Autocorrelation and Heteroskedasticity issue

[Jeevan sai]

hi, i have panel data of 22 cities with 17 years, and i used log transformations for my variables in xtreg , after carrying out the hausman test to decide between Random- effect or Fixed-effect model, the result was that Fixed effect model would be better. After this, I applied some tests to verify problems of heteroskedasticity, autocorrelation such as:

xttest2 (Breusch-Pagan statistic for cross-sectional independence in the residuals)
xttest3 (heteroskedasticity)
xtcsd, pesaran abs (cross sectional independence)

it turned out that my data has both autocorrelation and heteroskedasticity , how can i deal with it?

[Carlo Lazzaro]

Jeevan:
just invoke -cluster- or -robust- options (they do the very same job under -xtreg-) for standard errors to take both heteroskedasticity and autocorrelation into account.
After that, you cannot use -hausman- anymore to compare -fe- vs -re- specification (as -hausman- does not support non-default standard errors) but you have to switch to the community-contributed programme -xtoverid- (just type -search xtoverid- to spot and install it).

[Jeevan sai]

thanks Carlo, I invoked the robust in my code but the wald chi2 statistics are missing. and can i perform normality test for panel data as a diagnosis test?

[Carlo Lazzaro]

Jeevan:
as fas as the unreported chi2 statistic is concerned, see -help j_robustsingular-.
Normality is a (weak) requirement for residuals distribution only.

[Jeevan sai]

Ok Carlo.even after using robust in code autocorrelation and heteroskedasticity tests shown that still my models have them. Then I used xtgls ( n=22, t= 17) where it generated that it does not have heteroskedasticity or autocorrelation but can i avoid R square values in gls model.

[Carlo Lazzaro]

Jeeven:
it does not make any sense to test for heteroskedasticity and/or autoccorrelation when you imposed non-default standard errors, as the result of the tests will not change (because they work on residuals, that remain unchanged when you switch from default to non-default standard errors).
That said, -xtgls- sounds sensible here, because your T dimension is not negligible.
Hovever, -xtgls- does not allow to investigate fixed effect (see -xtregar-, instead).

[Jeevan sai]

Ok Carlo, i ran xtregar but with in r squares values are very low (0.3954) and rho ar values is 0.68 and how can i interpret r square values in xtgls?

[Carlo Lazzaro]

Jeevan:
I would not say that your within Rsq is low, not tah rho is negligible .
Unfortunately, -xtgls- does not return Rsq but -chi2- statistic.