POOLED OLS, test for heteroskedasticity and autocorrelation

[kresna mansueto]

Hello, everyone.
I run a hausman test to help me choose whether I will use random effect (RE) or fixed effect for my panel data. The result indicates that I will use RE. I further run Breusch-Pagan LM test, probability>chi=.4545. BP, I guess indicates that I will be using pooled OLS. May I know how to test for heteroskedasticity and autocorrelation? Thank you.

[Carlo Lazzaro]

Kresna:
welcome to this forum.
Would you please post what you typed and what Stata gave you back (via CODE delimiters; see the FAQ)? Thanks.
That said, if -xttest0- outcome shows no evidence to reject the null, -re- is the way to go.

[kresna mansueto]

Thanks, Carlo. When I execute
quietly xtreg $ylist $xlist, re
xttest0

Stata gave this:

Breusch and Pagan Lagrangian multiplier test for random effects

lmw[id,t] = Xb + u[id] + e[id,t]

Estimated results:
| Var sd = sqrt(Var)
---------+-----------------------------
lmw | .0932644 .3053922
e | .0764476 .2764915
u | .0053361 .0730486

Test: Var(u) = 0
chibar2(01) = 0.00
Prob > chibar2 = 0.4763

Given above, will I be using, RE?

[Carlo Lazzaro]

Kresna:
in my previous reply a "not" is missing.
Hence, as -xttest0- does not reject the null, you should go pooled OLS.

[kresna mansueto]

Thank you, Carlo.

[kresna mansueto]

Further, will I proceed conducting diagnostic test like collinearity, homoscedasticity, autorrelation and linearity? Thanks, Carlo.

[Carlo Lazzaro]

Kresna:
if you go pooled OLS, you should impose cluster standard errors: hence, surely autocorrelation and perhaps some form of heteroskedastcity are accounted for.
Perfect collinearity is autiomatically managed by Stata, by simply omitting variables.
You might be interested in the (probably a bit oversold: see https://www.hup.harvard.edu/catalog....40&content=toc, Chapter 23) -estat vif- for detecting quasi-extreme multicollinearity.
Squared predictors can be useful to investigate whether quadratic relationships with the regressand are more than a suspect.
Last but (by no means) least, I would check fopr possible model misspecification via -estat ovtest- or, even better, -linktest-.

[kresna mansueto]

Thank you, Carlo. I will execute your suggestions and have this book too.

[Malki Ranathunga]

regress var list, vce(robust)
can we use this command if we have reject null hypothesis in heteroskedastcity and use robust option to this issue? (under pooled OLS model)

[Carlo Lazzaro]

Malki:
as -robust- does not take serial correlation of non-independent observations into account, I would go:

regress <depvar> >indepvars> <controls> vce(cluster  panelid)