Geopolitical Factors and Cross-Border Capital Allocation in "Gravity" Model

[Chiara Rossi]

Dear Joao Santos Silva,

I am trying to replicate results from a regression in the IMF Global Financial Stability Report 2023 (Online Annex). They examine whether geopolitical factors matter for cross-border capital allocation using a bilateral panel, and they estimate the following "gravity" model:

log(X_c,c',t) = 𝛽 GEOPOLITICALDISTANCE_c,c',t-1 + 𝛿 GRAVITYCONTROLS_c,c' + 𝜈_c',t + 𝜈_c,t + 𝜖_c,c',t

Where X_c,c',t is the portfolio share, c is the destination country, c' is the origin country, and 𝜈_c',t and 𝜈_c,t are origincountry-time and destinationcountry-time fixed effects.

I say "gravity" because they estimate the model using OLS, while I read in many of your posts that gravity models should use for example PPML.

I have two questions:

1) If I want to estimate the model with OLS as they did, how do I code the origincountry-time and destinationcountry-time fixed effects in xtreg? This is what I have so far, I don't know how to add the country-time varying fixed effects.

xtset countrypair year, early

xtreg log_share L.GPD gravity_controls, fe robust

2) Would it be more sensible to estimate the same model with PPML?

I would be extremely grateful if you could help!

Best,

Chiara

[Joao Santos Silva]

Dear Chiara Rossi,

Welcome to the Forum.

1) You can use the user-contributed command reghdfe that allows you to include the fixed effects in a very flexible way (check the help file)
2) You should certainly use PPML to estimate the model. For that I suggest you use the user-contributed command ppmlhdfe (that that for PPML the dependent variable should not be logged).

Best wishes,

Joao

[Chiara Rossi]

Dear Joao Santos Silva ,

Thank you very much! This was extremely useful since I am quite new to Stata. A quick follow up on both questions:

1) To account for the above fixed effects, would absorb(i.origincountry#i.year i.destinationcountry#i.year) be correct? or should I use the c operator instead?
2) For PPML, would dropping the log but keeping the rest of the model unchanged as below be correct?

X_c,c',t = 𝛽 GEOPOLITICALDISTANCE_c,c',t-1 + 𝛿 GRAVITYCONTROLS_c,c' + 𝜈_c',t + 𝜈_c,t + 𝜖_c,c',t

Thank you very much again!

[Joao Santos Silva]

Dear Chiara Rossi,

1) do not use any operators, just write absorb(origincountry#year destinationcountry#year).
2) That is correct.

Best wishes,

Joao

[sabeer vc]

Dear Chiara Rossi,

Welcome to the Forum.

1) You can use the user-contributed command reghdfe that allows you to include the fixed effects in a very flexible way (check the help file)
2) You should certainly use PPML to estimate the model. For that I suggest you use the user-contributed command ppmlhdfe (that that for PPML the dependent variable should not be logged).

Best wishes,

Joao

Joao Santos Silva Why Dependent variable is not to be logged but independent variables are to be logged? (Sorry. I have asked the same question somewhere else also. But this recommendation seems providing me the context)

[Joao Santos Silva]

https://www.statalist.org/forums/for...ppmlhdfe/page2