ppmlhdfe with only fixed effects

[chantal Eleser]

Dear Statalist users,

Happy to join the community.


I am tring to using ppmlhdfe command regress the firm-level foreign investment data on the country-pair-industry-year fixed effect. the code is as follows:
ppmlhdfe foreigninv , a(sourcecountryid#destcountryid#industryid#year) cluster(parentcompanyid ) irr nolog separation(mu fe) d(fixeffectsum)

My purpose is to seperate the firm-level characteristics such as productivity and revenue or scale from the country-pair-industry-year factors that influene the entry decision and the amount of investment. I considered the observed variable foreigninv minus fixeffectsum is the firm-level characteristics that are reflected in the observed data foreigninv.
My questions are:
1. Can I do the seperation using ppmlhdfe command？What's the economic and statistic meaning of the "foreigninv - fixeffectsum" ? Can it be explained as the firm-level characteristics that are reflected in the investment??
2. If not, in what method can I do the seperation?
3.Can I even ppmlhdfe without any independent variables and just the sourcecountryid#destcountryid#industryid#year ?
4. What the meaning is the new-generated variable "fixeffectsum"? the "fixeffectsum" are mostly negative, which really confused me.

Thank you so much!!

[Joao Santos Silva]

Dear chantal Eleser,

I cannot answer all your questions, but I can answer some. The "fixeffectsum" variable you are creating are just the estimated fixed effects. If you want to get the fitted value, you need to use predict after the estimation. You can run PPML just with fixed effects, as you did. Check the ppmlhdfe help file to make sure you are using the fixed effects you want.

Best wishes,

Joao