significance between using different kinds of fixed effects in gravity model

[Page Alarcon]

Hello all,

My regression is a panel gravity-type regression, with sector level bilateral trade data.

My results are significant with importer-time + exporter-time + product-time FE , but not so with importer-product-time + exporter-product-time FE.

Can anyone please tell me if it is fine to use the first set of FE given I have industry-level data? What could the non-significant in the second set of FE mean?

Thank you!

[Phil Bromiley]

You didn't get a quick answer. You'll increase your chances of a helpful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex. Also, remember that we are not from your area - what a gravity model means is not defined in most statistics texts.

It is hard to understand your question. Is product the same as industry? What is the observation unit? You seem to have a very large number of fixed effects. I would worry that importer-product-time + exporter-product-time FE would be close to the number of observations. I'd be worried that you are actually fitting your variables with almost spare no degrees of freedom.

[Page Alarcon]

Hi Phil,

Thanks a lot for the reply, even though it was confusing! Yes I am running a trade (gravity) model regression, which typically uses a whole lot of fixed effects dummies. And yes by product I mean industry/sector, and the observation unit is a bilateral (country-)pair and product specific observation for a given year. So for instance it is the trade flow between country i and j for product/sector k in year t.

I interpret it that country-product-time FE is more "rigorous" than country-time plus product-time FE separately in terms of accounting for endogeneity, but is not useful for me since I lose significance with this level of FE, and also that it might be the type of RHS variables that I am using that affect this. I have both bilaterally varying trade policies and tariffs, but some of them may also be non-bilaterally varying (e.g. if the tariff is a non-discriminatory "most-favored nation" tariff that is the same for all partner countries, or the trade policy applies to all partners the same way). Because of this I know I cannot use country "pair" FE, but I wonder if I also cannot use country-product/sector-time FE?

Any help would be very appreciated!