Panel data estimatioin

[Sadam Khan]

Hello. Need help with panel data estimation.
I am working on panel data with T=96 and N=260. I used Hausman test and it gives FE an appropriate model for estimation as shown

 
Table: Hausman Test

Ho:        difference in coefficients not systematic

chi2(6) = (b-B)'[(V_b-V_B) ^(-1)](b-B)
			             =      251.86
			Prob>chi2 =      0.0000
			(V_b-V_B is not positive definite) 			
Table: Hausman Test

Ho:        difference in coefficients not systematic

chi2(6) = (b-B)'[(V_b-V_B) ^(-1)](b-B)
						             =      251.86
						Prob>chi2 =      0.0000
						(V_b-V_B is not positive definite)

But the hetero and auto test show both problems

 
			Table: Wald test for heteroskedasticity

H0: sigma(i)^2 = sigma^2 for all i
			chi2 (252)  =   2.8e+38
			Prob>chi2 =      0.0000

 
Table:Wooldridge test for autocorrelation

H0: no first-order autocorrelation
			F (1, 244) =    873.172
			Prob > F    =      0.0000

In accordance to the above data and results I need suggestions. Should I go for FGLS

xtgls

or use

xtreg

with

cluster

option? Plus if [CODE][cluster/CODE] then cluster with only Panel ID or Panel ID and Time both?

[Carlo Lazzaro]

Sadam:
welcome to this forum.
1) despite your T dimension is relevant, probably you can still use -xtreg-, clustering your standard errors on -panelid- and plugging -i.timevar- in the right-hand side of your regression equation.
You should test the joint significance of -i.timevar- as a post regression estimate via -testparm-;
2) -vce(cluster clusterid)- does not support more than one cluster variable. However, what you have in mind can be accomplished thanks to the community-contributed module -reghdfe- (just type -search reghdfe- to spot and install it), provided that -fe- is the right specification for your data;
3) -hausman- does not support non-default standard errors. Hence, you should switch to the community-contributed module -xtoverid- (just type -search xtoverid- to spot and install it).