Quantile regression with panel data (Random Effects)

[Bruno Chaihuaque]

Hi there!
I'm analyzing a data set with which I've constructed panel data. My question is is it mandatory/possible to perform a quantile regression with random effects? Considering that Hausman's Test suggests a random-effects model, but is a short almost balanced panel with 10 years (2011-2020)/500 observations. In this scenario, I decided to carry out an analysis using quantile regressions with qregpd package

[Joao Santos Silva]

Dear Bruno Chaihuaque,

The Random-Effects estimator is not particularly interesting in this context. I suggest that you use either pooled regression (with qreg2) or try the fixed-effects estimator implemented in xtqreg. If you really intend to use qregpd, you should be aware that the model it estimates is somewhat unusual; for more, see

http://www.econ.uiuc.edu/~roger/rese...rettes/DQR.pdf

Best wishes,

Joao

[Bruno Chaihuaque]

Thank you for your prompt reply. It is possible to perform a fixed-effects analysis even when Hausman test suggests trying random-effects? there is some literature that supports it? Excuse me for this basics questions...
QUOTE=Joao Santos Silva;n1658342]Dear Bruno Chaihuaque,

The Random-Effects estimator is not particularly interesting in this context. I suggest that you use either pooled regression (with qreg2) or try the fixed-effects estimator implemented in xtqreg. If you really intend to use qregpd, you should be aware that the model it estimates is somewhat unusual; for more, see

http://www.econ.uiuc.edu/~roger/rese...rettes/DQR.pdf

Best wishes,

Joao[/QUOTE]

[Joao Santos Silva]

Dear Bruno Chaihuaque,

The Hausman tests the null hypothesis that the RE estimator is valid; if you accept the null, that does not rule out using FE. A good textbook should explain this.

Also, (a robust version) the test may make sense in the context of least squares estimators, but not in the quantile regression context. So, you can do FE quantile regression whatever the value of the Hausman test.

Best wishes,

Joao

[Bruno Chaihuaque]

Thank you very much, professor Santos Silva. Got a lot of valuable information. I look forward to continuing to learn from your comments.

Dear Bruno Chaihuaque,

The Hausman tests the null hypothesis that the RE estimator is valid; if you accept the null, that does not rule out using FE. A good textbook should explain this.

Also, (a robust version) the test may make sense in the context of least squares estimators, but not in the quantile regression context. So, you can do FE quantile regression whatever the value of the Hausman test.

Best wishes,

Joao