I spent hours checking the archives and nothing came up, so I'll have to ask
you this one. I need to know if -xtregar- can be run with a lagged dependent
variable (LDV) in a large panel (T = 400, n = 70, unbalanced).
I ran -xtregar depvar indepvar, fe- and obtained rho = 0.64, indicating that
DW = 0.72 (Baltagi-Wu LBI = 0.78), i.e. there is positive serial correlation
in my model. I also checked -xtserial- and it was significant. When I
included a LDV, rho = -0.49, so that DW = 2.98 (Baltagi-Wu LBI = 2.56),
and -xtserial- had even greater F value, again very significant. Although
this would mean a negative serial correlation, Durbin-Watson is not valid
with a lagged dependent variable.
So my questions would be:
1. Can -xtregar- be used with a lagged dependent variable?
2. If it can be used, is the estimate of 'rho' correct or not?
3. If it cannot be used, what is the alternative (for FE)?
