Dear Statalist,

I am running a fixed effect model (with clustering at firm level) to model investment behaviour of firms in response to their past performance and past industry performance and I am unsure of whether I should include a lagged dependent variable in the regression (i.e. investment at t-1).

My supervisor now told me to start by reporting the results from the static regressions and potentially move onward to a dynamic regression.

I read in Cameron and Miller (2015) that the best rationale for deciding between a static vs a dynamic model is to run a test of serial correlation. They refer to Inoue and Solon 2006, a Portmanteau test .

Now, I already ran a xtserial test and it rejected the H0 of no serial correlation. Do I have to run an additional command like xtistest? Or would it essentially do the same work as xtserial?

Secondly, does clustering at firm level correct for serial correlation? I thought it does, but it would help me to get some confirmation here.

Many thanks for any advice on this!
Katharina