Trade Capacity

As it is apparent from the fact that the selected sample countries contain heterogeneous properties, like different market size, per capita GDP and exchange rate and they are not exactly the same, therefore, there is probability that standard error may not be normally distributed. In this regard, serial correlation, heteroskedasticity and also the problem of endogeneity may occur. For robustness check, and also for any undermentioned problems, this study uses fully modified least squares (FMOLS) and dynamic ordinary least square (DOLS) estimators. Pedroni proposes FMOLS and DOLS to attain the long-run cointegrating coefficients. In the existence of "unit root variables", the impact of super consistency may not control the endogeneity problem effect of the regressors if ordinary least squares (OLS) is used.

FMOLS estimator was formerly proposed in work by Phillips and Hansen to postulate optimal estimates of cointegrating regressions. This approach adjusts least squares to account for "serial correlation" effects and for the "endogeneity" in the regressors that result from the presence of a cointegrating association. Furthermore, the asymptotic behavior of FMOLS in models with full rank I(1) regressors, models with I(1) and I(0) regressors, models with unit roots, and models with only stationary regressors.

The fully modified (FM) estimator was originally devised to evaluate cointegrating links directly by modifying traditional OLS. One reason, this technique has proven beneficial in practice is that one can use the FMOLS corrections to determine how imperative these effects are in an empirical practice. This has assisted to make the approach less of a "black box" for practitioners. In cases where there are main differences with OLS the source or sources of those differences can generally be easily located and this in turn helps to stipulate the researcher with additional information about important features of the data. Contemporary simulation experience and empirical research indicates that the FM estimator performs well in relation to other procedures of estimating cointegrating relationships.

DOLS and FMOLS are superior to the OLS for many reasons; (1) OLS estimates are super-consistent, but the t-statistic obtained without stationary or I(0) terms are only approximately normal. Even though, OLS is super-consistent, in the presence of "a large finite sample bias' convergence of OLS can be low in finite samples (2) OLS estimates may suffer from serial correlation, hetero skedasticity since the omitted dynamics are captured by the residual so that inference using the normal tables will not be valid - even asymptotically. Therefore, "t" statistics for the OLS estimates are useless (3) DOLS and FMOLS take care of endogeneity issues by adding the leads and lags (DOLS). In addition, white Heteroskedastic standard errors are used. FMOLS does the same using a nonparametric approach, see Arize et al., and Arellano and Bond.

To overcome these problems, we applied FMOLS, and DLOS methods. These models are capable of dealing with above diagnostic issues attributable to the Padroni and Kao cointegration test and also to the PMG estimator. The results are reported in Tables 7 and  8 which are consistent with main models; however, there is reasonable improvement in explanation power of some of the indicators due to error correction.

Table 7 Fully modified OLS results (export is dependent variable)

Table 8 Fully modified OLS results (trade deficit is dependent variable)