I am running a panel data regression for 16 countries with 15 annual observations. Due to the fact that EViews does not provide an autocorrelation test for panel data (?), I had to switch for an absolutely unknown STATA. Hausman test (p=0,43) indicated that I should use Random Effects with GLS.

Based on guidelines here, I tested panel for heteroskedasticity, autocorrelation (HA) which are both present in the panel.

  1. To my understanding robust standart errors are fine to estimate approximate coefficients for variables with HA. I used xtreg rgdppc gerter trade empl, re cluster(id). Is this an "okay" method to remove HA? I am new to econometrics.

  2. In a post here, regressions' residual is tested for stationarity. Should I store the Panel RE GLS residual and test it for stationarity? Using Levin-Lin-Chu unit-root test, all series (rgdppc, gerter, trade, empl) showed stationarity (p=0,00), however, with Breitung none of them were stationary. Is this normal?

  3. And to the point - how should I test RE model for cointegration? With aforementioned GLS's residual or I should go instantly for the first difference?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.