SOCIAL EXPENDITURE AND ECONOMIC GROWTH: EVIDENCE FROM AUSTRALIA AND NEW ZEALAND USING COINTEGRATION AND CAUSALITY TESTS

SOCIAL EXPENDITURE AND ECONOMIC GROWTH: EVIDENCE FROM AUSTRALIA AND NEW ZEALAND USING COINTEGRATION AND CAUSALITY TESTS Habibullah Khan GlobalNxt University, Malaysia Omar K M R Bashar* Swinburne University of Technology, Australia ABSTRACT This study seeks to establish links between social expenditures and economic growth in Australia and New Zealand, and draw lessons for fast developing ASEAN economies as they aspire to be developed nations soon. Using annual data from 1980 to 2012, we deploy cointegration and error correction methods for establishing long-run relationship, and Granger causality tests for testing short-term direction of causality among the variables. For Australia, economic growth is found to have three main determinantseducation, health and social expenditures. For New Zealand, health and social expenditures have been found as the main determinants of growth. However, no long-run relationship could be established among the variables when we included budget deficit in our model. The Granger causality tests indicate that one way causality running from economic growth to health expenditure, and social expenditure to economic growth in Australia. In case of New Zealand, on the other hand, one-way causality runs from education expenditure to economic growth, health expenditure to education expenditure, economic growth to health expenditure, and education expenditure to budget deficit. Social welfare expenditures also Granger cause economic growth. Our findings suggest that social expenditures promoted growth in Australia and New Zealand with some implications for ASEAN countries that usually do not allocate sizeable portion of their budget for social welfare. The fast developing economies such as Singapore and Malaysia, which aim to achieve the developed country status by 2020 should adopt more generous social policies for the sake of a balanced development, argued in the paper. JEL Classifications: O23, O40 Keywords: Social expenditure, economic growth, Australia, New Zealand, ASEAN, cointegration, causality * Corresponding Author s Email Address: obashar@swin.edu.au INTRODUCTION Social expenditures such as education, health and social welfare expenditures are required for balanced development of an economy. In 2013, Nordic countries such as Iceland, Sweden, Norway, Denmark and Finland on average spent more than 30 percent of their GDP on social welfare. On comparison, Australia and New Zealand s share of social welfare was 19.5 and 22 percent of GDP, respectively (OECD, 2014). Table 1 in the appendix provides a glimpse of social welfare expenditures as percent of GDP in various OECD countries. We can see that high income OECD countries such as Denmark, Ireland, Luxembourg, the Netherlands, Japan, Sweden, Norway, United Kingdom, and the United States have persistently maintained a generous policy of social welfare (at least 20% of GDP) over the years. On contrary, Asian developing countries spend much less in social welfare expenditures. For instance, Singapore allocated only 2.8 percent of its GDP for social welfare in 2011, while the figure was just below 3 percent in Malaysia in 2012 (ILO). There is a growing consensus among economists and policymakers that intervention by the states through social welfare provisions can ensure sustained economic growth. Contemporary research on the link between social expenditures and economic development traditionally followed two approaches: social indicator approach and social expenditure approach. The social indicator approach relies on output indicators measuring progress in economic activities using various multivariate techniques, such as Principal Component Analysis, Factor Analysis, and Cluster Analysis. The social expenditure approach, on the other hand, uses certain input indicators to see if there is any direct links between such input variables and economic growth. Empirical evidence on the effects of social expenditures on growth has been mixed. Mercan and Sezer (2014) using cointegration analysis studied the link between education expenditure and economic growth on Turkish economy for the period of 1970-2012 and found positive effect of education expenses on growth. Beraldo et al.

(2009) using growth accounting framework on data from 19 OECD countries between 1971 and 1998 found that health and education expenditures positively affect growth. Similar findings were reported by Bellettini and Ceroni (2000), Schultz (1961), Psacharopoulos (1985) and Rosenzweig (1996). Fic and Ghate (2005) using Markov switching framework over the period 1950-2001 reported negative impact of welfare state regimes on economic growth. McDonald and Miller (2010) using US data from 1976 to 2006 found that welfare programs have no direct effect on a state s economy. Further, the study revealed negative effect of welfare programs through investment. Similar findings were reported by earlier studies such as Landau (1983) and Barro (1989). Baum and Lin (1993) using cross sectional data for 58 countries from 1975 to 1985 with the help of regression techniques based on constant returns to scale production function, reported mixed findings. The study found positive impact of education expenditures but negative impact of welfare expenditures on growth. Although there has been significant progress in social indicators research, relatively less attention has been put on social expenditure research, especially in the Asia-Pacific region. Our study aims to fill-in this gap in literature and draw lessons for fast growing economies in the ASEAN region. We ask: do social expenditures promote growth in Australia and New Zealand? Should ASEAN follow a more generous policy on social welfare? This paper is organized as follows: After an introduction to the subject matter that includes a brief review of the literature in section 1, the methodological issues are explained in section 2. The results of the analysis are presented in section 3 and the policy implications are discussed in section 4. Finally, conclusions are stated in section 5. METHODS AND DATA We specify the economic growth function for Australia and New Zealand as follows: Y f ( EDU, HEALTH, SOCIAL ) (1) Y f ( EDU, HEALTH, SOCIAL, DEF ) (2) Where, Y is GDP per capita (constant US$ 2005), EDU is education expenditure to GDP ratio, HEALTH is public health expenditure to GDP ratio, SOCIAL is social welfare expenditure as percentage of GDP, and DEF is government deficit as percentage of GDP. Equation (1) is the base model, whereas we augment it by deficit expenditure in equation (2), since budgetary deficit is likely to be created by excessive public expenditures. Thus, our growth function for base model (equation 1) becomes ln PCY EDU HEALTH SOCIAL u (3) t 0 1 t 2 t 3 And for equation 2, the growth model is ln PCY EDU HEALTH SOCIAL DEF u (4) t 0 1 t 2 t 3 t 4 Expected sign : 0 1 ; ; 2 0 3 0; 0 t 4 The error correction ) (EC term lagged one period, which integrates short-term dynamics in the long-run growth function is shown below through error correction model (ECM) in the following two equations: t t t

Where, EC t 1 is error correction term lagged one period. The ECM introduces an additional channel through which Granger causality could be detected. If two variables are cointegrated, there exists a causal relationship between them (Granger, 1988). We use annual data for the period 1980-2012. Data sources include various international compilations such as, UNESCO Institute for Statistics Online Database by the United Nations Educational, Scientific and Cultural Organization (UNESCO), International Financial Statistics by the International Monetary Fund (IMF), World Development Indicators by the World Bank, OECD ilibrary by the Organisation for Economic Co-operation and Development (OECD), and ILO Online Database by the International Labor Organization (ILO), and Australia and New Zealand publications by the Australian Bureau of Statistics and New Zealand Treasury. The modeling strategy follows a four-step procedure: i) determine the order of integration of the variables using Augmented Dickey-Fuller (ADF, 1981) and Phillips- Perron (1988) unit root tests. In case of contradictory findings, use Kwiatkowski-Phillips-Schmidt-Shin (KPSS, 1992) test of unit root. ii) if the variables are found to be integrated of same order, apply the Johansen-Juselius (1990, 1992, 1994) maximum likelihood method of cointegration to determine the number of cointegrating vectors. On the other hand, if the variables are found to be integrated of different order, make them integrated of same order through differencing before determining the number of cointegrating vectors. We will apply trace test and maximum eigenvalue test of cointegration. iii) if the variables are found to be cointegrated, estimate error correction model using standard methods and diagnostic tests. We will include the I(0) variables (which have been omitted in cointegration tests) while estimating vector error correction models. iv) run the Granger causality test in order to determine causal relationship among the variables. RESULTS Australia In order to analyze time-series properties of the data, we conduct Augmented Dickey-Fuller (ADF) and Phillips- Perron (PP) tests at both level and first difference for all variables in the model. Results of the unit root tests have been shown in Table 2. TABLE 2. UNIT ROOT TESTS FOR STATIONARITY (AUSTRALIA) Variables Level/ First Diff Aug. Dickey-Fuller (ADF) Test Statistic Phillips-Perron (PP) Test Statistic Without With Without With Conclusion

Trend Trend Trend Trend LNPCY EDU HEALTH SOCIAL DEF Level -08 (1) -2.8 (1) -02-2.32 First -45* (0) -4.57* (0) -40* -49* Level -2.95 (0) -2.83 (0) -2.93-2.83 First -5.71* (0) -5.75* (0) -5.73* -5.85* Level 0.75 (0) -1.36 (0) 00-16 First -4.36* (0) -4.87* (0) -40* -4.86* Level -2.16 (1) -29 (1) -1.83-1.99 First -3.10* (3) -3.75* (0) -3.77* -3.73* Level -2.36 (0) -2.56 (0) -2.36-2.56 First -5.33* (0) -5.17* (0) -56* -53* i) In ADF tests, optimum lag lengths, shown in parentheses in the test statistic column, have been determined using Schwartz Bayesian Criterion (SBC). ii) In PP tests, Bartlett kernel (default) spectral estimation method and Newey-West bandwidth (automatic selection) have been used. iii) Conclusion about the order of integration of a particular variable is based on the test that did not include the trend in the test equation. Test statistics with trend have been shown for the purpose of reporting only. iv) * denotes significant at 5 percent level. Mackinnon (1996) one-sided p-values have been used for this purpose. As all variables in the model are found to be, we conduct Johansen-Juselius cointegration analysis. We specify the relevant order of lags p 2 of the VAR model (implies a lag length of 1 in VEC model) before conducting cointegration tests. Given the nature of the data, which is annual, p 2 seems to be a reasonable choice as we can capture effects of events that occurred up to two years back (this is confirmed by the Schwarz information criterion). Results of the Johansen-Juselius cointegration analysis for the base model (without government deficit variable) have been shown in Table 3. TABLE 3. JOHANSEN-JUSELIUS MAXIMUM LIKELIHOOD COINTEGRATION TESTS (AUSTRALIA, BASE MODEL) Trace Test Maximum Eigenvalue Test Null Alternative Test Null Alternative Test Statistic Statistic r = 0 r > 0 44.98* r = 0 r = 1 23.99 r 1 r > 1 20.99 r = 1 r = 2 10.35 r 2 r > 2 104 r = 2 r = 3 6.39 r 3 r > 3 45 r = 3 r = 4 45 i) r refers to number of cointegrating equations. ii) The test has been conducted assuming linear deterministic trend. iii) * denotes rejection of null hypothesis of no cointegration at 10 percent significance level. MacKinnon- Haug-Michelis (1999) p-values have been used for this purpose.

At 5 percent significance level, the trace test and the maximum eigenvalue test indicate no cointegration among the variables. However, the trace test indicates 1 cointegrating equation at 10 percent significance level. When normalized for a unit coefficient on LNPCY, the cointegrating regression of economic growth in Australia can be given as follows (standard errors in parentheses): 1 LNPCY 97 07EDU 02HEALTH 0. 02SOCIAL (5) (02) (01) (004) In the estimated model above, none of the coefficients of explanatory variables of economic growth is found to be greater than unity, indicating low responsiveness of economic growth to changes in these variables. The coefficient estimates of the variables HEALTH and SOCIAL in the equilibrium relation are significant at 5 percent level and have the expected signs. The coefficient estimate of the variable EDU in the equilibrium relation is significant at 5 percent level with unexpected sign. However it should be mentioned that the correlation between education and growth could be negative because of various factors. First of all, certain types of education such as science and technology, research and innovation could be more growth enhancing than general types of education. Secondly, relatively richer and faster growing states may be in a better position to allocate more funds for educational institutions than lagging states. Finally, education or formal schooling may be seen as sacrificing current earnings for future earnings and thus anticipated growth is likely to boost educational expenditure. Thus the relationship between education and growth reflects the reverse causality as education responds to anticipated growth (Blis and Klenow, 2000). We estimate the error correction model in order to determine the dynamic behavior of economic growth, results of which are displayed in Table 4. TABLE 4. ESTIMATED ERROR CORRECTION MODEL (AUSTRALIA, BASE MODEL) Dependent Variable: ΔLNPCY Regressors Parameter Estimates T-Ratios (absolute value) Intercept 03 46* ΔLNPCY (-1) -0.10 09 ΔLNPCY (-2) -01 17 ΔEDU (-1) -009 0.57 ΔEDU (-2) -02 1.15 ΔHEALTH (-1) -04 1.59 ΔHEALTH (-2) -01 0.38 ΔSOCIAL (-1) -02 36* ΔSOCIAL (-2) -009 1.16 EC (-1) -07 2.71* Note: * denotes significant at 5 percent level. The estimated coefficient of the error term (-07) has been found statistically significant at 5 percent level with appropriate (negative) sign. This suggests that the system corrects its previous period s disequilibrium by 27 percent a year. The cointegrating relationship among the variables suggests existence of Granger causality in at least one direction, but it does not indicate the direction of temporal causality between the variables. In order to determine the direction of causality, we run the Granger causality test within the ECM, which results have been shown in Table 5. TABLE 5. GRANGER CAUSALITY TEST (AUSTRALIA, BASE MODEL) ΔLNPCY Dependent Variable ΔLNPCY ΔEDU ΔHEALTH ΔSOCIAL 1.92 (0.17) 63* (005) 06 (0.53) 0.17 04 10

ΔEDU (0.85) (0.96) (02) ΔHEALTH ΔSOCIAL 1.53 (03) 8.35* (002) 28 (0.10) 19 (05) 1.54 (03) 0.36 (0.70) i) A VAR lag length of 2 has been used in the Granger causality test. ii) Corresponding probabilities have been shown in parentheses. iii) * denotes significant at 5 percent level. It indicates causal relationship. The Granger causality test results indicate unidirectional causality from social expenditure to growth and growth to health expenditure. We found no causal relationship among other variables. These findings are plausible because social expenditure can be seen as an investment, which yield higher per capita income in future. Similarly, higher level of income is expected to result in better allocation of resources, e.g. healthcare. Cointegration Results with Government Deficit in the model When we include government deficit variable in our growth model for Australia, the Johansen-Juselius Maximum Likelihood Cointegration Tests result indicate no cointegration among the variables as shown in Table 6. TABLE 6. JOHANSEN-JUSELIUS MAXIMUM LIKELIHOOD COINTEGRATION TESTS (AUSTRALIA, MODEL 2 WITH DEF) Trace Test Maximum Eigenvalue Test Null Alternative Test Statistic Null Alternative Test Statistic r = 0 r > 0 58.38 r = 0 r = 1 247 r 1 r > 1 33.71 r = 1 r = 2 16.15 r 2 r > 2 17.56 r = 2 r = 3 10.78 r 3 r > 3 6.77 r = 3 r = 4 69 r 4 r > 4 08 r = 4 r = 5 08 i) r refers to number of cointegrating equations. ii) The test has been conducted assuming linear deterministic trend. iii) * denotes rejection of null hypothesis of no cointegration at 5 percent significance level. MacKinnon-Haug-Michelis (1999) p-values have been used for this purpose. Therefore, we can conclude that there is no long-run relationship among the variables in the model. In order to determine the direction of causality, we run the Granger causality test within the ECM, which results have been shown in Table 7. TABLE 7. GRANGER CAUSALITY TEST (AUSTRALIA, MODEL 2 WITH DEF) ΔLNPCY ΔEDU ΔHEALTH ΔSOCIAL ΔDEF Dependent Variable ΔLNPCY ΔEDU ΔHEALTH ΔSOCIAL ΔDEF 1.92 63* 06 3.85* (0.17) (005) (0.53) (03) 0.17 (0.85) 1.53 (03) 8.35* (002) 17 (0.36) 28 (0.10) 19 (05) 6.94* (004) 04 (0.96) 1.54 (03) 13 (06) 10 (02) 0.36 (0.70) 07 (0.52) 009 (0.99) 0.15 (0.86) 89* (002)

i) A VAR lag length of 2 has been used in the Granger causality test. ii) Corresponding probabilities have been shown in parentheses. iii) * denotes significant at 5 percent level. It indicates causal relationship. The Granger causality test results indicate unidirectional causality from growth to health expenditure and government deficit, social expenditure to growth and government deficit, and government deficit to education expenditure. We found no causal relationship among other variables. These findings are plausible because higher economic growth is expected to result in higher level of health expenditure, and this if excessive in amount, could result in government deficit. Similarly, higher level of social expenditure is expected to generate growth in future, but at present time it can create government deficit. Finally, excessive spending generating government deficit can be seen as an investment in human capital (education). Finally we perform diagnostic tests using correlogram of the residuals, which indicate presence of no serial correlation at 5 percent significance level. Figure 1 displays the diagnostic test results. Autocorrelations w ith 2 Std.Err. Bounds Cor(LNPCY,LNPCY(-i)) Cor(LNPCY,EDU(-i)) Cor(LNPCY,HEALTH(-i)) Cor(LNPCY,SOCIAL(-i)) - - - - Cor(EDU,LNPCY(-i)) Cor(EDU,EDU(-i)) Cor(EDU,HEALTH(-i)) Cor(EDU,SOCIAL(-i)) - - - - Cor(HEALTH,LNPCY(-i)) Cor(HEALTH,EDU(-i)) Cor(HEALTH,HEALTH(-i)) Cor(HEALTH,SOCIAL(-i)) - - - - Cor(SOCIAL,LNPCY(-i)) Cor(SOCIAL,EDU(-i)) Cor(SOCIAL,HEALTH(-i)) Cor(SOCIAL,SOCIAL(-i)) - - - - Figure 1: Diagnostic Test of the Residuals (Australia, Base Model) New Zealand In order to analyze time-series properties of the data, we conduct Augmented Dickey-Fuller (ADF) and Phillips- Perron (PP) tests at both level and first difference for all variables in the model. Results of the unit root tests have been shown in Table 8. TABLE 8. UNIT ROOT TESTS FOR STATIONARITY (NEW ZEALAND) Variables Level/ First Diff Aug. Dickey-Fuller (ADF) Test Statistic Phillips-Perron (PP) Test Statistic Conclusion

Without Trend With Trend Without Trend With Trend LNPCY EDU HEALTH SOCIAL DEF Level -06 (0) -1.93 (1) -09-1.79 First -44* (0) -4.15* (0) -43* -4.14* Level -0.58 (0) -1.70 (0) -09-1.98 First -48* (0) -42* (0) -47* -42* Level 0.75 (0) -1.36 (0) 00-16 First -4.36* (0) -4.87* (0) -41* -4.86* Level -2.16 (1) -29 (1) -1.83-1.99 First -3.10* (3) -3.75* (0) -3.77* -3.73* Level -2.36 (0) -2.56 (0) -2.36-2.56 First -5.33* (0) -5.17* (0) -56* -53* i) In ADF tests, optimum lag lengths, shown in parentheses in the test statistic column, have been determined using Schwartz Bayesian Criterion (SBC). ii) In PP tests, Bartlett kernel (default) spectral estimation method and Newey-West bandwidth (automatic selection) have been used. iii) Conclusion about the order of integration of a particular variable is based on the test that did not include the trend in the test equation. Test statistics with trend have been shown for the purpose of reporting only. iv) * denotes significant at 5 percent level. Mackinnon (1996) one-sided p-values have been used for this purpose. As all variables in the model are found to be, we conduct Johansen-Juselius cointegration analysis. We specify the relevant order of lags p 2 of the VAR model (implies a lag length of 1 in VEC model) before conducting cointegration tests. Given the nature of the data, which is annual, p 2 seems to be a reasonable choice as we can capture effects of events that occurred up to two years back (this is confirmed by the Schwarz information criterion). Results of the Johansen-Juselius cointegration analysis for the base model (without government deficit variable) have been shown in Table 9. TABLE 9. JOHANSEN-JUSELIUS MAXIMUM LIKELIHOOD COINTEGRATION TESTS (NEW ZEALAND, BASE MODEL) Trace Test Maximum Eigenvalue Test Null Alternative Test Null Alternative Test Statistic Statistic r = 0 r > 0 56.35* r = 0 r = 1 25.56* r 1 r > 1 30.80* r = 1 r = 2 210* r 2 r > 2 9.80 r = 2 r = 3 9.76 r 3 r > 3 04 r = 3 r = 4 04 i) r refers to number of cointegrating equations. ii) The test has been conducted assuming linear deterministic trend. iii) * denotes rejection of null hypothesis of no cointegration at 10 percent significance level. MacKinnon- Haug-Michelis (1999) p-values have been used for this purpose. Both the trace test and maximum eigenvalue test indicate 2 cointegrating equations at 10 percent significance level. When normalized for a unit coefficient on LNPCY and EDU, the cointegrating regression of

economic growth and education expenditure in New Zealand can be given as follows (standard errors in parentheses): 2 LNPCY 8.14 0.13HEALTH 0. 09SOCIAL (6) (03) (02) EDU 477 2.13HEALTH 3. 02SOCIAL (7) (09) (04) In the estimated model (equation 6) above, none of the coefficients of explanatory variables of economic growth is found to be greater than unity, indicating low responsiveness of economic growth to changes in these variables. The coefficient estimates of the variables HEALTH and SOCIAL in the equilibrium relation are significant at 5 percent level and have the expected signs. Thus, health and social expenditures are found to be the main determinants of economic growth. In the estimated model (equation 7) above, the coefficients of explanatory variables of education expenditure are found to be greater than unity, indicating high responsiveness of education expenditure to changes in these variables. The coefficient estimates of the variables HEALTH and SOCIAL in the equilibrium relation are significant at 5 percent level and have the expected signs. Thus, health and social expenditures are found to be the main determinants of education expenditure. We estimate the error correction model in order to determine the dynamic behavior of economic growth, results of which are displayed in Table 10. TABLE 10. ESTIMATED ERROR CORRECTION MODEL (NEW ZEALAND, BASE MODEL) Dependent Variable: ΔLNPCY Regressors Parameter Estimates T-Ratios (absolute value) Intercept 01 1.89 ΔLNPCY (-1) -08 08 ΔLNPCY (-2) -0.10 0.34 ΔEDU (-1) -01 0.58 ΔEDU (-2) 0003 02 ΔHEALTH (-1) 005 05 ΔHEALTH (-2) 008 0.36 ΔSOCIAL (-1) -006 0.76 ΔSOCIAL (-2) -008 0.95 EC (-1) -03 2.13* Note: * denotes significant at 5 percent level. The estimated coefficient of the error term (-03) has been found statistically significant at 5 percent level with appropriate (negative) sign. This suggests that the system corrects its previous period s disequilibrium by 23 percent a year. In order to determine the direction of causality, we run the Granger causality test within the ECM, which results have been shown in Table 11. TABLE 11. GRANGER CAUSALITY TEST (NEW ZEALAND, BASE MODEL) ΔLNPCY ΔEDU ΔHEALTH Dependent Variable ΔLNPCY ΔEDU ΔHEALTH ΔSOCIAL 3.54* (04) 0.30 (0.74) 3.30 (05) 5.92* (008) 74* (003) 07 (0.93) 11 (06) 3.14 (06) 1.78 (0.19)

ΔSOCIAL Proceedings of the Australian Academy of Business and Social Sciences Conference 2014 1.83 (0.18) 1.75 (0.19) 03 (05) i) A VAR lag length of 2 has been used in the Granger causality test. ii) Corresponding probabilities have been shown in parentheses. iii) * denotes significant at 5 percent level. It indicates causal relationship. The Granger causality test results indicate unidirectional causality from education expenditure to growth, growth to health expenditure, and health expenditure to education expenditure. We found no causal relationship among other variables. These findings are plausible because higher level of investment in education is expected to yield higher per capita income in future. Similarly, higher level of income is expected to result in better allocation of resources in healthcare, and higher level of health expenditure is expected to raise education expenditure (as investment in human capital) through increased productivity channel. Cointegration Results with Government Deficit in the model When we include government deficit variable in our growth model for New Zealand, the Johansen-Juselius Maximum Likelihood Cointegration Tests results based on Trace test (2 cointegrating equations) and Maximumeigen value test (no cointegration) give contradictory results as shown in Table 12. However, the sign of the error correction term in the VECM (vector error correction model) is found to be insignificant, indicating no cointegration among the variables. TABLE 12. JOHANSEN-JUSELIUS MAXIMUM LIKELIHOOD COINTEGRATION TESTS (NEW ZEALAND, MODEL 2 WITH DEF) Trace Test Maximum Eigenvalue Test Null Alternative Test Statistic Null Alternative Test Statistic r = 0 r > 0 861* r = 0 r = 1 33.72 r 1 r > 1 529* r = 1 r = 2 23.73 r 2 r > 2 28.95 r = 2 r = 3 141 r 3 r > 3 14.94 r = 3 r = 4 112 r 4 r > 4 3.92 r = 4 r = 5 3.92 i) r refers to number of cointegrating equations. ii) The test has been conducted assuming linear deterministic trend. iii) * denotes rejection of null hypothesis of no cointegration at 5 percent significance level. MacKinnon-Haug-Michelis (1999) p-values have been used for this purpose. Therefore, we can conclude that there is no long-run relationship among the variables in the model. In order to determine the direction of causality, we run the Granger causality test within the ECM, which results have been shown in Table 13.

ΔLNPCY ΔEDU ΔHEALTH ΔSOCIAL ΔDEF TABLE 13. GRANGER CAUSALITY TEST (NEW ZEALAND, MODEL 2 WITH DEF) Dependent Variable ΔLNPCY ΔEDU ΔHEALTH ΔSOCIAL ΔDEF 3.54* (04) 0.30 (0.74) 1.83 (0.18) 07 (0.93) 3.30 (053) 5.92* (008) 1.75 (0.19) 00 (0.82) 74* (003) 07 (0.93) 03 (05) 2.57 (0.10) 11 (06) 3.14 (06) 1.78 (0.19) 2.13 (0.14) i) A VAR lag length of 2 has been used in the Granger causality test. ii) Corresponding probabilities have been shown in parentheses. iii) * denotes significant at 5 percent level. It indicates causal relationship. 1.76 (0.19) 3.87* (03) 0.17 (0.85) 0.16 (0.86) The Granger causality test results indicate unidirectional causality from growth to health expenditure, health expenditure to education expenditure, and education expenditure to growth and government deficit. We found no causal relationship among other variables. These findings are plausible because higher economic growth is expected to result in higher level of health expenditure. Higher level of health expenditure, on the other hand, may result in higher level of investment in human capital due to expected productivity increase. Similarly, higher level of education expenditure is expected to generate growth in future, but at present time it can create government deficit. Finally we perform diagnostic tests using correlogram of the residuals, which indicate presence of no serial correlation at 5 percent significance level. Figure 2 displays the diagnostic test results.

Autocorrelations w ith 2 Std.Err. Bounds Cor(LNPCY,LNPCY(-i)) Cor(LNPCY,EDU(-i)) Cor(LNPCY,HEALTH(-i)) Cor(LNPCY,SOCIAL(-i)) - - - - Cor(EDU,LNPCY(-i)) Cor(EDU,EDU(-i)) Cor(EDU,HEALTH(-i)) Cor(EDU,SOCIAL(-i)) - - - - Cor(HEALTH,LNPCY(-i)) Cor(HEALTH,EDU(-i)) Cor(HEALTH,HEALTH(-i)) Cor(HEALTH,SOCIAL(-i)) - - - - Cor(SOCIAL,LNPCY(-i)) Cor(SOCIAL,EDU(-i)) Cor(SOCIAL,HEALTH(-i)) Cor(SOCIAL,SOCIAL(-i)) - - - - Figure 2: Diagnostic Test of the Residuals (New Zealand, Base Model) POLICY IMPLICATIONS First of all, our study found coefficients of the health and social expenditures in the growth model to be positive and significant, therefore they are the main determinants of growth in Australia and New Zealand. We also found educational expenditure negatively affecting growth in Australia in the long-run, which could be caused by factors often ignored in the traditional literature highlighting the positive effects of education through productivity enhancement. Education is viewed by most as sacrifice of current earnings for the sake of higher future earnings and thus educational spending is likely to get a boost from anticipated growth. Also certain types of education (e.g. science and technology) may be more growth enhancing than others (like general arts). Interestingly, when we included government deficit in the model, we could not find any long run relationship among growth, education expenditure, health expenditure, social expenditure, and government deficit. This implies that excessive spending in social sectors might have forced the governments to cut expenditure on many other productive sectors for fear of further deficit widening, and thus positive effects of social spending on growth might have been neutralized by negative effects of other sectors. Second, the coefficients of error correction terms in the vector error correction model (VECM) are found to be negative and significant, confirming that there exists a long-run relationship among the variables in the growth model that does not include government deficit. Social expenditure is also found to have negatively impacting economic growth in Australia in the short-run. Third, the Granger causality test results indicate one-way causality running from economic growth to health expenditure, social and education expenditures to growth, and health expenditure to education expenditure in both countries. However, if we include government deficit in the model, the results mostly remain same as before (base

model), except government deficit is found to be caused (one-way) by growth and social expenditure in case of Australia, and by education expenditure in case of New Zealand. From the policy point of view, our findings have important ramifications. The long-run positive link from health and social welfare expenditure to economic growth in Australia and New Zealand imply that developing countries, especially those experiencing rapid growth should adopt a policy of increasing their budget share in social welfare significantly as a prescription for sustained economic growth. The well-known maxim healthier is wealthier fits well in this context and the fast developing countries are also experiencing demographic features such as problem of ageing and increasing dependency ratio that may necessitate increasing budget for health and social welfare. Also, based on Granger causality test results, we can conclude that developing countries will gain significantly in the short-run from investment in social welfare due to two-way causality running to and from social welfare expenditures and economic growth. This is particularly important for ASEAN countries that do not allocate sizeable portion of their budget for social welfare and are seemingly opposed to the idea of social welfare. Though we are not generalizing on the experience of welfare countries, we would like to emphasis that there is nothing inherently wrong in higher public expenditure on social sectors as long as they are not at the expense of other highly productive sectors and are carried out most efficiently. The fast developing economies of the ASEAN, namely Singapore and Malaysia, which aim to achieve the developed country status by 2020 should adopt a more generous social policy in order to ensure a sustainable balanced development. CONCLUSION In this paper, we have examined the effects of social expenditures on economic growth in Australia and New Zealand by means of cointegration, error correction methods, and Granger causality test using annual data for a period of 33 years (1980-2012). Our study shows that long-run economic growth in Australia is largely explained by education, health and social expenditures; in case of New Zealand, on the other hand, economic growth is largely explained by health and social expenditures. It is also evident that there is a bi-directional relationship between social expenditures and economic growth, especially when the government allocates excessive budget for social welfare creating deficit. The estimated coefficients of the ECM indicate a moderate speed of adjustment to equilibrium. The sign of error correction term is negative and significant, confirming that there exists a long-run equilibrium relationship among the variables. Our findings suggest that in the long-run social expenditures impact economic growth positively. In the short-run also there is a two-way relationship between social expenditures and economic growth. These results tend to suggest that developing countries should not neglect social welfare since there is a positive link between social welfare expenditures and economic growth, and also social safety net is almost non-existent in those countries. Fast growing ASEAN nations, especially Singapore and Malaysia that usually rely on an extremely modest budget for social welfare, but at the same time are keen to achieve the status of developed nations by 2020, should adopt a more generous policy stance towards social welfare. ENDNOTES 1, 2 The standard errors for the cointegrating vector are computed following Boswijk (1995). REFERENCES Australian Bureau of Statistics, various publications. Barro, R., A Cross-country Study of Growth, Saving, and Government, NBER Working Paper No. 2855, 1989, National Bureau of Economic Research. Baum, N. D. and Lin, S., The Differential Effects on Economic Growth of Government Expenditures on Education, Welfare and Defense, Journal of Economic Development, 1993, Vol. 18, pp.175-185. Bellettini, G. and Ceroni, C. B., Social Security Expenditure and Economic Growth: An Empirical Assessment, Research in Economics, 2000, Vol. 54, pp49-275. Beraldo, S., Montolio, D. and Turati, G., Healthy, Educated and Wealthy: A Primer on the Impact of Public and Private Welfare Expenditures on Economic Growth, The Journal of Socio-Economics, 2009, Vol. 38, pp.946-956. Blis, M. and Klenow, P. J., Does Schooling Cause Growth? The American Economic Review, December 2000, pp 1160-1183. Boswijk, H. P., Identifiability of Cointegrated Systems, Technical Report, 1995, Tinbergen Institute.

Dickey, D. A. and Fuller, W. A., Likelihood Ratio Statistics for Autoregressive Time-Series with a Unit Root, Econometrica, 1981, Vol. 49, pp.1057-1072. Fic, T. and Ghate, C., The Welfare State, Thresholds, and Economic Growth, Economic Modelling, 2005, Vol. 22, pp.571-598. Granger, C. W. J., Some Recent Developments in a Concept of Causality, Journal of Econometrics, 1988, Vol. 39, pp.199-211. ILO, ILO Online database. International Monetary Fund, International Financial Statistics Online. Johansen, S. and Juselius, K., Maximum Likelihood Estimation and Inference on Cointegration with Applications to the Demand for Money, Oxford Bulletin of Economics and Statistics, 1990, Vol. 52, pp.169-210. Johansen, S. and Juselius, K., Testing Structural in a Multivariate Cointegration Analysis of the PPP and the UIP for UK, Journal of Econometrics, 1992, Vol. 53, pp11-244. Johansen, S. and Juselius, K., Identification of the Long-Run and the Short-Run Structure: An Application to the ISLM Model, Journal of Econometrics, 1994, Vol. 63, pp.7-36. Kwiatkowski, D., Phillips, P. C. B., Schmidt, P., and Shin, Y., Testing the Null of Stationary against the Alternative of a Unit Root, Journal of Econometrics, 1992, Vol. 54, pp.159-178 Landau, D., Government Expenditure and Economic Growth: A Cross-country Study, Southern Economic Journal, 1983, Vol. 49, pp.783-792. MacKinnon, J. G., Numerical Distribution Functions for Unit Root and Cointegration Tests, Journal of Applied Econometrics, 1996, Vol. 11, pp01-618. MacKinnon, J. G., Haug, A. A., and Michelis, L., Numerical Distribution Functions of Likelihood Ratio Tests for Cointegration, Journal of Applied Econometrics, 1999, Vol. 14, pp.563-577. McDonald, B. D. and Miller, D. R., Welfare Programs and the State Economy, Journal of Policy Modeling, 2010, Vol. 32, pp.719-732. Mercan, M., and Sezer, S., The Effect of Education Expenditure on Economic Growth: The Case of Turkey, Procedia- Social and Behavioral Sciences, 2014, Vol. 109, pp.925-930. OECD, OECD Factbook, 2014: Economic, Environmental and Social Statistics, OECD Publishing. OECD, OECD ilibrary. New Zealand Treasury, various publications. Phillips, P. C. B. and Perron, P., Testing for a Unit Root in Time Series Regression, Biometrica, 1988, Vol. 75, pp.335-346. Psacharopoulos, G. Education for Development, 1985, Washington, D. C.: The World Bank. Rosenzweig, M. R., When Investment in Education Matters and When It Does Not, Challenge, 1996, Vol. 39, pp5-27. Schultz, T. W., Investment in Human Capital, American Economic Review, 1961, Vol. 51, pp.1-17. UNESCO, UNESCO Institute for Statistics Online Database. World Bank, World Development Indicators Online. APPENDIX TABLE 1. SOCIAL WELFARE EXPENDITURES (% OF GDP) IN OECD COUNTRIES

Country Proceedings of the Australian Academy of Business and Social Sciences Conference 2014 Average GDP per capita 2009-2012 (Constant 2005 US$) Social welfare expenditures (% of GDP) 2009 2010 2011 2012 2013 Australia 36,550 17.8 17.9 18 18.8 19.5 Austria 39,232 29.1 28.9 27.9 27.9 28.3 Belgium 36,604 29.7 29.5 29.7 30.5 30.7 Canada 35538 19 18.7 18.1 18.1 18 Chile 8,826 11.3 10.8 10 10.. Czech Republic 14,171 20.7 20.8 20.8 21 21.8 Denmark 46,252 30 30 30 30.8 30.8 Estonia 10,955 20 20.1 18 17 17.7 Finland 38,100 29 29 29 30 30.5 France 34,009 32.1 32 32 32.5 33 Germany 36,587 27.8 27.1 25.9 25.9 26 Greece 20,698 23.9 23.3 24 24.1 22 Hungary --- 23.9 22.9 21.9 21 21 Iceland 52,783 18.5 18 18.1 17 17 Ireland 45,159 23 23.7 23.3 22 21 Israel 22,347 16 16 15.8 15.8 15.8 Italy 29,042 27.8 27.7 27.5 28 28 Japan 36,100 22 22.3...... Korea 20,726 9 9 9.1 9.3.. Luxembourg 79,388 23 23 22 23 23 Mexico --- 8 8.1 7.7 7.. Netherlands 40,939 23 23 23 24 24.3 New Zealand 27,818 21 21.3 21 22 22 Norway 64,977 23.3 23 22 22.3 22.9 Poland 10,163 21.5 21.8 20.5 20 20.9 Portugal 18,328 25 25 25 25 26 Slovak Republic 14,388 18.7 19.1 18.1 18.3 17.9 Slovenia 18,927 22 23 23.7 23.7 23.8 Spain 25,215 26 26.7 26 26.8 27 Sweden 42,735 29.8 28.3 27 28.1 28 Switzerland 54,569.. 20 19.5 18.8 19.1 Turkey --- 12.8........ United Kingdom 37,561 24.1 23.8 23 23.9 23.8 United States 44,240 19 19.8 19 19.7 20 OECD - Total --- 22.1 22.1 21.7 21.8 21.9 Source: 1) OECD ilibrary, http://www.oecd-ilibrary.org/, accessed 15 June 2014; and 2) World Development Indicators Online, accessed 29 June 2014.