[Fadi Ansar]

Dear Atrayee Choudhury,

I realize that you directed your question to Matthew, but while you wait for his reply, and for what it is worth, here is my view on it: There is nothing in Powell's method that prevents you from using interactions; these are treated as any other explanatory variable. I would add, however, my usual warning: Powell's method is extremely elegant but it estimates a model that is very different from what many users have in mind. so, make sure the estimator is doing what you think it is doing and if not consider using alternative estimators that are also available in Stata.

Best wishes,

Joao

Dear Joao Santos Silva ,
As you have mentioned we can use intreaction variable while using qregpd command ,I was getting the following error and showing that no interactions are allowed .How can I use interactions while using qregpd command .

 qregpd Zscore d_callmoney C5_Aggregate CapitalRatio NIM lAsset C5_Aggregate##d_callmoney,id(bankn
> ame) fix(Year)
interactions not allowed
r(101);

Thanking you,
Fadi

[FernandoRios]

Hi Fadi,
This means that you have to create the variables by hand, before you use them in your regression.
Unfortunately qregpd is programmed for an earlier version of Stata that didn't have factor notation, so you have to do it by hand.
F

[Fadi Ansar]

Thankyou Fernando Rios .After considering your commnets I have run the quantile regresion with qregpd command .but I am getting results like below (No sd value ,P value and confidence intrevel for C5 aggregate varaiable).is missing data is an issue for running this command .Does any one have any idea why I am getting results like this

 qregpd Zscore d_callmoney C5_Aggregate CapitalRatio NIM lAsset MPR_callmoney ,id(bankname) fix(Ye
> ar)
Nelder-Mead optimization
initial:       f(p) = -282.96707
rescale:       f(p) = -282.96707
Iteration 0:   f(p) = -282.96707  
Iteration 1:   f(p) = -206.73859  
Iteration 2:   f(p) = -206.73859  
Iteration 3:   f(p) = -182.30701  
Iteration 4:   f(p) = -122.72508  
Iteration 5:   f(p) = -122.72508  
Iteration 6:   f(p) = -119.88349  
Iteration 7:   f(p) = -119.88349  
Iteration 8:   f(p) = -119.88349  
Iteration 9:   f(p) = -119.88349  
Iteration 10:  f(p) = -117.18588  
Iteration 11:  f(p) =  -115.0798  
Iteration 12:  f(p) = -113.87909  
Iteration 13:  f(p) = -112.63421  
Iteration 14:  f(p) = -111.13069  
Iteration 15:  f(p) = -111.13069  
Iteration 16:  f(p) = -111.13069  
Iteration 17:  f(p) = -111.13069  
Iteration 18:  f(p) = -111.13069  
Iteration 19:  f(p) = -111.13069  
Iteration 20:  f(p) = -111.13069  
Iteration 21:  f(p) = -111.13069  


Quantile Regression for Panel Data (QRPD)
     Number of obs:               832
     Number of groups:             83
     Min obs per group:             1
     Max obs per group:            14
-------------------------------------------------------------------------------
       Zscore |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
--------------+----------------------------------------------------------------
  d_callmoney |  -.0577889   1.759453    -0.03   0.974    -3.506254    3.390676
 C5_Aggregate |  -2.415386          .        .       .            .           .
 CapitalRatio |   8.522331   162.4066     0.05   0.958    -309.7888    326.8334
          NIM |  -.0301694   .4509956    -0.07   0.947    -.9141045    .8537657
       lAsset |   1.269422   24.29435     0.05   0.958    -46.34664    48.88548
MPR_callmoney |   .1058607   2.903621     0.04   0.971    -5.585131    5.796853
-------------------------------------------------------------------------------
No excluded instruments - standard QRPD estimation.

Thanks in advance
Fadi ansar

[FernandoRios]

Hi Fadi
I believe you had a similar problem when using other commands like, xtqreg and xtreg.
Namely, C5_aggregate was highly collinear with other explanatory variables (or perhaps with the fixed effect). qregpd using the information of fixed effects different from other xtqreg, so its possible that the colinearity problem passes undetected, and only appears near the "solution" of the problem
So, my advice.
take a step back and figure out if your model gives you sensible results using simple panel data estimators (xtreg, fe). This will also help you figure out some basic data problems you may have.
Once you set a model with the lienar regression, you can try applying xtqreg or qregpd. Just keep in mind that these models identify different types of parameters

Best wishes
Fernando

[Fadi Ansar]

Thanks again for your suggestions FernandoRios .there is no such problem while running xtqreg ,xtreg fe and qreg commands .the only problem is with qregpd command.so now i am confused as can I go with other command

[Hamid muili]

Thanks for the insight. Can the qregpd be used for dynamic quantiles analysis with instrumental variable similar to Galvao (2011) estimator

[Joao Santos Silva]

No, it can not.

[Akbar Isayev]

Good day. Does anyone know why genqreg provide different results each time after running same code? Even I set seed, still I face with the same problem.

Thanks in advance

[Stefano Grillini]

Dear Matthew,

I am finding an issue while running the genqreg command. While estimates are computed for lower quantiles, at the 90th all coefficients are 0 and highly significant. I've tried running genqreg for the 85th and 95th quantiles and results are more in line with the results obtained across lower quantiles.
Any idea why this is happening?

See below the options I am currently using for the estimation.

 optimize(mcmc) noisy draws(10000) burn(3000) arate(.5)

Happy to provide a replicable code is necessary.

Thanks
Stefano

[Stefano Grillini]

Dear Matthew,

I am finding an issue while running the genqreg command. While estimates are computed for lower quantiles, at the 90th all coefficients are 0 and highly significant. I've tried running genqreg for the 85th and 95th quantiles and results are more in line with the results obtained across lower quantiles.
Any idea why this is happening?

See below the options I am currently using for the estimation.

 optimize(mcmc) noisy draws(10000) burn(3000) arate(.5)

Happy to provide a replicable code is necessary.

Thanks
Stefano

[Stefano Grillini]

Apologies all,

I'm had issues posting the message and it just got copied twice.

[Mugabil Isayev]

Subject: MCMC Chains in genqreg Command Dear Professor Matthew J. Baker, I'm using the genqreg command in my research: set seed 5000000 genqreg depvar indepvar1 indepvar2 indepvar3, q(10) optimize(mcmc) noisy draws(1000) burn(100) arate(.5) instruments(indepvar1 indepvar2) The number of MCMC chains isn't explicitly mentioned in my code and I couldn't find this detail in the genqreg documentation. Could you please inform me about the default number of MCMC chains used by genqreg, and if it's possible to modify this number? Thank you for your time.

[Junayed Hossain]

Dear Mathew,

Could you please provide me with an update why those above mentioned link for Powell(2015, 2016) doesn't work, please? or which paper they were?

Kind regards

[Kathi Kaeppel]

Hi all,

I am trying to apply these estimators to a panel data problem with year and age fixed effects.

My understanding is that the proneness() option of qgenreg is for the "control variables" (as opposed to treatment variables) - these controls do not change the interpretation of the estimates, but might be needed for identification.

My question is: If I would like to add covariates, while still getting unconditional QTE, should I use qgenreg and put my year and age dummies in proneness(), or do I have to use qregpd (as it is for panels) - however then, I can put my year var in the fix() but I am not sure where to put age FE if I want those to be accounted for, but not part of the model (as in Powell (2015) and implemented in qgenreg.

Many thanks!