Questions about qregpd - quantile regression for panel data

[Henry Nguyen]

I ran "Qregpd" by Stata 15 with the equation follow:

Equation 1:

qregpd e_ta_bk mblag lagtang proflag lagsize lagblev, id(id) fix(year) optimize(mcmc) noisy draws(1000) burn(100) arate(0.5) quantile(.1)

and equation 2:

qregpd c1bk mblag lagtang proflag lagsize lagblev, id(id) fix(year) optimize(mcmc) noisy draws(1000) burn(100) arate(0.5) quantile(.1)

note: c1bk = negative (e_ta_bk)

I got results to follow the link attached.


First question: Why the sign of independent variables in both equations above is not the same the sign of result in reg pool, xtreg (fixed effect, random effect). My meaning's because c1bk = negative (e_ta_bk), so coefficients in independence variables of equation 1 and equation 2 will inverse.

Second question: I run (qreg, xtqreg, qregpd) again and I got the different results when compared with the results in last time by (qreg, xtqreg, qregpd)

Thanks so much for Professor Nick Cox advice and any reply.

[Joao Santos Silva]

Dear Henry Nguyen

I am not sure if I understand the problem, but here are 2 comments that may help:

1 - qregpd uses random numbers, so if you do not set the seed it is likely that you will get different results each time you run the command.
2 - qregpd estimates a model that does not include the fixed effects, so its results are not comparable to the results you get with xtreg and xtqreg because these commands estimate models that include the fixed effects.

Best wishes,

Joao

[Henry Nguyen]

Dear Professor Joao Santos Silva
Thank for your reply.
Best wishes,

[Henry Nguyen]

Dear Prof Joao Santos Silva
I run "XTQREG" for my studies, but the same as the "QREGPD" command, the results do not have Pseudo R2 or R2_adj. My question is how to calculate the Pseudo R2 when running "XTQREG" and "QREGPD" command? And when I use the QREG command, I can easy to check robust tests (the equally of coefficients between quantile level), but in "XTQREG", I can not find anyways.
Thank you so much,
Best wishes,

[Joao Santos Silva]

Dear Henry Nguyen,

Those R² measures do not have much meaning in this context, that is why they are not reported. You can formally test the equality of the coefficients in different quantiles using bootstrap. However, informally, you can just use the ls option and see what are the variables with significant coefficients in the scale function.

Best wishes,

Joao

[tc peng]

Dear Joao Santos Silva

I am recently learning quantile regression for panel data and I learned a lot from your posts.

I wish to know the difference between Powell's two packages: qrpd and qregpd. But so far I cannot find any instruction materials that give details on the codes of qregpd... only a brief content from the helper.
Also is there any way to graph the results of qregpd (just like the qreg package)?

Your reply will be deeply appreciated. Thank you.

Regards,
TC

[Joao Santos Silva]

Sorry, tc peng, you need to ask that to the author of those packages.

Joao

[tc peng]

Joao Santos Silva Thank you for suggestions.

[Jake Ed]

Hello, I have a question about using qregpd with panel data.

My dependent variable and independent variables are both highly skewed. So I am using quantile regression to estimate the relationship between the two using qregpd. However, except for the 5th quantile (q = 0.05), the standard errors and p-values for the lagged value of the dependent variable as blank. Am I making a mistake including the lagged DV in the model? If not, appreciate any suggestions for fixing this issue. Thanks! My basic set up is as follows:

xtset Names1 year
panel variable: Names1 (strongly balanced)
time variable: year, 1 to 97
delta: 1 unit


qregpd mentions max careerage count mentionslag1, quantile(.25) id(Names1) fix(year) optimize(mcmc) noisy draws(1000) burn(100) arate(.5)

Results for q = 0.25.

Quantile Regression for Panel Data (QRPD)
Number of obs: 7125
Number of groups: 75
Min obs per group: 95
Max obs per group: 95
---------------------------------------------------------------------------------
mentions | Coef. Std. Err. z P>|z| [95% Conf. Interval]
----------------+----------------------------------------------------------------
max | -1.37e-20 3.98e-20 -0.34 0.730 -9.18e-20 6.43e-20
careerage | 1.05e-21 3.04e-21 0.34 0.731 -4.91e-21 7.00e-21
count | 6.43e-21 1.87e-20 0.34 0.730 -3.01e-20 4.30e-20
mentions | 1.07554 . . . . .
---------------------------------------------------------------------------------
No excluded instruments - standard QRPD estimation.


MCMC diagonstics:
Mean acceptance rate: 0.216
Total draws: 1000
Burn-in draws: 100
Draws retained: 900
Value of objective function:
Mean: -95.7133
Min: -696.4812
Max: -41.8394
MCMC notes:
*Point estimates correspond to mean of draws.
*Standard errors are derived from variance of draws.

[Devashish Singh]

@Joao Santos Silva i am getting different results with qregpd each time i run the command. sometimes the significant variable comes to be insignificant and vice-versa. you suggested to use set seed so that we do not get different results each time. my question is how accurate is this method of setting seed is then? i can just set the seed which gives me the result that i am looking for. this can be abused a lot i feel

[Joao Santos Silva]

Dear Devashish Singh,

I suggest you do not use that method.

Best wishes,

Joao

[usef mohammad]

dear @Joao Santos Silva
Thank you very much for your comments on this page

Please guide me
I am just learning the quantile panel
I had two questions
First, how to interpret the mcmc test results in qregpd? For example, how to determine from these results whether the results are appropriate or not
MCMC diagonals:
Mean acceptance rate: 0.103
Total draws: 1000
Burn-in draws: 100
Draws retained: 900
Value of objective function:
Mean: -32.8528
Min: -104.8704
Max: -22.9117


Second question, How do we bootstrap test for equality the coefficients in different quantiles?

[usef mohammad]

@Joao Santos Silva
Excuse me and one more question
In your opinion, which approach is better to use, xtqreg or qregpd?

[Joao Santos Silva]

Dear usef mohammad,

I proposed the method implemented in xtqreg and mmqr, and wrote xtqreg, and naturally prefer that approach.

Best wishes,

Joao

[usef mohammad]

thanks @Joao Santos Silva
In this approach, what test is needed for the roboost of the results? How can we test the equality of the coefficients?
Bootstrap or any other test