Marvin, kindly note that importer-year and exporter-year fixed effects are captured by multiplying the importer dummy by the year dummy and exporter dummy by the year dummy. Similarly, note that when you have dummies in your gravity equation, you cannot use a fixed effects estimator because such an estimator will wipe out all the dummies from your estimation results. You would rather use a random effects estimator (or pooled OLS).

I favour country-pairs as the cross-sectional index. Say you have 3 countries then the country pairs for export flows are a->b, a->c, b->a, b->c, c->a, c->b, i.e. n=6. If you enter your data according to this design for the cross-section then Eviews' cross-section and time fixed effects will account for country-pair effects and year effects.

Dear Marvin Sachs

I suggest you to use STATA for this purpose because Eviews may limit some estimations.

For STATA you can use these commands after XT ;

For creating fixed effects

qui tab year, g(dyear)

qui tab importer, g(dimporter)

qui tab exporter, g(dexporter>

After that add theses effects in your regression.

Good Luck

Cheers!

Saqib

Thank you for your answers!

@Eluid: I will take this into account as I use pooled OLS with dummies and Fixed Effects estimation without time-constant dummies.

@Muhammad: Unfortunately, I am not aware of using STATA properly. I have to stick with EViews.

@Vince: I will stick with your suggestion because my data has the form as in your example. Therefore the inclusion of time and cross-section fixed effects in EViews should be fine. This is in line with the idea, that the type of fixed effects depends on the structure of the underlying data.

One additional thought on the underlying data and the use of country-pair fixed effects.

I found one study that estimates the gravity model using only exports of one certain country and no trade pairs, e.g. country a with b (a-b), a-c, a-d ...

In this case, it would make sense to create an exporter dummy or exporter-time dummy that controls for the characteristics of the single country a.

If this conclusion is totally wrong, please let me know.

Which of the following fixed effects are relevant for estimating a gravity model with one country as exporter: export fixed effects, import fixed effects, country pair fixed effects and year fixed effects? my data set is panel data and my analytical software is stata 14