For matter of clarity, let us call the equation that tries to identify the determinants of FDI as
“the first equation” and that the equation that tries to identify the determinants of GDP growth as
“the second equation”. An ad hoc estimation technique was used in order to describe the best model
and consequently different models with different lags of dependent variables and different
estimation methods were applied.
The first model uses Ordinary Least Squares (OLS) estimation method, to identify the first
and second equations. t-statistics of all the independent variables in the first equation are
insignificant for 1%, 5%, and 10% levels of significance. In the second equation, t-statistic of gFDI,it
and hc(-5) is insignificant at all levels, while gX,it is significant at 1% level. Our test results
indicate us that OLS regressions do not produce statistically reliable/significant results.
In the second model, Two Stage Least Squares Method (TSLS) was used to estimate the
system. The results indicate that t-statistics of gY,it , hc(-5) and gX,it in the first equation are
insignificant. Moreover in the second equation hc(-5) is significant at the 10% level; gX,it is
significant at the 5% level, and gFDI,it is significant at the level of 1%.
In the third model, Three Stage Least Squares (3SLS) estimation technique was used in
order to estimate the system. hc(-5) is insignificant both in the first equation and the second
equation. On the other hand, in the first equation, gX,itis significant at the 10% level and gY,it is
significant at the 1% level. Moreover in the second equation of the system, while gX,it is significant
at the 5% level, gFDI,it shows significance for the level of 1%.
The fourth model, which was estimated by GMM technique, yield that gY,it and gX,it are
statistically significant at the 1% level and signs are positive, as expected, in the first equation.
However, hc(-5) is statistically insignificant in the same equation. All the coefficients are
statistically significant at 1% in the second equation and signs of variables are as expected.
However these results are not sufficient to make any interpretation about the fitness of the model.
As it was mentioned before we are applying an ad hoc estimation approach. Consequently, we must
continue to estimate other models with lagged values of dependent variables of the system.
Fifth model consists of one year lags of gFDI,it and gY,it and is estimated by GMM method,
as inclusion of one year lagged values of dependent variables implies that the model behaves as an
autoregressive model. As it can be seen from the table, all independent variables are significant in
the first equation, though at varying significance levels. However, in the second equation, one year
lagged value of GDP growth rate is statistically insignificant.
Sixth model consists both one-year and two-year lagged values of gFDI,it and gY,it ,
respectively. Our estimation results show that hc(-5) is insignificant in the first equation and two-
year lagged value of gY,it is statistically insignificant in the second equation. All other variables in
both equations are significant at different levels of significance and also their signs are as expected.
Finally, seventh run consists of three-year lags of dependent variables. In the first equation,
coefficients of hc(-5) and gFDI(-3) are statistically insignificant. Moreover, in the second equation
gY (-2) is insignificant.
Our analyses suggest that the best model for our system is model 5. In model 5, coefficients
of the variables show that FDI and economic growth are important determinants of each other.
Also, it is obvious from the results that export growth rate and human capital are statistically