24 models are estimated for the five Asian countries whose sample periods are the same as those in the BP test (see Table 2). While some models indicate the evidence of heteroscedasticity, non-normality and functional form problem, all the models are free from autocorrelation at the 10% significance level or better. If heteroscedasticity is detected, the results are computed by the White heteroscedasticity adjusted standard error.
Unit root and Cointegration Tests
For examining stationarity in each series, both the ADF- and PP tests identify that all the countries’ EG, FD, FC and FR are non-stationary in their levels (except a few results) but become stationary after taking the first difference. Thus all the underlying variables are confirmed as /. Subsequently, the Johansen cointegration test (with unrestricted intercepts and no trends) is conducted treating FR as an exogenous I variable in the cointegrating vector (Note 15). Before conducting the cointegration test, the lag order of each country’s estimation is selected as the Johansen test highly depends on the choice of lag length. Checking the test statistics at the maximum order of four, we choose three lags for Korea, Malaysia and Thailand and four lags for India and Indonesia, respectively. Then the trace statistics in Table 3 report that, there is a single cointegration relationship (r = 1) among EG, FD and FC at the 10% level or better in all countries. The bounds test is implemented at the maximum lag order of either four (for India and Indonesia) or three (for Korea, Malaysia and Thailand); we refer to the statistics of the lag order selection in the VECM assessment. The test statistics in Table 4 reveal that, there is cointegration relationship in: all EG, FD and FC for Korea; FD and FC for India and Malaysia; and only FC for Indonesia and Thailand (Note 16). Indeed, although several F-statistics in Table 4 are judged as inconclusive in the bounds test, the presence of cointegration has been detected through the conventional unit root tests (i.e., the ADF and PP tests). Next while we seek the lag length of each underlying variable, both AIC and SBC give us only the lag selections that seem to cause autocorrelation in both India and Indonesia’s models. Hence, the orders of the two countries are manually set as presented in Table 4. For the other three countries, Korea’s models are selected by SBC and Malaysia and Thailand’s models by AIC, respectively. Satisfaction level of teachers
In Table 5 the findings relevant to the five Asian countries’ finance-growth nexus are reported. “Yes” is based on the strong exogeneity statistics significant at the 10% level or better, whereas “No” is insignificant strong exogeneity or indicates that cointegration is not detected by the bounds test (see Table 4). The weak exogeneity test results significant at the 10% level or better are given by “§”. Irrespective of the significance level, financial development and economic growth are positively related to each other in all the countries.
Table 3: Johansen cointegration test results (trace statistics)
|r = 0||r = 1||47.57*||61.36*||59.20*||37.86**||56.72*|
|r <= 1||r = 2||17.12||20.12||16.21||12.5||13.8|
|r <= 2||r = 3||2.92||0.93||0.97||2.48||4.43|
Table 4: Bounds test results and selected orders
|Country||Dependent variable EG||FD||FC|
Table 5: Finance-growth-crisis nexus (1)
|India||Yes**||No||Yes* §||Yes* §|
|Malaysia||Yes** §||No||Yes* §||Yes* §|