EViews.com

Hello team,
I am new to Eview, doin a independent project (Projct: Determinants of FDI) for learning purpose.
I have taken panel data(13 observations) of various independent variables:"GDP, Foreign exchange, WPI, Electricity generation, Crudeoil, Taxrate, Inflation,External Debt,Exchange Rate,Interest rate" and my Dependent Variable is "FDI".
I have made the variables stationary by taking Log , 1st difference and 2nd Difference. Found 2nd difference of all variables to be more stationary by comparing Correlogram and Unit test for all 3 scenario.
Hence, I have estimated equation using 2nd difference. Output was satisfying Normality and Heteroskedasticity test but not Serial Correlation test
Error: insignificant Degree of Freedom for lags=2 , If I Reduce lag to "1" ,then it rejects Null hypothesis.
I read somewhere that the only way to tackle this error is by reducing variables, changing no of Observations or adding lag.
Hence 1) I added lag of Dependent Variable (FDI(-1)) to remove the error but no satisfactory output.
2)I reduced the no. of variables(had to omit important variables like GDP), output was satisfying all conditions(serial correlation,heteroskedasticity and normality).Here I tried to find the Root mean sqaure error was very high (165, desired being 2).

Kindly suggest where am I going Wrong.

You should consider using the "AIC/BIC" to find out the optimum number of lags. One problem that I see with omission of a variable is that you might land up under specifying your model and hence the problem of serial correlation might get worse as your error term would no more remain purely unsystematic but would have a systematic component to it. Hence, it is better to use the appropriate information criterion to estimate your model.