[Chiraz KARAMTI]

Yes found it many thanks

[Haritima Chauhan]

Hi Chiraz,
So were you able to find solution. I have a similar concern.
And if so, what package/software did you use ?

Thank you !

[Chiraz KARAMTI]

Hi Haritima
In fact, it's the Antonio Galvao paper. Besides, the STAT tenure adviced me to use for example:
qregpd ln_wage tenure union l_ln_wage, id(idcode) fix(year)

[Haritima Chauhan]

Great ! Thanx

[sedki zn]

hello
i have the same probleme. i have panel date and i will use QR but i don't know what i use ; qreg or qregpd

[Gabriella Dickens]

Hi,

I am wondering if you managed to sort out the code to successfully model a dynamic quantile regression with the presence of fixed effects. Did you use the IV command to deal with the lags or did you just add the dynamic lags into the model ignoring the fixed effects. Would appreciate any help!

Thank you

[Mohammed Kasbar]

Dear Statalist respected users,

I am trying to estimate a quantile regression in dynamic panel data model. I used the command syntax which was developed by Powell (2015). The syntax was as follows:

qregpd scoreofagencyconflicts lagscoreofagencyconflicts pc1-pc8, q(0.75) id(firmid) fix(time) optimize(mcmc) noisy draws(1000) burn(100) arate(.5) instruments(lag2scoreofagencyconflicts lagpc1-lagpc8)
mat list e(gamma)

qregpd scoreofagencyconflicts lagscoreofagencyconflicts pc1-pc8, q(0.25) id(firmid) fix(time) optimize(mcmc) noisy draws(1000) burn(100) arate(.5) instruments(lag2scoreofagencyconflicts lagpc1-lagpc8)
mat list e(gamma)

I used the lags of my regressors as instrumental variables. However, the results are really bizarre as all variables are extremely significant in the two models which made me quite suspicious.

Could you please advise whether I wrote the command syntax correctly or I am missing something?

Thanks a lot in advance.

[Travis A. Smith]

Hi Mohammed,

Without any knowledge of the research question or the data, I would point out that if the MCMC draws "get stuck" away from the optimum, you can small standard errors. Have you tried using the default optimization of Nelder-Mead? I'd also say that the number of draws is on the "low" side, but there is no correct number of draws.

Also, a common root of many problems we hear about with -qregpd- is due to the usage of a large number of covariates. Again, there is no set number as to what is too many covariates, but the optimization is difficult. When you include upwards of 9 covariates, the optimization routine can be strained.

I would suggest trying the following:

* standard quantile regression with time effects
qreg scoreofagencyconflicts lagscoreofagencyconflicts pc1-pc8 i.time, q(0.25)

* qregpd without instrumenting: Nelder-Mead
qregpd scoreofagencyconflicts lagscoreofagencyconflicts pc1-pc8, q(0.25) id(firmid) fix(time)

* qregpd without instrumenting: use MCMC to search for starting values by using the vector corresponding to the max obj function value
qregpd scoreofagencyconflicts lagscoreofagencyconflicts pc1-pc8, q(0.25) id(firmid) fix(time) noisy draws(1000) usemax

* qregpd without instrumenting: use MCMC with starting values from previous regression
qregpd scoreofagencyconflicts lagscoreofagencyconflicts pc1-pc8, q(0.25) id(firmid) fix(time) noisy draws(1000) from(e(b))

How do those results look?

Also, users of -qregpd- should note that MCMC and Nelder-Mead using a random number generator. So it is advised to set a seed before running regressions.

[Mohammed Kasbar]

Travis A. Smith

Thanks a lot indeed.
I also contacted Powell, who is the author of the qregpd command and he advised me, as you did, to increase the number of draws given that I have many covariates.
Thanks a lot again. Much appreciated

[Francesca Murphy]

Hi Mohammed,

I have a similar problem, did you find that increasing the number of draws helped?

Many thanks,
Francesca Murphy

[Chiraz KARAMTI]

Hi,
I was reading Galvao paper about quantile regression with dynamic panel data. I'm not sure if my question is a little bizarre. Galvao said in his paper that the bias for the dynamic panel quantile regression variable can be ameliorated through the instrumental variables, namely the lagged regressors yi t−2 as instruments of yi t−1,. Ok, but we know that we can do that only for the differenciated equation, which is not the case in the Galvao paper! the parameters estimation procedure is based on the minization of an objective function in level! I saw the same reasoning in other papers!!!
I missed something???? How can we use the Arellano-Bond instruments along with equation in level? Please any help?

[merve kaya]

Dear Silva,

How can we estimate wald test after qregpd or xtqreg quantile model? After qregpd or xtqreg model how can we take R2 value?

[Joao Santos Silva]

Dear merve kaya,

What Wald test do you want to perform? The R2 does not make much sense in the context of quantile regression.

Best wishes,

Joao

[merve kaya]

Dear Joao Santos Silva,

I want to test whether the coefficient obtained from quantile regression model are equal to each other. For example, X1 is an explanatory variable.
[X1 from q(0.5)=X1 from q(0.75)].

This quantile wald test obtained from eviews easily, but 'eviews 10' estimates pooled quantile regression's wald test. I need fixed quantile or dynamic fixed quantile regression's wald test code.

[Joao Santos Silva]

Dear merve kaya,

If you use xtqreg with the ls option, you can test whether the variable has a significant coefficient in the scale function; if it has, you reject the null hypothesis that its coefficients do not vary across quantiles.

Best wishes,

Joao