[Joao Santos Silva]

Dear Lukas Ferner,

Indeed, all models require assumptions, but some are stronger than others... Anyway, with such short "T", it is difficult to do anything very meaningful, but maybe you can try Ivan Canay's approach (also based on a strong assumption), but I do not think there is a command for it.

Best wishes,

Joao

[River Huang]

Dear Joao, Is it possible to test equality of coefficients (on the same variable) in different quantiles using -xtqreg- command?

[Joao Santos Silva]

Dear River,

The method allows that quite easily, but at the moment it in not implemented in the command. The alternative is to use bootstrap.

Best wishes,

Joao

[Suresh Paul]

Joao Santos Silva Is there any update on what River Huang posted?

Is it possible to test equality of coefficients (on the same variable) in different quantiles using -xtqreg- command? 

I would like to do the same - test the equality of coefficients of the same variable across quantiles. How would one go about doing this?

[Joao Santos Silva]

Dear Suresh Paul,

Unfortunately I did not have the time to update the command. However, you can do that using bootstrap. Alternatively, use the option ls and look at the results for the scale function; if a variable has a significant coefficient in the scale function it will have significantly different coefficients in some quantiles.

Best wishes,

Joao

[Suresh Paul]

Joao Santos Silva Thanks for the prompt advise. I will give it a try.

[Joao Martins]

Dear Joao Santos Silva

I am using the xtqreg command but my panel has a high (n/T).
In your paper I read about applying a Jacknife bias correction as a way to mitigate the inference problems that the high (n/T) may bring. Is there a way of applying this correction using the xtqreg command?

Best regards,

João

[Joao Santos Silva]

Dear Joao Martins

There is no option to do that in the command, but you can write a simple bit of code to do that. If you email me, I am happy to share the code we used in the paper to do that.

Best wishes,

Joao

[Ivan Ticofere]

Dear Joao Santos Silva

I am using the xtqreg command but my panel has a high (n/T=3800/3). Yet I obtain resuts that make sense.
However, in your paper I read about applying a Jacknife bias correction as a way to mitigate the inference problems that the high (n/T) may bring.
Could you possibly share how to apply this correction using the xtqreg command?
I supposestandard bootstrapping or jackknifing the standard errors is not necessary in this case?

Best regards

Ivan

[Joao Santos Silva]

Dear Ivan Ticofere,

If you email me I'll be happy to send you the code, but with T=3 there is hope to get reliable results.

Best wishes,

Joao

[Ivan Ticofere]

Dear @Joao Silva,

Thank you very much.
You mean "there is NO hope to get reliable results", I suppose?
Best,
Ivan

[Joao Santos Silva]

Indeed; thanks for pointing the embarrassing typo.

Joao

[Indrani Manna]

Dear Professor Joao,

Sorry for following up on an old thread. But I am currently running a quantile regression and I am facing a problem similar to Lukas Ferner above and I am quoting him "the calculated estimators for all variables at most of the quantiles are highly insignificant with high p-values". I have done an extensive regression analysis. This is not possible in all cases. Is there a way to get around this problem. I have used the xtqreg command in stata. I have 13 years of observations for each unit.

Many Thanks
Indrani

 xtqreg ef5_gdp crar z_score reg_post pvt_bond_cap_gdp five_bank_conc ff stock_mkt_cap_gdp, q(0.5)



                              MM-QR regression results
Number of obs = 351
WARNING: some fitted values of the scale function are negative
.5 Quantile regression
-----------------------------------------------------------------------------------
                  |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
------------------+----------------------------------------------------------------
             crar |   1.953971   56.45447     0.03   0.972    -108.6948    112.6027
          z_score |   1.815558   34.23989     0.05   0.958    -65.29338     68.9245
         reg_post |  -.0623857   1.691462    -0.04   0.971    -3.377591     3.25282
 pvt_bond_cap_gdp |    .373695   5.620276     0.07   0.947    -10.64184    11.38923
   five_bank_conc |   .9433943   11.00344     0.09   0.932    -20.62296    22.50975
               ff |  -.3657501   9.748104    -0.04   0.970    -19.47168    18.74018
stock_mkt_cap_gdp |  -.1017847   1.598762    -0.06   0.949    -3.235301    3.031732
-----------------------------------------------------------------------------------

[Joao Santos Silva]

Dear Indrani Manna,

I suggest you compute the standard errors by using clustered bootstrap, as illustrated here.

Also, note that the warning suggests that you may need to make your model more flexible.

Best wishes,

Joao

[Indrani Manna]

Dear Professor Joao,

Thank you for your kind response. I had two reasons for choosing the variables I have chosen for my regression.
1) I have a small sample
2) I preceded by panel regression with a bayesian model averaging to actually single out the covariates.

Is this the right way to go?

Regards
Indrani