How to deal with heteroskedasticity in panel regression (gretl) I'm currently analyzing the profitability determinants of Isamic banks in GCC countries and I'd like to run a regression in which ROA is the dependent variable and the independent variables are 5 bank-specific variables (Size, Capital Adequacy Ratio, NPL ratio, Cost-to-Income ratio, Liqudity ratio), 3 macro-variables (GDP growth, Inflation and Money Market interest rate) and a Dummy variable (1 for Islamic banks and 0 for Conventional ones).
I have a panel of 114 banks (45 Islamic and 69 Conventional banks operating in 6 countries) over a time period of 5 years.
I run in gretl a pooled OLS and according to the White's test there is heteroskedasticity.
Could someone tell me the steps to follow through gretl in order to correct this bias? 
 A: As AdamO said, calling it a "bias" is a little misleading, but I suppose you want to do correct inference (that is, estimate the standard errors of your coefficients consistently).
When you say pooled OLS I understand it (in contrast to AdamO) to be a model without group-specific intercepts, so a standard regression.
In Gretl's GUI interface you have a tickbox "robust standard errors" in the model specification dialog. Tick it. Next to it you have a button where you can choose the precise way of doing heteroscedasticity-consistent errors, but you can probably leave it at the default.
If you use it in a script (CLI way), then add the option "--robust" at the end of your ols command.
[Sorry for not answering earlier, I haven't been checking Crossvalidated in recent weeks.]   
A: Heteroscedasticity does not bias linear models unless the model is misspecified. It leads to possibly incorrect calculation of standard errors. If by pooled OLS you mean that you are aggregating values and conducting an ecological study of averages, then you will need to use precision or frequency weighting. Irrespective of that, if there are more than 40 observations, the Huber White error estimator gives valid SE estimates regardless of the variance structure of your data.
