Dear all,

I am using ppmlhdfe to estimate a gravity model that includes an auto-regressive (AR) term to adjust the dependent variable's potential serial correlation.

In my knowledge, the ppmlhdfe assumes the linear parameters enter the model exponentially, leads me to think of log-transforming the AR if I want to add it linearly. But, it drops the zero values since log(0) = -inf.

Is there a proper way to add a lagged dependent variable, or AR, into my specification using ppmlhdfe?
Or is it not appropriate to add AR at all in ppml?

Best regards,
Dongin