If errors are not N(mue,sigma), some kind of transformation of data may be required to apply. It depends on the data itself. There are numerous data transformations are available in any text book.

My Dear

If the data not Normal distribution ..you can use Robust Methods to analysize your data in Regression ..

With Best Wishes

How do you know that the residuals are not normal? Is it by the Jarque-Bera test? Is it by the normal plot? What are the p-values?

That the f-statistic is significant implies the adequacy of the regression. It is overriding. The normality of residual requirement is not as important. Uncorrelatedness of the residuals is more important.

dear Prof. Ette Etuk

I used Jarque-Bera test and the 'p' value of the test is 0.000. Hence, I infer that the residuals are not normally distributed.

Does uncorrelatedness mean absence of autocorrelation of the residuals. If so, it is not relevant for me. Because, i have cross-sectional data of the companies in my sample.

Please advise me further

Yes uncorrelatedness of the residuals means that there is no autocorrelation amongst them. Check the autocorrelation function of the residuals too. If less than 5%/of them are significant the model may be considered adequate.

Dear Srikanth Potharla

Winsorize you data prior to estimating the GMM or 2OSLS. Hopeully you will find better outcome.

thank you prof. Muhammad Asif Khan for your valuable response