PPMLHDFE use

[Jose Antonio Espejel]

Hello everybody,

I am studying the impact of infrastructure on horticultural trade using a gravity equation at a product level. I will be using country-product and product-year fixed effects. The new "ppmlhdfe" command seems to be the best to use. However, it is taking a very long time to calculate. My dataset is about 2 million observations. Does this has something to do with the observations dropped due to be singletons or separated by fixed effects? Around 10443 observations are dropped due to this. (I have a lot of zero trade observations in this dataset)

Has someone experiences this? Any solutions?

Thank you.

[Jose Antonio Espejel]

The program does not go further than below:

[Joao Santos Silva]

Dear Jose Antonio Espejel,

Tom Zylkin may be able to provide further insights, but it may just be that with that number of observations and fixed effects the estimation takes a long time.

Best wishes,

Joao

[Tom Zylkin]

Hi Jose Antonio Espejel

It looks like in your case the program is never actually reaching the estimation step. Most likely it is because it is having a hard time determining whether your data is suffering from separation or not. I suggest adding the option "sep(fe simplex)". This will simplify the steps used to look for separation. Hopefully it will speed things up.

Regards,
Tom

[Jose Antonio Espejel]

Dear Tom Zylkin ,

Your suggestion works perfectly. Thank you for your help.

BR,
José