QREG - robust standard errors

[Joana Passinhas]

Dear STATA users,

I want to estimate quantile regressions using time series data but I wanted to know if there is already any way in STATA to obtain standard errors robust to autocorrelation for quantile regressions. I see that qreg2 has heteroskedasticity robust standard errors, but what about autocorrelation?

Thank you,
Joana Passinhas

[Joao Santos Silva]

Dear Joana Passinhas,

I am not aware of the existence of any command that does that.

Best wishes,

Joao

[Joana Passinhas]

Dear Professor Joao Santos Silva,

Thank you for the reply.

Can I also ask if it is normal to obtain smaller SE with qreg2 (robust) than with qreg when the test performed by qreg2 can not reject the null hypothesis of constant variance?

Best regards,
Joana

[Joao Santos Silva]

Dear Joana Passinhas,

If the null is not rejected, the variances should be asymptotically equivalent, so the difference should be cause just by sampling noise. Feel free to contact me directly if you want to discuss this further.

Best wishes,

Joao

[Joro Kolev]

This is an advanced question which has only advanced solutions.

But generally any bootstrap scheme which accommodates the error structure you assume, can be used with quantile regression.

With the error structure that you are ready to assume, the block bootstrap would be appropriate. So in principle you can apply block bootstrap on the quantile regression you want to run, and this would give you correct standard errors.

Block bootstrap is not a simple matter in Stata. To my knowledge there is no package doing it automatically.

[Joana Passinhas]

Dear Joao Santos Silva ,

Thank you for the reply - the differences are not very large so it should be that.

Dear Joro Kolev,

Thank you for the answer. I have been using block bootstrap as you suggest, I was just wondering if there was any other option in STATA already since 1) optimal length of blocks is not that trivial to choose and 2) my code might be a little inefficient and I wanted to estimate several quantile regressions.

Best,
Joana