[Joao Santos Silva]

Indeed, I would use xtpoisson or ppmlhdfe, which is more flexible and possibly faster.

Best wishes,

Joao

[Bernard Mamo]

Dear Joao Santos Silva ,

Thank you for the latter. I have two additional follow-up questions, if you don't mind.

1. As my variable of interest (colonial penetration; which is interacted with stock of migration) is a fixed effect, would it be safe to assume that -xtreg, fe- and other fe models are to be disregarded?
2. Secondly, given that I am operating at the HS-6 digit level, the relative amount of zero trade flows is relatively significant. Hence, when assessing on the basis of trade elasticity, would it make sense (for either -xtreg- or -ppml-) to limit the data values as follows?

xtreg LImports_Homogenous LDistcap Contig WTO GATT Lrgdp Com_lang Low_Col Int_Col High_Col Lmignocol i.Year if LImports_Homogenous>0, robust cluster(LDistcap)

Thank you once again; Best wishes,

Bernard.

[Joao Santos Silva]

Dear Bernard Mamo,

On 1, I guess you mean your variable of interest is time invariant? The expression "fixed effects" is now used to mean many different things, so depending on the structure of your data and the nature of the problem, you may want to introduce different sets of the so-called fixed effects. For example, trade models often include time-origin and time-destination fixed effects, and these are compatible with bilateral variables without time variation such as distance.

On 2, do not restrict your data like that, use all the observations if at all possible.

Best wishes,

Joao

[Bernard Mamo]

Dear Joao Santos Silva,

Yes, precisely; time-invariant. To be more exact, it is computed as follows:

gen double Low_Col= Low_Col_Pen* LMigration
gen double Int_Col= Int_Col_Pen * LMigration
gen double High_Col= High_Col_Pen * LMigration

In terms of data structure, I have reconfigured the data from HS-6 digit to HS-2 digit, and is as follows (eg)

CommodityCode	Year	PartnerISO	Import	Export
82	2002	ARG	344744	681986
70	2002	ARG	763063	2598111

If I might ask to check whether I am correct; I would imagine that time-destination effects are captured by simply:

i.Year

due to the fact that I am only assessing Bilateral trade flow between Spain and 47 trading partners. On the other hand, time-origin fixed effects would be captured by interacting the time dummy with each Spain-PartnerISO combination. Although, wouldnt such a substantail amount of fixed effects reduce the predictive power of the model?

Once again, thank you so much for your input.

Best wishes,

Bernard.

[Joao Santos Silva]

Dear Bernard Mamo,

I do not want to give you detailed advice because I am not sure to understand exactly what you are doing and therefore my advice can be misleading. However, I suggest you check this free book; it has details on all of this and it is easy to read.

Best wishes,

Joao