Hello.

I am estimating a stochastic frontier model using panel data for about 3000 firms over 20 years.

My key command is - sfpanel lnIK l.lnIK dlnSalesK l.lnSalesK dsalessq l.dsales_stdev l.stdev dumyear1-dumyear33, mode(tfe) distribution(exponential) usigma(lnCK dumold dumlev dumsize RV_P RV_N) vce(robust) (1)

where lnIK is investment rate and the other variables are explanatory covariates. I am able to get the point estimates of the coefficients using maximum likelihood technique, So I am using a true fixed effects model and assuming exponential distribution for one sided inefficiency term.

I would like to estimate the marginal effects of inefficiency covariates - lnCK etc in STATA. However, with exponential distribution, I can not use ' Predict, margin' post estimation command for this.

I therefore tried to use the conditional estimates of inefficiency given by - predict e_jlms, jlms

and then I run a fixed effects regression -

xtreg e_jlms lnCK dumold dumlev dumsize RV_P RV_N, fe (2)

I want to confirm is this is a valid way to derive the marginal effects of explanatory variables such as lnCK on firm level efficiency?

Thanks,

Anubha.