Dynamic panel data model estimation results for equation 3 in model 1 are presented in appendices 1 through 6 and are reported in table 5.1. This includes four separate results in columns 1 to 4 of the table. Column 1 comprises the OLS estimated results, column 2 has the one-step system GMM results and column 3 includes the two-step system GMM results. Column 4 is made up of the results for the LSDV estimated model. An underlying advantage of the dynamic system GMM estimation is that all variables from the regression that are not correlated with the error term (including lagged and differenced variables) can be potentially used as valid instruments (Greene, 2008). Optimal set of internal instruments were utilized by engaging the collapse option in the system GMM results. All estimations are robust to heteroskedasticity or autocorrelation. This is irrespective of whether they are considered under OLS, system GMM, or LSDV.
In this model specification, lagged YGR and WR are predetermined and endogenous variables respectively. Hence, I control for the endogeneity of these variables in its lagged form as regressors by using internal instruments;
namely, lagged levels of the standard differenced equation (equation 5) and lagged differences of the levels equation (equation 3). The list of these internal instruments can be found in appendices 1 and 3. As an additional check of potential endogeneity problems I investigate the correlation coefficients (see
115 appendix 2) between residuals from the base regression and independent variables. The coefficients of correlations suggest that none of the independent variables is highly correlated with predicted residuals.
116 Table 5.1: Estimated Empirical Results in the Output-Remittances Model (Model 1) Dependent Variable: YGR
OLS SYSTEM-GMM LSDV
One-Step Two-Step
Instrument Weight collapsed collapsed
Regressors (1) (2) (3) (4)
YGR(-1) 0.33958** 0.26652** 0.26529* 0.03141
(0.136) (0.105) (.064) (0.129)
Log(labour) 0.64111*** 0.67894*** 0.74795* -9.40942
(0.361) (0.362) (0.199) (19.829)
Log(capital) -1.68758*** -2.04399 -1.87093*** -0.81266
(0.911) (1.352) (0.983) (1.643)
REER -0.00026 1.12000 0.0003 0.00389
(0.002) (0.001) (0.001) (0.003)
INF -0.09096** -0.10451** -0.10773* -0.15324*
(0.035) (0.040) (0.037) (0.041)
INV 0.00111** 0.00146*** 0.0015** 0.00117
(0.001) (0.0007) (0.001) (0.001)
WR -0.00026 -0.00001 -0.00112 0.00034
(0.002) (0.002) (0.001) (0.003)
Constant 13.162 15.41628 14.2918** 19.06527
(5.678) (7.948) (5.746) (22.631)
Time Dummy Yes Yes Yes Yes
Country Dummy No No No Yes
Observations 133 133 133 133
No. of countries 19 19 19 19
Instrument count - 18 18 -
F-stat (Wald χ2 ) 4.24 43.10 158.89 3.45
F-stat (p-value) [0.0000] [0.0000] [0.0000] [0.0002]
AR(2) - [0.394] [0.411] -
AR(3) - [0.231] []0.220 -
Sargan Test (OIR) - [0.930] [0.930] -
Hansen Test (OIR) - [0.870] [0.870] -
Notes: Robust standard errors, consistent in the presence of any pattern of heteroskedasticity and autocorrelation within panels are reported in curly brackets.
Robust standard errors are with Windmeijer (2005) finite-sample correction for the two-step covariance matrix
P-values are reported in square brackets * indicates significant at 1 percent level ** indicates significant at 5 percent level *** indicates significant at 10 percent level
117 An examination of results in Table 5.1 begins with some specification or diagnostic tests. As a starting point, the system GMM estimators assume that the idiosyncratic errors are serially uncorrelated for consistent estimations.
The presence of autocorrelation will indicate that lags of the dependent variable (and any other variables used as instruments that are not strictly exogenous), are in fact endogenous, thus bad instruments. Arellano and Bond develop a test for this phenomenon that would potentially render some lags invalid as instruments. Of course, the full disturbance is presumed autocorrelated because it contains fixed effects, and the estimators are designed to eliminate this source of trouble.
The Arellano-Bond test for autocorrelation is applied to the differenced residuals in order to purge the unobserved and perfectly autocorrelated idiosyncratic errors. These results are reported as AR(2) and AR(3) in the lower portion of table 5.1. The null hypothesis here that
for k = 1, 2 and 3 is rejected at a level of 0.05 if . If are serially uncorrelated, then the null of no serial correlation will be rejected at order 1 but not at higher orders. This indeed is the case with results in columns 2 and 3. Here, it can be concluded that there is no evidence of serial correlation at the five percent level of significance. Given this results, the estimates can be regarded as consistent.
The next specification test is a test of overidentifying restrictions of whether the instruments, as a group, appear exogenous. This test of instrument validity has to do with a comparism of the number of instruments used in each case and the related number of parameters. It is implemented by the Sargan and Hansen J tests. For one-step, non-robust estimation, the Sargan statistic which is the minimized value of the one-step GMM criterion function, is applicable. The Sargan statistic in this case is however not robust to autocorrelation. So for
118 one-step, robust estimation (and for all two-step estimation), the xtabond2 command also reports the Hansen J statistic, which is the minimized value of the two-step GMM criterion function, and is robust to autocorrelation. In addition, xtabond2 still reports the Sargan statistic in these cases because the Hansen J test has its own problem: it can be greatly weakened by instrument proliferation. Only the respective p-values are reported for this test results in the lower part of table 1. Here, the null hypothesis that the population moment condition is valid is not rejected if The summary statistics in columns 2 and 3 indicate that the one-step and two-step system GMM dynamic panel models of the selected 21 SSA countries have 18 instruments and 14 parameters each. This represents a total of 4 overidentifying restrictions in each case. In both specifications, the Hansen–J statistic does not reject the Over-Identifying Restrictions (OIR), thus confirming that the instrument set can be considered valid.
The F-statistic is the small-sample counterpart of the Wald (Chi Squared) statistic and it is a measure of the overall significance of the estimated models and the values here in each of the specifications are considerably satisfactory with level of significance being one percent in each case. This of course is indicative that all the exogenous variables jointly explained significantly, the economic growth process across the sampled SSA countries over the study period.
Results on the control variables are broadly and satisfactorily consistent with theoretical expectations. The Blundell–Bond (system-GMM) robust estimates (in specifications 2 and 3) indicate that growth dynamics are crucial and significant across the sampled SSA countries. An inspection of these results reveals that past realizations of economic growth produced some contemporaneous positive impact on economic growth. Precisely, a 100
119 percent increase in the past realizations of growth explained positively, about 27 percent of current growth levels. This is irrespective of whether the one-step or two-step collapsed instruments options is considered. In both cases therefore, it cannot be concluded that growth dynamics do not retard economic growth in the study group over the study period. This finding clearly agrees with that of Ahortor and Adenutsi (2009).
Size of labour force also produced some very meaningful and interesting results in the Blundell–Bond robust estimates. One striking observation here is that labour input produced a contemporaneous positive impact on economic growth across the sampled countries over the study period. This variable is also highly significant at the one percent level in the two–step system GMM option.
In more definitive terms, a one hundred percent increase in size of labour force under the two –step system GMM estimates, explains about 74.80 percent of the increase in economic growth across the study group. This result is not surprising since labour supply is in relative abundance in most of the SSA countries. It is therefore expected that the average production function in these economies will be characterized by enormous labour intensity. The transmission mechanism here is such that additional labour input in any of the selected SSA countries will directly impact on output growth. However, this argument will only hold as long as these economies operate within the positive region of the production function (that is before diminishing returns set in).
The implication of this result for theory is that economic growth inducing role of labour input is mostly applicable in the selected SSA economies.
Surprisingly, capital input is negatively signed and weakly significant at the ten percent level when the two-step system GMM with collapsed instrument option is considered. This result indicates that a one hundred percent increase in capital input in these SSA economies will explain about 187 percent
120 reduction in economic growth rate for these sampled economies. Capital input in this sense turns out not to be a major consideration in driving economic growth in the sampled SSA economies. This fact may not be unconnected with the relative dominance of the labour intensive sectors in most SSA economies.
Real effective exchange rate (REER) is found to be insignificant as an explanatory factor of economic growth in SSA. This of course is not unexpected given the small size of the foreign trade sector in most SSA economies. The foreign traded goods in SSA countries are dominated by imports of consumption goods which in the long-run do not bring about much economic growth.
Inflation rate variable should attract some comments as it explains economic growth in the sampled group over the study period significantly at the one percent level. This variable under the two-step system GMM with collapsed instrument option specifications produced a contemporaneous negative impact on economic growth across the sampled countries over the study period. As can be seen, a 100 percent increase in inflation rate explains about 108 percent reduction in economic growth in the selected SSA economies. A negatively signed coefficient for the variable, inflation rate is of course not unexpected as can be explained by the following transmission mechanism.
It is well known that borrowers benefit from major episodes of inflation, but lenders of loanable funds (that also double as profit maximizers) do counter the tide of inflation to minimize loss from the phenomenon. They achieve this by frequently adjusting the rate of interest upward probably to compensate themselves for loss of value in loaned funds and to keep track with the trend of inflation. The consequence of this kind of behaviour is that rising interest rate will constitute a disincentive to investment in these SSA economies and of
121 course lead to a decline in economic growth. Theoretically, this result confirms that while some mild inflation rate may be consistent with the goal of economic growth, persistently high inflation rate definitely will impede economic growth.
Given the two-step collapsed instruments option in the Blundell–Bond estimates, the domestic investment variable has a significant (at the five percent level) contemporaneous positive impact on economic growth across the sampled countries over the study period. Precisely, a 100 percent increase in domestic investment will explain about 0.15 percent increase in economic growth. The very low nature of share of economic growth explained by domestic investment may actually be explained by the high dependence on, and of course, the dominance of foreign investment component of total investment in many of the SSA economies. Overall, this relationship is not unexpected from the viewpoint of theory as investment remains a traditional driver of economic growth in every economy. This finding is similar to those of Chami et al (2003) and Faini (2006) who also found that domestic investment and private capital flows were positively related to growth.
However, the workers‘ remittances variable has an insignificant contemporaneous negative impact on economic growth across the sampled countries over the study period. What this finding suggests is that a significant proportion of remittances inflow to SSA is directed (intentionally or otherwise) at some economically unproductive uses. This result is in agreement with findings in Chami et al (2005). It is however in contrast with Ahortor and Adenutsi (2009). The relatively small volume of workers‘ remittances inflows to SSA countries could actually be the explanation for the insignificant result obtained for this variable. In terms of basis for comparism of the two works, Ahortor and Adenutsi (2009) used the Blundell and Bond GMM on a set of
122 dynamic panel models while Chami et al (2003) employed panel data methodology or the fixed effect estimation procedure. On grounds of data type, the outcomes of the two works may be comparable but this may not be true in the case of estimation techniques. The policy implication of this result is that for now, workers‘ remittances inflows may not be effectively relied on in driving economic growth in these SSA economies.