fixed effect parameter estimates with xtregar, fe

[Pierre Paul]

Hello everyone,

I am running a panel fixed-effects model with AR(1) disturbances using xtregar, fe. My data consist of firms observed at different various years.
When getting the predicted values of the fixed-effects (using predict command), the predicted FE don't show for years for which previous years are missing. For exemple, let's consider a firm i observed at years t=2005, 2006, 2009, 2011, 2013, 2014. Then, after using predict fix_eff, u, the column named fix_eff has an empty value for the year 2009 (as well as for 2005, 2011 and 2013). What is the reason behind that ?
Besides, I am wondering how the fixed-effects estimates are computed under xtregar, fe. I compared, within each firm, the given estimate to mean(dependent variable) minus mean(fitted values), as the fixed-effects are estimated that way under xtreg, fe, and found some differences. My question is simple : how are the fixed-effects estimates computed under xtregar, fe ?

Thank you.

[Phil Bromiley]

I can't be sure, but the documentation says these are only estimated within the sample. I suspect you're not getting fixed effects for observations that could not be used in the estimation and the previous value must be not missing due to the ar1 correction.