Economics Discussion Paper Series EDP-1502

Transcription

1 Economics Discussion Paer Series EDP-150 Education, Health, and Economic Growth Nexus: A Bootstra Panel Granger Causality Analysis for Develoing Countries Hüseyin Şen Ayşe Kaya Barış Alaslan January 015 Economics School of Social Sciences The University of Manchester Manchester M13 9PL

2 Education, Health, and Economic Growth Nexus: A Bootstra Panel Granger Causality Analysis for Develoing Countries This Version: February 13, 015 Hüseyin Şen Deartment of Public Finance, Faculty of Political Sciences, University of Yıldırım Beyazıt, Ankara, Turkey Ayşe Kaya Deartment of Public Finance, Faculty of Economics and Administrative Sciences, University of İzmir Kati Çelebi, İzmir, Turkey Barış Alaslan* Deartment of Economics, School of Social Sciences, University of Manchester, UK & Deartment of Public Finance, Faculty of Political Sciences, University of Yıldırım Beyazıt, Ankara, Turkey Abstract This aer emirically analyzes the ossible existence of Granger causality among three variables; education exenditure, health exenditure, and economic growth for the selected eight develoing countries Argentina, Brazil, Chile, India, Indonesia, Mexico, South Africa, and Turkey over the eriod For this urose, we emloy the Bootstra Panel Granger Causality aroach. Our analysis shows no robust evidence of Granger causality among education exenditure, health exenditure and economic growth for all the countries considered in this aer; only in two of eight develoing countries Brazil and Mexico a ositive and significant causality running from both education and health exenditures to economic growth was observed; however, this result was significantly negative for Indonesia. 1 Keywords: Education Exenditure, Health Exenditure, Human Caital, Economic Growth, Develoing Countries, Bootstra Panel Granger Causality Analysis. JEL Classification Numbers: I15, I5, O11, E6 *Corresonding Author Deartment of Economics, School of Social Sciences University of Manchester, Arthur Lewis Building, Oxford Road, Manchester, M13 9PL, UK baris.alaslan@manchester.ac.uk ** We are grateful to László Kónya for roviding his TSP codes, which we adated for our analysis, and helful comments on a revious version. We would also like to thank Arusha Cooray for helful suggestions. The views exressed in this aer and any errors are our own.

3 1. Introduction In the history of develoment economics, no matter how develoed countries are, the question of how countries can boost economic growth has been a controversial and much disuted subject for more than a half century. In fact, the discussions which have focused on the role of human caital in economic growth have grown in imortance with endogenous growth models since the mid-1980s. In articular, the existence of a ossible interlay among education, health, and economic growth has received an increased interest among researchers and olicymakers. Indeed, education and health, which are commonly regarded as a considerable comonent of human caital accumulation, lay a key role, as a catalyst, in a structural change in a society and economic transformation, and therefore stimulate long-run economic growth not only in low-income countries, but also in many develoed countries. Education is, for instance, learning and training rocess by which an individual acquires skills and knowledge. It is also regarded as a tool in romoting economic efficiency and social cohesion. Furthermore, countries with individuals who have a higher level of education can adot imorted technologies and develo technological innovation, thus fostering economic growth and develoment in the long-run. Moreover, a higher level of education increases marginal roductivity of hysical caital and labor force, and therefore romotes national income of a country. On the other hand, as reorted by World Develoment Reort (1993), good health may affect economic growth in a number of asects: firstly, good health eliminates roduction losses which can result from illness; secondly, it raises a number of children enrolled to school and erforming better in a cognitive and learning task; thirdly, it creates an oortunity for individuals to use existing resources, which would otherwise have to be sent on treating illness. Last, but by no means least, individuals with good health have higher income and contribute to a country s income by boosting roductivity. So far, there have been a large volume of ublished aers that have only catured the role of either education or health in economic growth; however, the direction of causality between these variables has been a controversial and much disuted subject within the field. This aer therefore intends to make a contribution and to rovide a value-added to the existing emirical literature in a number of asects: Firstly, unlike revious studies in the literature which have focused only on one-way or two-way causality between either education or health and economic growth, in this aer causality among these three variables, that is, education, health and economic growth, is emirically analyzed. Secondly, we emloy a more fitting aroach for our analysis, the bootstra anel causality method roosed by Kónya (006), which allows us to cature cross-sectional deendence and heterogeneity across countries under consideration. Finally, we focus on a grou of develoing countries: Argentina, Brazil, Chile, India, Indonesia, Mexico, South Africa, and Turkey, which have almost similar growth and develoment atterns, over the eriod The remainder of the aer is organized as follows: Section begins by the theoretical dimension of the research framework and looks at the interactions among education, health, and economic growth, and then reviews the related emirical literature. Section 3 describes the analytical framework in which model secification, data set, and estimation method are resented, whereas Section 4 reorts the emirical findings of the aer. And finally, Section 5 offers concluding remarks.

4 . Theoretical Background and Review of the Related Literature The role of education and health in the rocess of growth as well as economic develoment has been recently of rimary imortance. In the context of endogenous growth models, Romer (1986) and Lucas (1988) are well-known examles of studies that focused mainly on the role of human caital in economic growth. For instance, Lucas (1988) considered human caital as a cumulative variable with ositive externalities and as the main driving force behind economic growth. In other words, the main idea behind his argument is that individuals with a higher level of education will be more efficient and more roductive in their work life. Moreover, education will enhance roductivity, not only through the acquisition of skills individuals obtain, but also through romoting hysical caital and the adotion of technological develoment. A large body of recent research suggests that educational attainment is in fact a key driver of the acquisition of skills, better emloyment outcomes, individuals and country s well-being, and therefore economic growth [See, for examle, Romer (1990), Barro (1991), Barro and Lee (1993), Benhabib and Siegel (1994), Islam (1995), Barro and Sala-i-Martin (1995), Gemmell (1996), Sala-i-Martin (1997), Temle (1999), Hanushek and Kimko (000), Bils and Klenow (000), Kruger and Lindahl (001), Sianesi and Reenen (003)]. On the other hand, as Bloom et al. (004), Sala-i-Martin et al. (004), Gyimah-Bremong and Wilson (004), Jamison et al. (005), and Weil (007) remind us; good health imroves human welfare as well as labor roductivity, and ositively affects economic growth in both develoing and industrial countries. Conversely, a large number of studies, for instance, UNAIDS (004), UN (005), McDonald and Roberts (006), and WHO (007), have documented the adverse effects of articular diseases such as malaria, HIV/AIDS, influenza andemic, which is the case esecially in low-income countries, as well as in many other countries. Numerous studies, such as Strauss and Thomas (1998), Wang and Taniguchi (003), Hoddinott et al. (005), and Jensen and Lleras-Muney (01) also emhasize that inadequate nutrition, malnutrition, inadequate consumtion of rotein, energy and vitamin, smoking, and drinking, which are all closely linked to child and adult mortality, may cause oor health, which results in low level of labor roductivity and shortens life exectancy, and therefore have an adverse, indirect effect on economic growth. 3 It may be ossible, however, that these effects are overestimated or underestimated due to indirect effects of education on health or vice versa. For instance, in his very recent work, Agénor (01) reorted that good health and nutrition may hel children erform better in a cognitive and learning task, which increases school enrolment and educational attainment. Similar arguments have been also done in several revious studies, such as Behrman (1996), Bloom et al. (004), Ahmed and Arends-Kuenning (006), and Bleakley (007). Numerous studies have also attemted to exlain that longer life exectancy as a result of imroved health conditions increases the roensity to save and allows individuals to invest more in education and to be more roductive, which therefore has a growth-enhancing effect [See, for instance, Zhang et al. (003), Miguel and Kremer (004), Soares (006), Jayachandran and Lleras-Muney (009), Agénor (01)]. On the other hand, several studies have investigated the effect of education outcomes on health; the studies carried out by Tamura (006) and Agénor (01) are well-known examles of this oint. For instance, Agénor (01) as well as some other studies, such as Hurt et al. (004), Arendt (005), Albouy and Lequien (009), and Clark and Royer (013), suggest that individuals with better education are well-informed about nutritional and health risks not only for their own health but also for their family members, esecially for their children and souses.

5 In reviewing the emirical literature, to a large extent, the research has, however, tended to focus on one-way causality between either education or health and economic growth. Indeed, a large and growing body of literature has mostly used a single-equation aroach to estimate the imact of either variable, namely, education or health on economic growth. For instance, Barro and Lee (1993) emloyed a set of anel data to estimate the determinants of economic growth, hysical investment, and human caital accumulation as well as fertility for 19 countries over five-year eriods from Based on the findings of their study, educational attainment has a considerable exlanatory ower on economic growth; in other words, education is ositively correlated with economic growth. In the same vein, Benhabib and Siegel (1994) used Cobb-Douglas aggregate roduction function with hysical and human caital stocks and estimated cross-country growth-accounting regressions using Ordinary Least Squares (OLS) with Heteroscedasticity-consistent covariance method for the eriod Unlike the findings of Barro and Lee (1993), they concluded that human caital is insignificantly correlated with er caita growth rates. However, in an alternative model they develoed, human caital stock lays a significant role in the growth rate of total factor roductivity. Cheng and Hsu (1997) used the Johansen cointegration test and Granger causality technique by Hsiao (1981) to study the causality between human caital and economic growth in Jaan for the eriod They found a bi-directional causality between human caital and economic growth. In other words, the findings of their study showed that an increase in human caital has a growth-enhancing effect; at the same time, economic growth ositively affects human caital. Likewise, In and Doucouliagos (1997), who alied a Granger causality test to a new data set and used the canonical cointegration regression (CCR) estimation aroach, found a bi-directional causality between human caital formation and economic growth in the US over the eriod Using ooled aggregate data, Freire-Serén (00) estimated the equations of the dynamic system to investigate the relationshi between human caital and economic growth for the Sanish regions over the eriod According to their study, human caital ositively accounts for income growth and vice versa, indicating the existence of two-way causality between human caital and income growth. Furthermore, Nomura (007) estimated the model by Mankiw et al. (199) for a samle of 85 countries over the eriod Based on the Ordinary Least Square (OLS) regression method, the findings of the study reveal that the contribution of human caital to economic growth matters more and is statistically significant esecially in the countries where a low level but higher quality of education exists. In the similar vein, Tsamadias and Prontzas (01) followed the model by Mankiw et al. (199) to analyze the effect of education on economic growth in Greece during the eriod and showed a significant and ositive effect on economic growth during the eriod for which the study was carried out. A recent study by Boccanfuso et al. (013) used the analytical model develoed by Islam (1995) who considered a anel data aroach to study cross-country growth convergence over the eriod and introduced a new tye indicator of human caital to show the imortance of the qualitative asects of human caital and to analyze the question of whether human caital has a growth enhancing effect for a samle of African countries using anel data over the eriod According to the findings of their study, human caital lays a ositive role in the rocess of economic growth and convergence for the African countries. Another recent study by Uneze (013) imlemented anel cointegration and causality testing

6 aroaches for 13 Sub-Saharan Africa countries during the eriod and found a bidirectional causality between caital formation and economic growth. On the other hand, a large and growing body of studies have investigated the link between health and growth [See, e.g, Fogel (1994), Barro (1997), Sachs and Warner (1997), Bloom and Williamson (1998), Bhargava et al. (001), Mayer-Foulkes (001), Gyimah-Bremong and Wilson (004), Bloom et al. (004), Eide and Showalter (011)]. For instance, Barro (1997) used a anel data of around 100 countries over the eriod His study indicated that higher initial schooling and life exectancy have a growth-enhancing effect. In the same vein, Bhargava et al. (001) used a anel data aroach and studied the effects of health indicators, such as adult survival rates on GDP growth rates at 5-year intervals for a number of countries. They found that adult survival rates have a ositive imact on GDP growth rates in low-income countries. What is more, to investigate the role of health status in roductivity, Rivera and Currais (1999) used an extended version of the Solow model, which is closely related to the model by Mankiw et al. (199), and run a log-linear equation which is estimated using Ordinary Least Square (OLS) with White s heteroscedasticity-consistent covariance estimation method for OECD countries over the eriod The results of this study suort the revious research underlining the fact that health has a ositive imact on economic growth. Mayer-Foulkes (001) alied Barro s (1995) convergence model to a five-yearly database to exlore the long-term effect of health on economic growth in Mexico during the eriod Health imrovements were found to cause ermanent income increments in this country during the aforementioned eriod. The findings of Mayer-Foulkes (001) are consistent with those of the study by Fogel (1994) who reorted that better nutrition and health account for a third of economic growth in Great Britain over the last 00 years. Another study on the OECD countries by Hartwig (010) investigated the role of health caital formation in GDP growth for a samle of 1 OECD countries over the eriod by alying a anel Granger causality method. Interestingly, the findings of his study are, however, not consistent with those of other studies in the existing emirical literature, which revealed that human caital accumulation in the form of health affects long-term economic growth. A more recent study by Cooray (013) emloyed both Ordinary Least Squares (OLS) and Generalized Method of Moments (GMM) to analyze the differential effects of health on economic growth for a samle of 10 countries using anel data over the eriod The results for the full samle showed that health caital has no robust and significant effect on economic growth, unless through their interactions with health exenditure and education. However, based on the countries divided by income grous, health caital has no robust imact on economic growth in high and uer-middle income economies, whereas in low and low middle-income countries it has a statistically significant imact only through their interaction with education and health exenditure. 5 To the best of our knowledge, Li and Liang (010) is erhas the most relevant study to our aer. Based on an augmented version of Mankiw et al. (199) model, they emirically investigated the sources of economic growth for a grou of East Asian economies using a anel data set over the eriod According to their findings, the effects of the human caital in the form of both health and education on economic growth are statistically significant not only for the whole samle, but also for the sub-samle eriod. Their study also considers the sub-samle estimation of the ost-1997 Asian financial crisis for comarison uroses. Their findings suggest that it is more lausible for olicymakers in the East Asia to invest more in health than education.

7 All the studies reviewed so far have, however, overlooked Granger causality among these three variables; education, health, and economic growth. This aer, therefore, shows and critically evaluates a good awareness of the existing literature to identify unexlored/unsolved issues that are both theoretically interesting and real-world relevant in the methodological aroach to analyzing this for a grou of selected develoing countries. We now turn to data set and estimation method. 3. Data Set and Estimation Method 3.1. Data Set In this aer, we used annual data set abstracted from the World Develoment Indicators (WDI) database of the World Bank for the selected eight develoing countries Argentina, Brazil, Chile, India, Indonesia, Mexico, South Africa and Turkey. The data set covers the eriod sanning from 1995 to 01. There are several reasons why we have considered in articular these countries: Firstly, in comarison to other develoing countries, they are all fast-growing countries. Secondly, desite their different economic structures as well as their olicies and olitical systems, they have recently made a remarkable economic rogress, which makes it ossible for these countries to be among the world s largest and most owerful economies in the near future. Thirdly, excet Chile, they are all a member of G-0 countries and are more likely to have a significant voice within their own geograhic region and a growing global influence in the time to come. Finally, although these countries have recently shown a sustainable rate of growth, albeit below the world s average, and roved an economic take-off in recent years, their role in contributing to adoting new technologies is still limited, which enables us to study an in-deth analysis of causality among education, health [which are both imortant comonents of human caital accumulation, as noted earlier], and economic growth. 6 In reviewing the emirical literature, as noted earlier, most of the studies rely on causality between either education or health and economic growth. However, in this aer, we analyze the causal relation among these three variables. For this urose, we use education exenditure and health exenditure [as a share of GDP], and annual ercentage growth rate of GDP as a measure. 3.. Estimation Method In general, there are three alternative estimation methods that can be alied to examining the direction of Granger causality in a anel data analysis: The first aroach is based on estimating a anel vector error correction model (VECM) by means of a generalized method of moments (GMM) estimator that estimates a anel model by eliminating the fixed effect. However, this method accounts neither for heterogeneity nor for cross-sectional deendence. A second method roosed by Hurlin (008) resents a anel data Granger causality test which does account for sloe heterogeneity but disregards cross-sectional deendence; even so, substantial biases and size distortions may occur. However, the third aroach roosed by Kónya (006) allows us to study both heterogeneity and cross-sectional deendence. Overall, we believe that Kónya s (006) aroach has three sueriorities over other alternative aroaches: Firstly, this aroach is based on the seemingly unrelated regression (SUR)

8 estimation which considers cross-sectional deendence across countries. Secondly, based on the Wald test with country-secific bootstra critical values, this aroach does not require the joint hyothesis for all members of a anel. And finally, considering the fact that unit root tests may suffer from low ower, it requires no re-testing for unit roots and any cointegrating relationshis. In the light of all the methods reviewed above, this aer follows the bootstra anel Granger causality aroach by Kónya (006), which considers cross-section deendency and crosscountry heterogeneity. On the basis of country-secific bootstra critical values, this method allows us to test the Granger causality for each individual country by taking into account the ossible contemoraneous correlation across countries. A brief account of the econometric method used in this aer is resented below: Cross-Sectional Deendence To investigate the existence of cross-sectional deendence, we emloy three different tests: Lagrange multilier test statistic of Breusch and Pagan (1980) for cross-sectional deendence and two cross-sectional deendence test statistics of Pesaran (004), one based on Lagrange multilier and another based on the air-wise correlation coefficients. The first is the Lagrange Multilier (LM) test develoed by Breusch and Pagan (1980), which requires the estimation of the following anel data model: Y it = α i + β i X it + μ it [1] for i = 1,, 3,, N; t = 1,, 3,, T 7 where i is the cross section dimension; t is the time dimension; X it is kx1 vector of exlanatory variables, while α i and β i are the individual intercets and sloe coefficients that are allowed to differ across states. In the LM test, the null hyothesis of no cross-sectional deendence H 0 : Cov (μ it, μ jt ) = 0 for all t and i j is tested against the alternative hyothesis or cross-sectional deendence H 1 : Cov (μ it, μ jt ) 0 for at least one air of i j. For testing null hyothesis, the Lagrange multilier test statistic for cross-sectional deendence (hereafter, CD BP ) of Breusch and Pagan (1980) is given by: CD BP = T N 1 i 1 N ρ ij j i 1 [] where ρ ij is the estimated correlation coefficient among the residuals obtained from individual OLS estimation of Equation 1. Under the null hyothesis, the LM statistic has an asymtotic chi-square distribution with N (N 1)/ degrees of freedom. However, Pesaran (004) indicates that the CD BP test has a drawback when N is large, imlying that it is not alicable when N. To overcome this roblem, the following Lagrange multilier statistic for the cross-sectional deendence (hereafter, CD LM ) were develoed by Pesaran (004). The CD LM statistic is given as follows:

9 CD LM = 1 N(N 1) N 1 i 1 N (Tρ ij -1) [3] j i 1 Under the null hyothesis of no cross-sectional deendence with T and then N, CD LM asymtotically follows a normal distribution. On the other hand, CD LM test is likely to indicate substantial size distortions when N is large relative to T. Pesaran (004) therefore rooses a new test for cross-sectional deendence (hereafter, CD) that can be used where N is large and T is small. The CD statistic is calculated as follows: CD = T N(N 1) N 1 i 1 N ρ ij [4] j i 1 According to Pesaran (004) under the null hyothesis of no cross-sectional deendence with T and N in any order, the CD test is asymtotically normally distributed. However, Pesaran et al. (008) state that while the oulation average air-wise correlations are zero, the CD test will have less ower. Therefore, they roose a bias-adjusted test that is a modified version of the LM test by using the exact mean and variance of the LM statistic. The bias-adjusted LM statistic is calculated as follows: CD adj = T N(N 1) N 1 i 1 N (T k)ρ ij utij ρ ij j i 1 v Tij [5] 8 where u Tij and v Tij are the exact mean and variance of (T-k) ρ ij, which are rovided by Pesaran et al. (008). Under the null hyothesis of no cross-sectional deendence with T first followed by N, the results of the CD adj test follow an asymtotic standard normal distribution Sloe Homogeneity Tests The standard F test is the most widely used way to test the null hyothesis of sloe homogeneity H 0 : β i = β for all i against the hyothesis of heterogeneity H 1 : β i β j for a non-zero fraction of air-wise sloes for i j. This requires that the exlanatory variables are strictly exogenous, and the error variances are homoscedastic. In order to relax the assumtion of homoscedasticity in the F test, Swamy (1970) develoed the sloe homogeneity test that examines the disersion of individual sloe estimates from a suitable ooled estimator. Pesaran and Yamagata (008) state that both the F test and Swamy s test require anel data models where N is relatively small comared to T. Therefore, they roose a standardized version of Swamy s test (hereafter, test) for testing sloe homogeneity in large anels. The test is valid when (N, T) without any restrictions on the relative exansion rates of N and T when the error terms are normally distributed. Swamy s statistic can then be modified as:

10 S = N (β i β WFE ) X i Mτ X i (β i β WFE ) [6] σ i i 1 where β i is the ooled OLS estimator; β WFE is the weighted fixed effect ooled estimator of the Equation 1; M τ is an identity matrix of order T and σ i is the estimator of σ i. Pesaran and Yamagata (008) then develoed the following standardized disersion statistic: = N ( N 1 S k k ) [7] Under the null hyothesis with the condition of (N, T) and so long as N/T, and when the error terms are normally distributed, the test has an asymtotic standard normal distribution. The small samle roerties of the test can be imroved when there are normally distributed errors by using the following mean and variance bias adjusted version: adj = N ( N 1 S E(Z it ) var(z it ) ) [8] where the mean E(Z it ) = k, and var(z it ) = k(t-k-1)/(t+1) Panel Granger Causality Test The anel Granger causality technique roosed by Kónya (006) entails describing a system which includes three sets of equations. His aroach can be formulated as follows: 9 1 EG 1t = α 11 + β 1 11lEG 1t l + δ 1 11lEE 1t l + φ 11lHE 1t l + ε 11t 1 EG Nt = α 1N + β 1 1NlEG Nt l + δ 1 1NlEE Nt 1 + φ 1NlHE Nt l + ε 1Nt [9] EE 1t = α 1 + β 1l EG 1t l + δ 1l EE 1t l + φ 1l HE 1t l + ε 1t EE Nt = α N + β Nl EG Nt l + δ Nl EE Nt l + φ Nl HE Nt l + ε Nt [10] HE 1t = α 31 + β 31l EG 1t l + δ 31l EE 1t l + φ 31l HE 1t l + ε 31t HE Nt = α 3N + β 3Nl EG Nt l + δ 3Nl EE Nt l + φ 3Nl HE Nt l + ε 3Nt [11]

11 where EG, EE, and HE denote economic growth, education exenditure, and health exenditure, resectively. N is the number of countries of anel (i = 1,, 3,, N), t is the time eriod (t = 1,, 3,, T), and l is the lag length. The error terms, ε 1Nt, ε Nt and ε 3Nt, are suosed to be white-noises (i.e. they have zero means, constant variances and are individually serially uncorrelated) and may be correlated with each other for a given country. Moreover, it is assumed that EG, EE and HE are stationary or cointegrated so, deending on the time series roerties of the data, they might denote the level, the first difference or some higher difference. To test for the anel Granger causality in this system, alternative causal relations for a country are likely to be found. For examle, there is one way Granger causality from EE to EG if not all δ 1,i are zero, but all β,i are zero; there is one-way Granger causality from EG to EE if all δ 1,i are zero, but not all β,i are zero; there is two-way Granger causality between EE and EG if neither δ 1,i nor β,i is zero; there is no Granger causality between EE and EG if all δ 1,i and β,i are zero. This definition can easily be extended to causal relations between education exenditure and health exenditure, and economic growth. To determine the direction of causality, Wald statistics for Granger causality are comared with country-secific critical values that are obtained from the bootstra samling rocedure. 4. Emirical Findings In this section, we reort the emirical results of the aer. Before considering anel data Granger causality analysis, we tested for cross-sectional deendency and sloe homogeneity among the countries that we considered in this study. The results are resented in Table 1. Table 1: Cross Sectional Deendency and Sloe Homogeneity Tests Cross-Section Deendency Tests Statistic -value LM (Breusch and Pagan, 1980) * CD lm (Pesaran, 004) * CD (Pesaran, 004) 4.471* LM adj (Pesaran and Yamagata, 008).33* Sloe Homogeneity Tests 3.37* * adj Note: (*) indicates rejection of the null hyothesis at 1% level of significance. The data covers the whole samle eriod from 1995 to 01. Source: Authors calculations 10 As can be seen from Table 1, the results show that the null hyothesis of no cross-sectional deendence across the countries is strongly rejected at 1% level of significance, imlying that the seemingly unrelated regressions (SUR) method is aroriate rather than country by country OLS estimation. This result also shows that a shock, which may occur in one of the selected develoing countries, seems to influence other countries. Our results thus indicate that selected eight develoing countries have highly integrated economies, and when a shock occurs in one of them, it will then affect the others. On the other hand, the results significantly reject the null hyothesis, and indicate not only that education and health influence economic growth in each country, but also that the regression error terms among countries also affect each other.

12 Table 1 also reveals the results from the two sloe homogeneity tests which show that the null hyothesis of the sloe homogeneity is rejected thus suorting the country-secific heterogeneity. This result imlies that the direction of Granger causality between variables in our eight develoing countries might be heterogeneous and the direction of causal linkages among the variables may differ across countries. Our results suort the alternative hyothesis that heterogeneity exists among countries, and thus that individual countries are affected by their own secific characteristics. The existence of cross-sectional deendence and sloe heterogeneity among our selected eight develoing countries means that it is aroriate to use the Bootstra anel Granger causality method by Kónya (006). Having established the existence of cross-sectional deendence and the heterogeneity across countries, we determine the otimum lag structure by following Kónya (006) where the maximal lags are allowed to differ across variables but to be the same across equations. Due to the fact that the results from the Granger causality test may be sensitive to the lag structure, determining the otimal lag length(s) is crucial as to the robustness of the findings. Kónya (006) oints out that the selection of the otimal lag structure is very imortant since the Granger causality test results rely on this. To determine the otimal lag structure, we follow Kónya's aroach in which maximal lags are allowed to vary across variables, but to remain the same across equations, as noted earlier. We estimate the system for each ossible trinity of 1 1 1, and by assuming from one to four lags, and then choose the combinations which minimize the Akaike Information Criterion (AIC) and Schwartz Information Criterion (SIC). Table : Granger Causality between Education Exenditure and Economic Growth Country Estimated Coefficient Wald Test. Stat. Bootstra Critical Values 1% 5% 10% H 0 = EE does not cause EG Argentina *** Brazil Chile India Indonesia *** Mexico South Africa Turkey H 0 = EG does not cause EE Argentina Brazil Chile India Indonesia Mexico South Africa *** Turkey *** Note: The data covers the whole samle eriod from 1995 to 01. (***) indicates statistical significance at 10%. Critical values are based on 1000 bootstra relications. Source: Authors calculations 11 The results in Table show that only for Argentina and Indonesia, there exists a significant and ositive Granger causality at 10% level of significance running from education exenditure to economic growth, whereas for the other countries there is no significant causality between these variables. On the other hand, the same table also indicates that there is a significant and

13 ositive Granger causality running from economic growth to education exenditure for two countries, i.e. at 10% level of significance for South Africa; at 5% and 10% levels of significance for Turkey. Table 3: Granger Causality between Education Exenditure and Health Exenditure Country Estimated Coefficient Wald Test. Stat. Bootstra Critical Values 1% 5% 10% H 0 = EE does not cause HE Argentina *** Brazil Chile India *** Indonesia *** Mexico South Africa Turkey *** H 0 = HE does not cause EE Argentina *** Brazil *** Chile *** India Indonesia Mexico South Africa Turkey Note: The data covers the whole samle eriod from 1995 to 01. (***) indicates statistical significance at 10%. Critical values are based on 1000 bootstra relications. Source: Authors calculations Table 3 reorts the results of Granger causality between education exenditure and health exenditure. The results indicate a ositive Granger causality running from education exenditure to health exenditure in the case of Argentina, India, Indonesia and Turkey. Table 3 also shows that only for Indonesia, there is a significant negative Granger causality running from education exenditure to health exenditure, whereas for the other countries Brazil, Chile, Mexico and South Africa there is no causal relation between education exenditure and health exenditure. On the other hand, one can also see from the table that there is a significant and ositive Granger causality running from health exenditure to education exenditure at 10% level of significance for both Argentina and Brazil, whereas it is significantly negative for Chile. 1

14 Table 4: Granger Causality between Health Exenditure and Economic Growth Country Estimated Coefficient Wald Test. Stat. Bootstra Critical Values 1% 5% 10% H 0 = HE does not cause EG Argentina Brazil Chile India *** Indonesia *** Mexico South Africa Turkey H 0 = EG does not cause HE Argentina Brazil *** Chile India *** Indonesia Mexico *** South Africa Turkey Note: The data covers the whole samle eriod from 1995 to 01. (***) indicates statistical significance at 10%. Critical values are based on 1000 bootstra relications. Source: Authors calculations Table 4 resents the results from Granger causality analysis between health exenditure and economic growth. For India and Indonesia, there is a one-way ositive Granger causality running from health exenditure to economic growth at 10% level of significance. However, there is a ositive Granger causality between economic growth and health exenditure at 10% level of significance for Brazil, India and Mexico. 13 Table 5: Granger Causality between Education Exenditure, Health Exenditure, and Economic Growth Country Estimated Coefficient Wald Test. Stat. Bootstra Critical Values 1% 5% 10% H 0 = EE and HE do not cause EG Argentina Brazil *** Chile India Indonesia *** Mexico *** South Africa Turkey Note: The data covers the whole samle eriod from 1995 to 01. (***) indicates statistical significance at 10%. Critical values are based on 1000 bootstra relications. Source: Authors calculations Finally, the results in Table 5 generally indicate that there is no causal relationshi for the most countries we incororated into our emirical analysis. In other words, the null hyothesis of non-causality is acceted for Argentina, Chile, India, South Africa and Turkey. The results show that there is a significant and ositive causality running from both education exenditure and health exenditure to economic growth for Brazil and Mexico, whereas there exists a significant and negative causality for Indonesia at 10% level of significance. Overall, in this study, weak evidence of causal relation between education exenditure, health exenditure, and economic growth was found for all the develoing countries, excet Brazil,

15 Mexico and Indonesia. However, it is imortant to note that in some cases, the resent findings seem to be consistent with other research in the literature, which found a significant and ositive Granger causality between either education exenditure or health exenditure and economic growth. For instance, for Turkey, the emirical findings of this aer show a significant and ositive Granger causality running from economic growth [education exenditure] to education exenditure [health exenditure], whereas, as can be seen from the tables, in all other cases, insignificant Granger causality between these variables was reorted. Table 6 summarizes the direction of Granger causality among education, health, and economic growth for all the countries under consideration. Table 6: Direction of Granger Causality between Develoing Countries Direction of Causality EE EG EG EE EE HE HE EE HE EG EG HE EE, HE EG Develoing Countries Argentina and Indonesia: Significant and ositive Brazil, Chile, India, Mexico, South Africa and Turkey: Insignificant South Africa and Turkey: Significant and ositive Argentina, Brazil, Chile, India, Indonesia and Mexico: Insignificant Argentina, India and Turkey: Significant and ositive Indonesia: Significant and negative Brazil, Chile, Mexico and South Africa: Insignificant Argentina and Brazil: Significant and ositive Chile: Significant and negative India, Indonesia, Mexico, South Africa and Turkey: Insignificant India and Indonesia: Significant and ositive Argentina, Brazil, Chile, Mexico, South Africa and Turkey: Insignificant Brazil, India and Mexico: Significant and ositive Argentina, Chile, Indonesia, South Africa and Turkey: Insignificant Brazil and Mexico: Significant and ositive Indonesia: Significant and negative Argentina, Chile, India, South Africa and Turkey: Insignificant 14 Notes: EE, HE, EG denote education exenditure, health exenditure and economic growth, resectively. reresents the causal direction. Source: Authors summary 5. Concluding Remarks In this aer, we emirically studied the ossible existence of the Granger causal relationshi among education exenditure, health exenditure, and economic growth for the selected eight develoing countries: Argentina, Brazil, Chile, India, Indonesia, Mexico, South Africa, and Turkey for over the eriod To do so, we emloyed the bootstra anel Granger causality technique roosed by Kónya (006), which considers cross-sectional deendence and heterogeneity across the countries. Our analysis showed one-way causality from education exenditure to economic growth for Argentina and Indonesia, and one-way causality from economic growth to education exenditure for South Africa and Turkey. The findings of this aer also indicated one-way causality from education exenditure to health exenditure for India and Turkey and one-way causality from

16 health exenditure to education exenditure for Brazil; however, interestingly, two-way causality between these two variables was observed only for Argentina. Also, one-way causality from health exenditure to economic growth for Indonesia and one-way causality from economic growth to health exenditure for Brazil and Mexico but two-way causality between these variables for India were found. In addition, only in two of eight develoing countries Brazil and Mexico a ositive and significant causality running from both education and health exenditures to economic growth was observed; however, this result was significantly negative for Indonesia. In all cases, only for Chile insignificant or no causal relation among these three variables was found. It is worth noting that our emirical findings indicate mixed results that cannot be generalized. There could be several ossible exlanations for these results. Firstly, a ossible exlanation for this might be that given the limited amount of resources governments have, share of government sending on education and health in GDP is very low. Secondly, it might be due to teaching quality, which has recently come to the fore esecially in develoing countries as in many low-income countries. Finally, it also seems ossible that these findings are due to inefficient or corrut bureaucracy in education and health, which is again the case in a number of develoing countries. 15

17 References Agénor, P.-R. (01), Public Caital, Growth and Welfare, Princeton University Press, Princeton: New Jersey. Ahmed, A. and Arends-Kuenning, M. (006), Do Crowded Classrooms Crowd Out Learning? Evidence from the Food for Education Program in Bangladesh, World Develoment, Vol: 34, Albouy, V. and Lequien, L. (009), Does Comulsory Education Lower Mortality?, Journal of Health Economics, Vol: 8(1), Arendt, J.N. (005), Does Education Cause Better Health? A Panel Data Analysis Using School Reforms for Identification, Economics of Education Review, Vol: 4(), Barro, R. (1991), Economic Growth in a Cross-Section of Countries, Quarterly Journal of Economics, Vol: 106 (), Barro, R. J. (1997), Determinants of Economic Growth: A Cross-Country Emirical Study, MIT Press: Cambridge, Mass. Barro, R. and Lee, J-W. (1993), International Comarisons of Educational Attainment, Journal of Monetary Economics, Vol: 3(3), Barro, R. J. and Sala-i-Martin, X. (1995), Economic Growth, McGraw-Hill: New York. Behrman, J.R. (1996), The Imact of Health and Nutrition on Education, World Bank Research Observer, Vol: 11(1), Benhabib, J. and Siegel, M. (1994), The Role of Human Caital in Economic Develoment: Evidence from Cross-Country Data, Journal of Monetary Economics, Vol: 34, Bhargava, A., Jamison, D. T. L., Lau, J. and Murray, C. J. L. (001), Modeling the Effects of Health on Economic Growth, Journal of Health Economics, Vol: 0(3), Bils, M., and Klenow, P. J. (000), Does Schooling Cause Growth, American Economic Review, Vol: 90(5), Bleakley, H. (007), Disease and Develoment: Evidence from Hookworm Eradication in the American South, Quarterly Journal of Economics, Vol: 1, Bloom, D. E., Canning, D. and Sevilla, J. (004), The Effect of Health on Economic Growth: A Production Function Aroach, World Develoment, Vol: 3, Issue: 1, Bloom, D. E. and Williamson, J. (1998), Demograhic Transitions and Economic Miracles in Emerging Asia, World Bank Economic Review, Vol: 1, Boccanfuso, D., Savard, L., and Savy, B.E. (013), Human Caital and Growth: New Evidences from African Data, International Economic Journal, Vol: 7, No: 1,

18 Breusch, T. and Pagan, A. (1980), The Lagrange Multilier Test and Its Alication to Model Secifications in Econometrics, Reviews of Economics Studies, Vol: 47, Cheng, B.S. and Hsu, R.C. (1997), Human Caital and Economic Growth in Jaan: An Alication of Time Series Analysis, Alied Economics Letters, Vol: 4, Clark, D. and Royer, H. (013). The Effect of Education on Adult Health and Mortality: Evidence from Britain, American Economic Review, Vol: 103, No: 6, Cooray, A. (013), Does Health Caital Have Differential Effects on Economic Growth?, Alied Economics Letters, Vol: 0, Eide, R. E. and Showalter, M. H. (011), Estimating the Relation between Health and Education: What Do We Know and What Do We Need to Know?, Economics of Education Review, Vol: 30, Fogel, R.W. (1994), Economic Growth, Poulation Health and Physiology: The Bearing of Long-term Processes on the Making of Economic Policy, American Economic Review, Vol: 84, Freire-Serén, M.J. (00), On the Relationshi between Human Caital Accumulation and Economic Growth, Alied Economics Letters, No: 9, Gemmell, N. (1996), Evaluating the Imacts of Human Caital Stocks and Accumulation on Economic Growth: Some New Evidence, Oxford Bulletin of Economics and Statistics, Vol: 58(1), Gyimah-Bremong, K., and Wilson, M. (004), Health, Human Caital, and Economic Growth in Sub-Saharan African and OECD Countries, Quarterly Review of Economics and Finance, Vol: 44, Hanushek, E. A. and Kimko, D. D. (000), Schooling, Labor-Force Quality, and the Growth of Nations, American Economic Review, Vol: 90(5), Hartwig, J. (010), Is Health Caital Formation Good for Long-Term Economic Growth? Panel Granger Causality Evidence for OECD Countries, Journal of Macroeconomics, Vol: 3, Issue: 1, Hoddinott, J., Alderman, H. and Behrman, J. (005), Nutrition, Malnutrition and Economic Growth in Health and Economic Growth: Findings and Policy Imlications, ed. by Guillem Lóez-Casasnovas, Berta Rivera, and Luis Currais, MIT Press: Cambridge, Mass. Hsiao, C. (1981), Autoregressive Modeling and Money Income Causality Detection, Journal of Monetary Economics, Vol: 7, Hurlin, C. (008), Testing for Granger Non Causality in Heterogeneous Panels, Mimeo. Deartment of Economics: University of Orleans.

19 Hurt, L.S., Ronsmans, C., and Saha, S. (004). Effects of Education and Other Socioeconomic Factors on Middle Age Mortality in Rural Bangladesh, Journal of Eidemiology and Community Health, Vol: 58(4), In, F. and Doucouliagos, C. (1997), Human Caital Formation and US Economic Growth: A Causality Analysis, Alied Economics Letters, Vol: 4, Islam, N. (1995), Growth Emirics: A Panel Data Aroach, Quarterly Journal of Economics, Vol: 110(1195), Jamison, Dean T., Lau, L. J. and Wang, J. (005), Health s Contribution to Economic Growth in an Environment of Partially Endogenous Technical Progress, in Health and Economic Growth: Findings and Policy Imlications, ed. By Guillem Lóez-Casasnovas, Berta Rivera, and Luis Currais, MIT Press: Cambridge, Mass. Jayachandran, S., and Lleras-Muney, A. (009), Longevity and Human Caital Investments: Evidence from Maternal Mortality Declines in Sri Lanka, Quarterly Journal of Economics, Vol: 14(1), Jensen, R. and Lleras-Muney, A. (01), Does Staying in School (and Not Working) Prevent Teen Smoking and Drinking? Journal of Health Economics, Vol: 31(4), Kónya, L. (006), Exorts and Growth: Granger Causality Analysis on OECD Countries with a Panel Data Aroach, Economic Modelling, Vol: 3(6), Krueger, A. B. and Lindahl, M. (001), Education for Growth: Why and for Whom?, Journal of Economic Literature, Vol: 39(4), Li, H. and Liang, H. (010), Health, Education, and Economic Growth in East Asia, Journal of Chinese Economic and Foreign Trade Studies, Vol: 3, No:, Lucas, R.E. (1988). On the Mechanics of Economic Develoment, Journal of Monetary Economics, Vol:, Mankiw, G., Romer, D. and Weil, D. (199), A Contribution to the Emirics of Economic Growth, Quarterly Journal of Economics, Vol: 107, Mayer-Foulkes, D. (001), The Long-Term Imact of Health on Economic Growth in Mexico, , Journal of International Develoment, Vol: 13(1), McDonald, S., and Roberts, J. (006), AIDS and Economic Growth: A Human Caital Aroach, Journal of Develoment Economics, Vol: 80, Miguel, E. and Kremer, M. (004), Worms: Identifying Imacts on Education and Health in the Presence of Treatment Externalities, Econometrica, Vol: 7(1), Nomura, T. (007), Contribution of Education and Educational Equality to Economic Growth, Alied Economics Letters, Vol: 14,