Tips and tricks how to speed up ppmlhdfe

[Theresa Spickers]

Hi all,

I am working with ppmlhdfe with a very large dataset (approx. 175 mio obs). I run a regression with entity and time trend fixed effects (about 5 mio and 8 k each). I have enough computing power available (I use 10% of CPU and 35% of RAM with ppmlhdfe), but still each regression takes about 30-45 minutes to run.

This is the regression specification:

 ppmlhdfe dep_var treat0_post treat1_post treat2_post treat3_post treat4_post , absorb(event_user event_trend) cl(event_id)

Do you have any tips and tricks how I could speed up the estimation?

Thanks!

Theresa

[Wei Hai Deng]

While I don't have any experience with ppmlhdfe, I have on occasion used parallel to speed up work on especially large datasets.
There is also multishell. Both are available from ssc.

Best,
Wei