Savings Investment Correlation in Developing Countries: A Challenge to the Coakley-Rocha Findings Abu N.M. Wahid Tennessee State University Abdullah M. Noman University of New Orleans Mohammad Salahuddin* Assistant Professor Southeast University Dhaka, Bangladesh Email: salahuddin0000@gmail.com Cell: 0088-01732843499 *Corresponding author Abstract- In contrast to Coakley Rocha findings, this paper discovers another puzzle such assmaller savings-investment correlation for developing countries than for developed ones. Furthermore, this doesn t disappear even when heterogeneity and cross sectional dependence are explicitly taken into account in a panel of developing countries. JEL Classifications: C23, F21, F32 Keywords- Capital mobility, Coakley, Cross section dependence, Feldstein Horioka puzzle, heterogeneity, Rocha, Saving-investment correlation coefficients for developing countries have been surprisingly lower (Dooley et. al 1987, Kesuga 2004) and the authors attempt to provide explanation for this poor correlation by disregarding the view that it reflects high capital mobility. Rocha (2005) claims that this puzzling correlation between savings and investment in developing countries disappears when country specific heterogeneity is taken into account. Contrary to their claim, this paper shows that the puzzle (lower correlation between savings and investment) does not disappear even when heterogeneity and cross section dependence are explicitly taken into account. I. INTRODUCTION Feldstein-Horioka (1980) puzzle (F H puzzle, henceforth) about the relationship between savings and investment is criticized for its failure to consider country specific heterogeneity and cross section dependence. Authors such as Mark et al. (2003), Coakley et al (2004) and Rocha (2005) addressed these crucial issues. Much of the literature analyzes savings investment correlation from the perspectives of developed economies while only a few studies focus on developing economies. The estimated II. A BRIEF METHODOLOGICAL DISCUSSION Let I t be the (domestic) investment GDP ratio of a country at time t, and S t be the (domestic) savings GDP ratio, and the relationship between the two be represented by Equation (1). Under free capital mobility, as is true in the developed countries, the correlation between these two is expected I t = α +βs t +u t, u t ~ iid (0, σ 2 ) (1) to be very poor, hence, β should approach zero. On the other hand, if the international 89.1
Abu N.M. Wahid, Abdullah M. Noman, and Mohammad Salahuddin capital mobility is low, as in the developing countries, the coefficient of the only regressor in the above equation would equal unity implying a one for one correspondence between domestic savings and investment. If the variables of the above equation are stationary, then one can apply OLS to estimate the slope coefficient. However, if the variables of the model are nonstationary and there exists no cointegrating relationship between them (i.e. the residuals must be stationary), then estimating Equation (1) using OLS would result in spurious regression. Under such circumstances, Phillips and Moon (1999), and Kao (1999) suggest that panel data be used as a way to overcome the spuriousity to estimate the said equation. In particular, they establish that one can obtain consistent estimates of long run parameters by gathering the mean effect. In this paper, we use the Mean-Group (or MG, for short) approach of Pesaran and Smith (1995) because of its ability to deal with I(1) regressors as well as innovations. The paper investigates the savings investment correlation in a world panel of 20 developing countries using the following model that allows heterogeneity both in the intercept and slope coefficients I it = α i +β i S it + u it, u it ~ iid (0, σ 2 ) (2) where, i = 1..20 and the rest of notations are as in Equation (1). There are certain assumptions that underlie the MG approach. These assumptions are discussed in Pesaran and Smith (1995) and Coakley et al. (2004, 2005). O Connell (1998) warns that panel data may suffer from cross-sectional dependence (CSD). Therefore, we use another estimator that is designed to take CSD into account. This estimator is known as the cross sectionally augmented MG (or CMG, for short) proposed in Pesaran (2004b) and would seize the unobserved common variables or shocks that may induce the cross-sectional dependence in the data. The basic framework remains the same but the underlying equations are now augmented with cross sectional averages. In addition to using a different estimator that accounts for CSD, we also conduct a diagnostic test, namely, the PCD test (Pesaran, 2004b) for the presence of CSD in the residuals of MG regressions. Data III. DATA AND RESULTS The sample includes annual observations of investment-gdp and savings-gdp ratios for a panel of 20 developing countries: Bangladesh, Indonesia, Kenya, Lesotho, Niger, Togo, Zambia, Bolivia, China, Colombia, Domonican Republic, Egypt, Peru, South Africa, Sri Lanka, Swaziland, Turkey, Hungary, Oman, and Uruguay. The data have been gathered from the World Bank Development Indicators (WDI 2007) database. Results and Findings Table 1 presents two unit root test results, namely, the ADF and KPSS tests. The test equations include both a constant a time trend. As for the ADF test, one can reject the null hypothesis in only 13 out of 40 cases. On the other hand, the KPSS test rejects the null hypothesis of stationarity in 20 out of 40. Together, these results indicate that the variables at hand, i.e. I t and S t, are mostly indistinguishable from I(1) process. The last column of Table 1 presents the augmented Angle Granger (AEG) test of cointegration. Again in most of the cases, one cannot reject the null of nonstationarity in the residuals series obtained from OLS regression. This finding justifies the use of the MG panel regression of Pesaran and Smith (1995) which is presumably capable of estimating long run relationship among variables in the presence of nonstationary regressors that do not cointegrate. Special Issue of the International Journal of the Computer, the Internet and Management, Vol. 19 No. SP1, June, 2011 89.2
Savings Investment Correlation in Developing Countries: A Challenge to the Coakley-Rocha Findings TABLE 1 UNIT ROOT TEST RESULTS Notes: The optimal lag length for ADF test is selected using testing down method from a maximum lag of 8 and for the KPSS test, the lag truncation parameter equals the integer part of 4(T/100 1/4 ).Critical values for the ADF and AEG tests are based on MacKinnon (1996) and for the KPSS tests on Kwiatkowski et al. (1992). * (**) denotes rejection of the null at 5% (10%) level. Table 2 reports the estimation results. The slope coefficient of the savings investment regression within the MG framework is approximately 0.6. The estimated value of the coefficient is far smaller than the expected value of unity in the developing countries. We test two hypotheses, namely, β = 0 (i.e. perfect capital mobility) and β = 1 (i.e. no capital mobility). The associated t statistics reject both the null hypothesis of unit and zero coefficients at 1% level. These findings, as also in Rocha (2005), are consistent with the fact that the developing world has achieved some degree of capital mobility, even if not perfect, in the recent past following deregulation and liberalization policies. The PCD test result for the presence of cross section dependence rejects the null of no CSD at 1% level of significance. Therefore, we apply CMG estimator that accounts for CSD in the data. TABLE 2 SAVING INVESTMENT CORRELATION IN A DEVELOPING PANEL 1973 2002 Notes: MG stands for mean group estimation while CMG for cross sectional augmented MG estimation and CS is the between of cross section estimator.* Rejects the null of no correlation at 1%.. The associated p values for the JB test are in brackets. 89.3
Abu N.M. Wahid, Abdullah M. Noman, and Mohammad Salahuddin The results of CMG estimation are also presented in Table 2. The estimated value of the coefficient gets even smaller with CMG procedure, which is only 0.45 indicating only a poor relationship between domestic savings and investment. The associated t statistic again rejects the null hypothesis of both zero and unit savings investment coefficients. Overall these findings are more supportive of the F H puzzle than challenging it. The between or cross sectional estimates of the panel coefficients are also reported in the Table which are very much surprising. The slope coefficient is negative and insignificant. While on prima facie basis, it is difficult to explain this result, it may be reinforcing the findings of other estimators that the savings investment relationship is poor in the panel. TABLE 3 SAVING INVESTMENT CORRELATION: A COMPARISON Source: Coakley et al. (2004) and Rocha (2005) In order to ensure that inferences of the paper are valid, we follow Coakley et al. (2004) and test for normality of the sample individual estimates of both MG and CMG procedures. The Jarque Bera (JB) test values indicate that the null hypothesis of normality cannot be rejected at any significance level. This validates the small sample inferences in this paper based on Student t distribution 1. Table 3 makes a comparison of the findings of this paper with that two other most relevant studies, namely, (Coakley et al. 2004) and (Rocha, 2005). While the former investigates the F H puzzle in an OECD panel, the latter looks into a developing panel. Both of the studies employ the MG and CMG estimators. The studies show that the F H puzzle disappears when heterogeneity and cross section dependence are explicitly modeled. The findings of the present study, however, seem to disagree with the conclusions made in those two studies. Especially, both the MG and CMG estimates in this study are lower than those of Rocha (2005) implying that savings investment relationship is rather low which is contrary to expectation given the higher degree of control and regulation imposed in the financial markets in general and in the capital markets in particular. Also, the findings of this paper are based on more recent data than those of Rocha (2005). Overall, the findings of this study contradict the claim of (Coakley et al. 2004) and (Rocha, 2005) that the F H puzzle disappears when heterogeneity and cross sectional dependence are explicitly modeled in a panel setting. IV. CONCLUSIONS This paper does not support the claim of earlier findings indicating disappearance of the F H puzzle of lower correlation between savings and investment in developing countries in the presence of heterogeneity and cross sectional dependence. Both the hypotheses of perfect capital mobility and perfect capital immobility are rejected implying intermediate degree of capital mobility among these countries. This finding sounds plausible in the sense that the developing countries are moving fast toward free trade. The volume of merchandized trade and mobility of capital flows have been Special Issue of the International Journal of the Computer, the Internet and Management, Vol. 19 No. SP1, June, 2011 89.4
Savings Investment Correlation in Developing Countries: A Challenge to the Coakley-Rocha Findings burgeoning over the recent years. Both MG and CMG estimators provide with lower values for savings investment correlation indicating the presence of the puzzle of low correlation between savings and investment even in the presence of heterogeneity and cross sectional dependence. REFERENCES [1] Coakley, J. Fuertes, A M, Spagnolo, F. (2004) Is the Feldstein Horioka Puzzle History? The Manchester School, Vol. 72 No. 5, pp. 569-590 [2] Coakley, J., Flood, R., Fuertes, A. and Taylor, M. (2005). Purchasing power parity and the theory of general relativity: the first tests, Journal of International Money and Finance, Vol. 24, pp. 293 316. [3] Dooley, M., Frankel, J. and Mathison, D. J. ( 1987). International Capital Mobility: What Do Saving-Investment Correlations Tell US? IMF Staff Papers, 34, 503-29. [4] Kao, C. (1999). Spurious regression and residual-based tests for cointegration in panel data, Journal of Econometrics, Vol. 90, pp. 1 44. [5] Kwiatkowski, D., Phillips, P. C. B., Schmidt, P. and Shin, Y. (1992) Testing the Null of Stationarity Against the Alternative of a Unit Root: How Sure Are We That Economic Time Series Have a Unit Root?, Journal of Econometrics, 54, pp. 159 78. [6] MacKinnon, J. G. (1996) Numerical Distribution Functions for Unit Root and Cointegration Tests, Journal of Applied Econometrics, 11, pp. 601 18. [7] Mark, N.C., Ogaki, M. and Sul, D.(2003). Dynamic Seemingly Unrelated Cointegrating Regression NBER Working Paper T0292 [8] Pesaran, M. and Smith, R. (1995). Estimating long-run relationships from dynamic heterogeneous panels, Journal of Econometrics, Vol. 68, pp.79 113. [9] Pesaran, M. H. (2004a). Estimation and inference in large heterogeneous panels with a multifactor error structure, Cambridge Working Paper in Economics No. 0305. University of Cambridge. [10] Pesaran, M. H. (2004b). General diagnostic tests for cross section dependence in panels, Cambridge Working Paper in Economics No. 0435. University of Cambridge. [11] Phillips, P. and Moon, H. (1999). Linear regression theory for non-stationary panel data, Econometrica, Vol. 67, pp.1057 112. [12] Rocha, F. (2005), Heterogeneity, cross section dependence and capital mobility in developing countries, Economics Letters, Vol. 89, Issue 1, pp. 18-23 [13] World Development Indicators (WDI, 2007) database, CDROM, World Bank, Washington DC. 1 The Kolmogorov Smirnov (KS) test of normality produces similar results. 89.5