SWAMY—Random coefficients estimator in panel data

[TomDoan]

@SWAMY implements the Swamy random coefficients estimator from Swamy(1970), "Efficient Inference in a Random Coefficient Regression Model", Econometrica, vol 38, no 2, pp 311-323. @MEANGROUP is a similar procedure which does pooled mean group (equally-weighted) estimates.

Detailed description

[behnam]

Hi TOM,
I am working in fixed effects panel data but this is a heteroskedastic panel with serial correlation.and i must estimate a homoskedastic without any correlation and i don't know exactly what should i do. in stata there is a code named xtgls that can solve this two problems jointly, is there any rats code like that? i try to do @swamy procedure but results are completely different with stata and serial correlation is still remaining.Could you kindly help me on this?
thanks in advance,
behnam

[TomDoan]

SWAMY is completely different from anything that xtgls does. AR1 corrections, whether in single equation or multiple equations/panel data are generally an indication that you have inadequately dealt with the dynamics of the model—there is almost always a better way to handle that using lags in the regressors.

[behnam]

Dear tom,
thanks for your prompt answer,
what about heteroskedasticity? how i can solve this problem and get homoskedastic result?can adding lags solve heteroskedasticity or there isn"t relation between them?

[shimarats]

hi tom.

my problem is same behnam problem . if i do swamy -random coefficients estimator in panel data. i am doing to model of panel and another pooled method. i find to two problem like hetroskedasticity and serial correlation .which coding can solve both these problems ? is it possible i used swamy code.? after solve these problems ,can i used by EGLS method that means after i solve these problem ,how i can to estimate the equation of panel data. thanks dear

[TomDoan]

As with the previous poster, the Swamy procedure has nothing to do with either heteroscedasticity or serial correlation. And, as with the previous poster, if you have serially correlated errors, fix the dynamics of the model. Slapping an AR correction on a poorly specified model has been considered bad form for at least 20 years.