growth than East Asia. But in general, Latin America had relatively poorer human resources—in terms of lower educational attainment and lower life expectancy. The regional differences have widened over time. The net effect of human resources contributed to about slower growth by about 0.3 percentage points in Latin America relative to Asia over the whole period.

The investment rate and fertility had strong effects on Latin America's performance relative to Asia, by lowering the per capita growth rate by about 0.6 and 0.5 percentage points per year respectively over the past three decades. Without this difference, Latin America would have had a level of per-capita income 25% higher after the thirty years ending in 2000.

The institution and policy variables turned out to have a significant effect on differences in growth rates. The differences in growth may be due to low (human and physical) capital accumulation, or low productivity growth. The growth effects of institutions and policies can occur by reducing productivity and the speed of catch-up to the technological frontier, and changing the incentives for (physical and human) capital accumulation.

The combined effect of the differences in the five policy variables— government consumption, rule of law, inflation, democracy, and trade openness— accounted for 1.6 percentage points slower growth in Latin America relative to Asia over the period from 1970 to 2000. The institution and policy variables contributed most to the difference in growth rates, registering 2.0 percentage points in the 1980-90 period. That is, during the debt crisis, policies and institutions deteriorated significantly in Latin America. As we emphasize again below, although external conditions could have led to a deterioration of internal policies and institutions, the poor growth performance, even in a period with negative external environment such as the eighties, can be traced to bad policies and institutions.

Among the institution and policy variables, trade openness was the most important variable. Latin America's relatively inward-oriented trade strategy accounted for slower growth of about 0.6 percentage points per year. Not only does it have lower trade share (GDP share of exports plus imports), but most of the countries are smaller in size and population than Asian countries, which further reduces effective trade openness.

The higher inflation in Latin America also reduced growth by 0.3 percentage points relative to Asia over the whole period from 1970 to 2000. The negative effect of high inflation was more significant in the 1980s, lowering growth by 0.7 percentage points in Latin America relative to Asia. During the 80s, the average inflation rate in Latin America was 48.5%, while for our sample of 9 East Asian countries it was 2.6%. As discussed above, this effect does not include the likely detrimental effects of inflation on investment; however the evidence shows that the effects of inflation on investment are much smaller than those of inflation on productivity growth (De Gregorio, 1996).

Government consumption and rule of law also contributed to the lower growth rate of Latin America, by 0.3 and 0.4 percentage points per year respectively over the three decades. By contrast, democracy turned out to play an insignificant role on growth difference between the two regions. On average East Asia's low political freedom, far lower than the "critical turning point," has been relatively unfavorable to economic growth. Because democracy has a nonlinear effect on per capita growth, very low or high values tend to be detrimental to growth. In this respect, the democracy level for PRC, Indonesia and Haiti were very low, while Costa Rica, and Trinidad and Tobago were on the high side of the distribution. But for most East Asian and Latin American countries, the democracy level has not made a significant difference to growth rates.

Table 4 [ PDF: 215kb | 1 page ] shows that the effect of the relatively unfavorable terms-of-trade shock was also small in Latin America. This result casts doubt on the view that the problem of Latin America was due to its patterns of specialization, which faced a particularly unfavorable external scenario. Advocates of the import substitution strategy of Latin America in the sixties argued that the countries ought to pursue internal industrialization since the products they exported faced declining terms of trade. However, the evidence from our regression shows that the latter argument is wrong and that it is precisely openness, as part of good policies and institutions, that boosts fast and lasting growth.

In addition, the external environment could explain part of the poor performance during the debt crisis. The largest difference between predicted and actual growth occurs in the 80s. This difference of 1.4 percentage points, even after we control for policies, institutions, terms of trade and balance of payments crisis, cannot be explained by the growth regressions. Of course, as we discuss in the next section, the output losses from currency crises do not only depend on external factors, but also on some internal factors such as initial conditions and policy responses.

On the other hand, the balance-of-payments crisis contributed to lower growth in Latin America by about 0.2 percentage points than in East Asia over the whole period. It had the biggest effect in the 1980-90 period, explaining the growth differential of 0.5 percentage points. But in the 1990s, when East Asian economies also suffered from balance-of-payments crises, its contribution to the growth differential became negligible.

Thus, while initial income and external conditions explain only moderate differences in growth rates, the major differences are produced by investment, human resources and the institution and policy variables. The traditionally important growth factors such as investment, fertility, and quality of human resources contributed significantly to the difference in per capita GDP growth between East Asia and Latin America. Moreover, relatively poor economic policies, such as trade protection, high inflation, high government consumption, and lack of good institutions, have been very important factors contributing to the relatively slow growth of Latin American countries during the past three decades.

Table 4 [ PDF: 215kb | 1 page ] focuses on the relatively poor performance of Latin American countries compared to East Asia. But there were also tremendous variations in growth performance among Latin American countries. While the best performing—Dominican Republic and Chile—grew by 3.2 and 2.4 percent per year during the 1970 to 2000 period, the worst performers—Nicaragua and Venezuela—registered negative growth rates of -2.7 and -1.7 percent. In addition, growth rates fluctuated greatly within each country. For instance, average per capita growth for Chile was only around 1.2 percent over the period 1970-90 but increased dramatically to 4.8 percent over the period 1990- 2000. On the contrary, with the exception of Philippines, the 9 Asian countries all had strong growth throughout most of those three decades without significant variations.

For this reason, we investigate the extent to which variations in the growth performance of individual Latin American countries can be attributed to the factors that explain the international growth variations, and in particular, the extent to which the variations are due to differences in domestic institutions and policies. We therefore extend Table 2 [ PDF: 171kb | 1 page ] of the entire period, breaking down Latin America into 21 individual countries.

Based on the regression result of column 2 in Table 3 [ PDF: 132kb | 1 page ] , we can assess how much of the variations in the growth performance of individual Latin American countries relative to that in East Asia can be attributed to each explanatory variable. Table 5 [ PDF: 182kb | 1 page ] shows the effect of various factors on the difference between predicted and actual growth to 10 selected Latin American economies over the whole period from 1970 to 2000. For instance, the predicted growth rate for Chile is 2.6 percentage points lower on average than the East Asian region, while the actual difference was 2.1. The relatively higher income level of Chile compared to East Asia led to a drop of 0.5 percentage points in Chile's growth due to the convergence effect. Investment represented a 0.5 percentage points increase, while fertility and human resources made no significant contribution to the growth differential. Chile's sound rule of law accounts for the 0.2 percentage point edge over East Asia but other institutions and economic policies were relatively unfavorable to growth. The combined effect of the differences in the other four policy and institutional variables—high government consumption, high inflation, low levels of democracy, and low trade openness—accounted for 1.5-percentage-point slower growth in Chile relative to the group of nine East Asian countries from 1970 to 2000. Most of these indicators improved during the nineties, contributing to growth above the East Asian average, and only surpassed by PRC and Taipei,China over the whole decade.

The negative effect of poor institutions and policies on growth is evident in all Latin American countries. In some countries, such as Bolivia, Haiti, and Nicaragua, poor institutions and economic policies completely outweigh the favorable factor of lower initial income, leading to far slower average per capita growth compared to the East Asia region. For example, Haiti should have grown by 3.7 percentage points more than East Asia given its relatively lower initial income level. But it turned out that Haiti's average growth rate over three decades was about 2.8 percentage points lower than the East Asian average because of poor human resources and economic institutions.

Some Latin American countries that were more prone to crisis over the past three decades (Argentina, Brazil, Ecuador and Mexico) show significantly lower growth rates, about 0.7 to 0.9 percentage points below East Asia.

3.4 Growth Prospects for East Asia and Latin America

The results from cross-country regressions can be used to construct forecasts of economic growth for individual countries. The projected growth rates for 2001-2010 are obtained by multiplying the 2000 values (or the 1995-99 period average) of the explanatory variables by the estimated coefficients in the panel regression of column 2 in Table 3 [ PDF: 132kb | 1 page ]. Terms-of-trade shocks are assumed to be equal to those in the 1990s. We assume no balance-of-payments crisis.

The results of the growth projections for East Asia and Latin America are presented in Table 6 [ PDF: 147kb | 1 page ]. We only provide the regional averages.

For the 21 Latin American countries, the predicted growth rate is estimated to be 2.3 percent over the 2001-2010 period, increasing from the average of 1.6 percent during the nineties. On the contrary, the average growth rate for the East Asian region is predicted to be 3.8 percent—a figure very close to the average of 4.0 percent in the last decade. Hence, the growth differential between two regions will shrink substantially to 1.4 percentage points, compared to 3.1 percentage points over the whole period from 1970 to 2000 and 1.9 percentage points over the 1990-2000 period. This is explained basically by convergence, since the high initial income in East Asia slows growth vis-àvis Latin America.

Overall, growth in Latin America compared to East Asia is predicted to be higher than in any single decade of the previous 40 years, with the exception of the sixties. Although modest when compared to East Asian performance, this rate of percapita income growth is almost twice that of 1960-2000. Improved institutions and policies help to explain why Latin America could do better.

The predicted difference in average growth rates over the period of 2001-2010 can be broken down separately into the contributions from the 12 explanatory variables.

Table 6 [ PDF: 147kb | 1 page ] shows that the convergence effect becomes quite unfavorable to East Asia due to its higher income relative to that of Latin America in 2000. The net convergence effect during the 2001-2010 period is predicted to make the average growth rate of the East Asian region 1.7 percentage points per year lower than that of the Latin American region. Therefore, considering the effect that convergence has in reducing differences across regions, the rest of the factors still explain a large difference of about 3 percentage points.

The increasing gap between Latin America and East Asia in terms of human resources is likely to contribute to slower growth in Latin America, with a net effect of some 0.6 percentage points over the 2001-2010 period. Although both regions have experienced improvements in human resources (see Table 2 [ PDF: 171kb | 1 page ] ), the differences have widened and the variables explain a larger difference than in the past. The difference in investment still explains an about 0.7-percentage-point growth differentials.

The institution and policy variables are expected to continue to have strong effects on differences in growth rates. The combined effect of the differences in the five policy variables—government consumption, rule of law, inflation, democracy, and trade openness—is expected to account for growth 1.3 percentage points lower in Latin America relative to East Asia over the period from 2001 to 2010.

Note that we assume that no crisis will occur to any region. But in reality, crises often occur. A crisis could make a big difference in our predictions. The estimation shows that a balance-of-payments crisis would lower the growth rate by 1.7 percentage points per year. This is equivalent to the predicted differential in growth rates between the two regions over the 1990-2000 period.

3.5 Extensions: Quality of Education and Income Distribution

Empirical studies of the determinants of economic growth suggest numerous additional explanatory variables. Our framework captures the most important growth determinants, but some other variables may also have a bearing on performance, particularly in Latin America and East Asia. These variables could be relevant growth determinants although the regressions may not capture them well because of lack of data or collinearity problems with other independent variables, which may hamper the search for sensible estimates.

One important additional variable is the quality of schooling.7 The schooling variable considered in basic regressions refers to the quantity of education, as measured by years of schooling, rather than its quality. An alternative measure of educational stock, which is considered to reflect variations of educational quality between countries, is the scores achieved in internationally comparable tests in the subjects of science and mathematics. Conceptually, the quality of education is reflected in the performance of students and graduates. One shortcoming of these data, however, is that the observations apply to different years and are most abundant for the 1990s. Based on the limited sample, Barro (1999) and Hanushek and Kimko (2000) find that test scores are positively related to growth rates of real per capita GDP in cross-country regressions.

Table 7 [ PDF: 199kb | 1 page ] shows the average test scores on mathematics and science for seventh grade students in the countries that participated in the cross-national achievement tests. In 1991 the International Assessment of Educational Progress (IAEP) conducted tests of mathematics and science achievements for 13-years-old students. The International Association for the Evaluation of Educational Achievement (IEA) also carried out the Third International Mathematics and Science Study (TIMSS) in 1994 and 1995.

Among the 44 countries that participated in the IAEP and/or TIMMS projects, students in the East Asian economies—PRC, Hong Kong, China, Japan, Korea, Singapore and Taipei,China—showed the highest achievements in mathematics. For example, in the IAEP mathematics test, PRC ranked first with an average score of 80.2, followed by Korea and Singapore. In contrast, Brazil, the only participating Latin American country, came last, with an average score of 34.7, following Mozambique. Among the 39 countries that participated in the TIMSS, Singapore, Korea, Japan, and Hong Kong, China were the top four in mathematics, with average scores ranging from 56.4 for Hong Kong, China to 60.1 for Singapore. In contrast, Colombia, the only participating Latin American country, performed significantly more poorly with a mean of 36.9, placing it second to last after South Africa.

The results are also favorable for Asia in science tests, for which Asian students performed much better than Latin Americans. Although evidence on the quality of schooling is still scarce, there is a very clear gap between Latin America and Asia, which adds to the deficiencies in the quality of human resources we already discussed in the previous sections.

Another area where the differences between Latin America and Asia are evident, although not included in the independent variables in our regressions, is income distribution. Figure 1 shows Gini coefficients for Latin American and Asian countries, with Japan and the United States shown for comparison. The data are taken from the World Development Report of the World Bank, for the closest year, which in most cases is between 1996 and 1998. There are several problems that make it difficult to make cross-country comparisons. For example, differences depend on whether the unit of analysis is the household or individual, whether income is measured before or after tax, and whether the surveys refer to income or expenditure. However, given all of those caveats, the conclusion is undisputable: inequality in Latin America is much greater than in Asia, and as we argue below it can explain differences in human resources, policies and institutions between the two regions.

Source: World Bank, World Development Report.

The relationship between income distribution and growth has recently become a hot topic. Theoretical discussions often predict negative effects of inequality on growth (Alesina and Rodrik 1994, and Persson and Tabellini 1994). Most cross-country empirical studies also find support for a negative relationship between income inequality and growth (Alesina and Rodrik 1994, and Perroti 1996). However, some recent studies based on panel-data estimation find a positive relationship (Li and Zou 1998, and Forbes 2000). The main problem in carrying out cross-country empirical investigations is the quality and comparability of the data measured, with small differences often resulting in large differences in the estimated relationship between inequality and growth.

We have investigated the effects of inequality on growth using our panel framework. Our measure of income inequality is the Gini index. The data come from the UNU/WIDER—UNDP World Income Inequality Database (WIID), which extends the data set of Deininger and Squire (1996).

The first row of Table 8 [ PDF: 93kb | 1 page ] reports the estimated coefficient on the Gini index when it is added to the systems in row 2 of Table 3 [ PDF: 132kb | 1 page ]. The overall sample size for the panel regressions decreases from 464 to 277, because many fewer observations of Gini coefficients are available than for the full sample considered in Table 3 [ PDF: 132kb | 1 page ]. In the system, the Gini value for around 1970 appears in the equation for growth from 1970 to 1975, and so on. The five-year lagged values of the Gini coefficients are added to the list of instruments.

The estimation result shows that there is no significant impact of Gini coefficients on economic growth. The estimated coefficient, -0.001 (s.e.=0.018) is essentially zero. Thus, with the other explanatory variables considered in growth regressions held constant, differences in income distribution have no significant effect on subsequent economic growth.

Although income inequality has no direct impact on growth, additional effects can arise from the influence of inequality on the explanatory variables. One of the effects suggested by previous studies involves the impact of income distribution on fertility. Row 2 of Table 8 [ PDF: 93kb | 1 page ] shows the estimation result for a panel system in which the log of the fertility rate is the dependent variable. In this system, the explanatory variables include the log of per capita GDP and the Gini index. The lagged values of the log of per capita GDP and the Gini index are used as instruments. The result confirms a strong positive impact of inequality on fertility.

In theories based on political economy arguments, inequality affects government expenditures and thereby growth. In unequal societies, there are more incentives for redistributive politics (Meltzer and Richard, 1981). Row 3 of Table 8 [ PDF: 93kb | 1 page ] shows direct consideration of a panel system in which government consumption ratio is the dependent variable. We find a significant influence from the Gini index.

Another channel by which income inequality influences growth is educational attainment. Poor families with borrowing constraints are not able to invest in their children even when the returns on education are very high. They have problems sending their children to school even under free schooling, since they often need income from their children's employment. This occurs relatively less in more equal societies, given the same level of income, since the parents are able to pay the costs of education. A more equal distribution enables more households to send their children to school. Row 4 of Table 8 [ PDF: 93kb | 1 page ] confirms that income inequality has a strong negative impact on secondary school enrollment. Lower secondary school enrollment leads to smaller secondary educational stock in time, and consequently has an adverse impact on economic growth. Thus, income distribution affects growth through the human capital channel.

We also find a strong negative impact of income inequality on institutional quality. Row 5 of Table 8 [ PDF: 93kb | 1 page ] shows the estimation result for a panel system in which the log of the rule-of-law index is the dependent variable. We find a significantly negative impact of the Gini index on the rule of law. Political economy considerations can also help to explain why corruption, rule of law, and institutional quality in general are weaker in more unequal societies.

Overall, we find substantial evidence that inequality affects growth indirectly by influencing fertility, government consumption, education and rule-of-law. Consider as an example the estimated coefficient of 0.143 on the Gini coefficient in row 3 of Table 8 [ PDF: 93kb | 1 page ] . This point estimate implies that an increase in the Gini coefficient by 0.1 (its standard deviation), that is 10 percentage points, raises the government-consumption ratio by 1.4 percentage points of GDP. If we multiply this value by the estimated coefficient of the government-consumption ratio in the growth regression (-0.07 in column 2 of Table 3 [ PDF: 132kb | 1 page ] ), we get -0.001. Thus, this indirect channel lowers economic growth by about 0.1 percentage point. Similarly, the point estimate of -0.87 of the Gini coefficient in the regressions for rule of law (row 5 of Table 8 [ PDF: 93kb | 1 page ] ) and the estimated coefficient of the rule-of-law index in the growth regression (0.018 in column 2 of Table 3 [ PDF: 132kb | 1 page ] ) imply that an increase in the Gini coefficient by 0.1 leads indirectly to a decrease in growth rate by about 0.16 percentage point through deteriorating institutional quality.

This evidence suggests that although income distribution has no significant direct effect on economic growth in our regressions (row 1 in Table 8 [ PDF: 93kb | 1 page ] ), inequality may be detrimental to economic growth by increasing distortions, weakening institutions and reducing the quality of human resources. More research needs to be done to establish the definite connections, since up to this point we have seen some very suggestive correlations.

In addition, the accounting exercises overall show that although the gap in the growth rate between Latin America and East Asia will narrow over the next decade, it will still remain substantial due to the differences in investment, fertility, schooling, as well as certain policy variables such as government consumption, rule of law, and inflation.

Therefore, Latin America must do more to improve the investment rate, fertility, schooling, and institutions. Two important policies to achieve this objective involve increasing public saving and expanding educational enrollment. An increase in public saving will contribute to raising the investment rate, containing pressures on government consumption, tax distortions, and high inflation rates. An increase in educational enrollment, particularly at the secondary level, will help to lower fertility and increase educational attainments.

However, improving public finance and education investment is not easy. As we analyzed in Section 3.2, high government expenditures and low educational enrollment are to a certain extent the outcome of unequal income distribution. Latin American countries have a more unequal income distribution than East Asian countries, and the evidence on the determinants of income distribution show that this gap cannot be closed in a short period of time. For example, improvements in education take time to pass through to a large share of the labor force (see, e.g., De Gregorio and Lee, 2002).


  1. empirical framework includes a representative set of the explanatory variables that have been widely used in previous work. See Barro and Sala-i-Martin (2004, Ch. 12) for details.
  2. We do not include the 1960s period in the regression because the currency-crisis variable is only available from 1970.
  3. The estimation gives countries equal weights but allows for different error variances in each period and for a correlation of these errors over time. Some studies suggest estimating panel growth regressions by the fixed-effects estimation technique, considering for an unobservable country fixed effect. However, the fixed-effects technique eliminates information from cross-section variations (see Barro, 1997, pp.36-39). Temple (1999) discusses other statistical problems concerning the estimation and interpretation of crosscountry growth regressions.
  4. Barro and Sala-i-Martin (2004) show that some additional regressors are statistically significant when added one at a time to the regression, in a similar way to our framework in Table 3. Notably, schooling quality and geography variables are found to be significant.

The views expressed in this paper are the views of the authors and do not necessarily reflect the views or policies of the Asian Development Bank Institute (ADBI), the Asian Development Bank (ADB), its Board of Directors, or the governments they represent. ADBI does not guarantee the accuracy of the data included in this paper and accepts no responsibility for any consequences of their use. Terminology used may not necessarily be consistent with ADB official terms..





[previous chapter] [next chapter]


Post a Comment

We welcome your feedback on this publication. Post a comment. ADBI is not obliged to acknowledge or publish comments and may abridge or edit them before web posting.

Comment(s)

There are [0] comment(s) for this entry. Post a comment.

    Back to Top 
    © 2014 Asian Development Bank Institute.