Panel help

[Julie]

Dear Gareth and other Eviews helpers,

My data almost looks identical to those of the second format. I managed to import them in Eviews.
I have my dependent variable (abnormal_return) and several explanatory variables (trade, fina,...), each for 66 countries (cross-section) for 2 years (period).

Now I want to run simple regressions with them like abnormal_return c trade (both of which are expressed in percentages).

I was warned however about the existence of fixed effects for the panel variable.

Therefore, I ran the regression with fixed effects for both period (2) & cross-section (66), did the Redundant Fixed Effect - Likelihood Test and I obtain fixed effects for both.

The problem now is, when I run the regression with "none" (no fixed effects, no random), then I more or less get the results I want. The results with "fixed" for both period and cross-sections just isn't realistic. The output states if trade goes up with 1%, the abnormal return decreases with almost 4%. That just isn't possible!

Is it possible I should adjust something (f.e. only fixed for period?).

Please take a look at my attachments, you would help me enourmosly.

[arora]

Hi,

I face same problem as you, but what I estimate is direct investment. I also used fixed-effect model (according to F-test and Hausman-test), but the result with pooled regression is far much better and more theoretically makes sense.
Perhaps what I can suggest is to do some theoretical cross-check with other similar literature using the same method, and compare the results. If it's totally different, maybe there is something to fix in the model specification, or that the model spec you adopt is unsuitable for the sample countries, or there are some events happened in the time frame, which will affect the data. But if the results are quite similar, there must be an argument for that case.
Cheers :)