[River Huang]

Dear Joao Santos, Is it possible to have heteroscedastic errors in the (panel quantile) regressions? If positive, how can we deal with this situation (using bootstraps)? The usual -xtreg- command has the "robust" option, which accounts for the clustered (heteroscedastic/serially correlated) errors.

[Joao Santos Silva]

Dear River Huang,

Heteroskedasticity is not a problem, but serial correlation may be. In my personal webpage you find an example oh how to deal with it by bootstrap.

Best wishes,

Joao

[River Huang]

Dear Joao Santos, Thanks for the reply. I just visited your website (https://www.surrey.ac.uk/people/joao-santos-silva) but did not find the example. Could you kindly be more specific to inform me the address?

[Joao Santos Silva]

Sorry River Huang, I should have been more explicit: https://jmcss.som.surrey.ac.uk/MM-QR-JK.do

Best wishes,

Joao

[River Huang]

Dear Joao Santos, Got it and many thanks.

[Nik Anic]

Dear @Joao Santos - First thank you for the elaboration on the previous feeds. I'm new to Stata (experienced R user but for my professor I need to conduct a quantile analysis in Stata and that is actually my first time operating this program) so I am somewhat uncertain on especially three things:

1) In the paper on which you build the argumentation for the xtqreg function, it is stated that jackknife is able to deal with attenuating the inference problems identified, showing that the coverage of the confidence intervals is much improved. Is it correct to assume that this means that, even for small n to T combinations (e.g. when T would only be 2-4) the jackknife approach described could reduce the standard errors imposed? I am asking b/c I ran the xtqreg command with T = 3 and n > 1500 and, basically, the SE just exploded

2) If so, I tried applying the code you posted to River Huang but it did not work - when I run the modified code it gives the message that xtqreg does not allow the "if option" - e.g. if c>=3 & s==0 is not readable I assume. Do you know how to circumvent this issue?

3) Maybe I should open another thread to this, but it could be that you already know the answer: I was now experimenting also with the usual qreg2 function and using dummy coded variables as time and stratum fixed effects (again, small t but many strata). For the analysis, I have two outcomes of interest (let's call them y1 and y2), and when running the quantile regression, I can only obtain results for y1, but not for y2. To be more precise, I get results for all percentiles (0.05-0.95) for y2, but I only obtain results for y2 when running quantile regressions for the 0.3 - 0.8 percentile, but when trying to run it for e.g. pct. 0.1, 0.2 or 0.9, it renders the error: "Convergence not achieved". Do you have any intuition on why this might be (too little variation within one percentile maybe?)

Thank you very much for your guidance!

[Joao Santos Silva]

Dear Nik Anic,

With T=3 there is no hope to estimate quantile regression with fixed effects.

You can use "if" with the command, but note that it is not an option (so, it goes before the ,).

You can try to play with the wls option to see if you get convergence. Anyway, including "fixed effects" in that way is not valid.

My suggestion is that you try the so called correlated random effects approach as in

Abrevaya, Jason and Christian M. Dahl (2008). "The effects of birth inputs on birthweight," Journal of Business and Economic Statistics, 26, 379–397.

Best wishes,

Joao

[Daniel Lasso]

Dear Joao Santos Silva,

First, thank you very much for the answers on the xtqreg command.

I am in trouble since i can not get a constant on fixed effects quantile regression. However, you indicated to Lim that there was a way if you have the results of the regression. Could you please expand to me, How can I compute the constant in xtqreg?

Thank you,

Daniel Lasso,

[Joao Santos Silva]

Dear Daniel Lasso,

I could not find the post by Lim you mention so I am not sure if I understand what you mean because there is no constant as such in fixed effects models. If you are talking about estimates of the fixed effects, there is now an option to save them.

Best wishes,

Joao

[Daniel Lasso]

Dear Joao Santos Silva

Thank you very much for answering me.

I am new so I ended up answering a different post.

My question was about this post. I know that in fixed effects models the constant is not particularly meaningful, as you replied Li Ma, also the constant is not reported, however for me the constant is particularly useful as it might reflect some genetical (unobserved) aspects. Therefore, how can i compute or calculate the constant for each quantile when I am doing quantile regression. (xtqreg)

Thank you, and hope that I was clear.

[Joao Santos Silva]

This was answered in a different thread.

[Aviral Kumar Tiwari]

Dear @Joao Santos Silva,
I am trying to save results (for scale, location and quantile regression parameter and their standard errors and significance denoted by *) in excel using following codes. However, i get result only for regression coefficients for all quantile. I am new to Stata can you please help me to modify the codes so that I can save all the results. Manually doing is so time consuming.

local qs "0.05 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.95"
foreach tau in `qs' {
xtqreg x y z w, i(id) quantile(`tau') ls
outreg2 using "Machado and Silva (2019)", excel dec(3) ctitle(`tau') `replace'
local replace append
}

[Joao Santos Silva]

Dear Aviral Kumar Tiwari,

I can try to make that possible the next time I revise the command, but for the moment I do not think it is possible to do what you want. Note, however, that the scale and location results are the same for each quantile, so you only have to add them manually once.

Best wishes,

Joao

[Aviral Kumar Tiwari]

Ok dear Prof. @Joao Santos Silva,
Thanks

[Joao Santos Silva]

Dear All,

With the usual thanks to Kit Baum, an updated version of xtqreg is now available in SSC; the changes are not big, but all users are encouraged to update.

Please do let me know if you find any problems with the new version.

Best wishes,

Joao