Fixed effects in PPML gravity model regression

[Adithya Kolloju]

Hello everyone,
I am performing a PPML fixed effect gravity model to check the attractiveness between universities. My data set contains 7459 observations from the period 2001-2020. The variable list has Institutional factors, Regional factors, and Economic factors.
I would like to use a variable from Institutional data as my dependent variable and all other factors as independent variables. I want fixed effects in this regression model. I selected some variables like size, ranking, and type of the institution (which are string variables), with other independent values (integer or numeric values). I have some variables which contain large observations. I converted the variables (dependent/independent) to log values to run the PPML regression with fixed effects.
How many fixed effects can I use in the PPML model? Can anyone suggest me?
Thank you.

Best regards,
Adithya.

[Maxence Morlet]

You'll definitely want to use the community-contributed ppmlhdfe command. Literature on the incidental parameter problem for Poisson models is plethoric, but I guess a good place to strt is Weidner and Zylkin (2021). They specify the magnitude of the bias according to the fixed-effects vectors included in your estimation. They focus on three-way gravity models.

[Maxence Morlet]

Also, when using the ppmlhdfe command, if I were you I would not "pre-transform" the variables and feed level variables into the programme.

[Joao Santos Silva]

Dear Adithya Kolloju,

Just to reiterate what Maxence said above, do not log the dependent variable when estimating by PPML; you can log the regressors if that helps.

Best wishes,

Joao

[Adithya Kolloju]

Thank you for your suggestions, I will try to execute this package.