Quantile model with fixed effects

[Lucia Gracia]

Dear @Joao Santos Silva,

I have panel data spanning 31 countries from 1990 to 2018, encompassing 29 years. My goal is to explore how independent variables influence the dependent variable across different quantiles. I intend to incorporate fixed effects for both years and countries to account for time and country-specific factors. My initial command for conducting quantile regression is as follows:

xtset cnt year
xtqreg manEMP lnPop lnPop_sq lngdppc lngdppc_sq lnfdi i.year i.cnt, q(0.25 0.5 0.75)

However, the results are puzzling, since the coefficients for countries appear as omitted, prompting me to question if I may be overlooking something or if fixed effects are not applicable in this context. Additionally, I seek guidance on addressing heteroskedasticity within the quantile regression framework while simultaneously incorporating fixed effects.

Could you kindly provide insights or suggest alternative approaches to address these concerns?

[Joao Santos Silva]

Dear Lucia Gracia,

The command xtreg automatically includes the cnt fixed effects (because of how you xtset the data), so you should not include i.cnt in your model. You do not have to worry about heteroskedasticity, but should worry about serial correlation: you can use bootstrap by countries to compute valid standard errors, as in this example.

Best wishes,

Joao

[Lucia Gracia]

Thanks for your answer @Joao Santos Silva
It is the first time I am working with the quantile model and I am a bit lost. I did what you suggested and I also looked to other posts in the forum. I have now this command: xtqreg manEMP lnPop lnPop_sq lngdppc lngdppc_sq ln_fdi i.year, ls q(0.10(0.40)0.90) which should be correct for fixed effects.

But I am facing problems to compute standard errors, is this code correct?


bootstrap, cl(cnt_enc) r(100) id(id) seed(123): xtqreg manEMP lnPop lnPop_sq lngdppc lngdppc_sq fdi ln_fdi i.year

Thanks for your time.
Best,
Lucía