Fixed country effects and year fixed effects

[Eba Dijkstra]

Dear All,

I want to test the effect of trade agreements on trade using panel data. I composed a model and eventually performed a regression with time fixed effects.
The output : 'fyear 2-17 can be seen as the years 1996-2011.
How do I have to interpret those different years? what do these different years say about what? In 1996 (fyear2) .....

Dependent variable (GLI6) = trade.


Furthermore, my r-squared is very low.. how can I interpret this? I already changed my independent variables but the r-squared stays low when I include fixed effects (country pair fixed effects/year fixed effects) , does this mean that the dependent variable is not good to measure with this model?

Thank you in advance!

[Eba Dijkstra]

CORRECTION: : * my r-squared is very low if I use 'country pair fixed effects" and 'year fixed effects"

[Carlo Lazzaro]

Eba:
- I suspect there's something wrong in your -re- model (r--sq between=1.000 sounds like a warning message);
- it sounds also strange that you decide to switch from -re- to -fe- specification, as they arer pretty different beasts (by the way: all the time-invariant predictors are wiped out due to the -fe- machinery);
- I would check whether you have some omitted predictors.

[Eba Dijkstra]

Thank you for your reaction Carlo, very helpful! Do you mind to take another look?
I need a fixed effects model instead of a random effect model, I got rid of the omitted predictors and performed a new regression including code:
(1) country pair fixed effects
(2) country pair and year fixed effects

still very low R-squared..

[Carlo Lazzaro]

Eba:
- deleting by hand the time-invariant predictors is almost immaterial, as your results do not change that much;
- left aside the issue of omitted predictors, the low within sq-R (that is the sq-R you should look at if you go -xtreg,fe.) could be explained by a scant within panel variation of the time-varying predictors.