[OWUSU ANSAH]

No, I do not think there is a command for Galvao's estimator.

Thankyou. I want to ask, if there is a way I can implement the Galvao technique since you applied this in your paper. I have a lagged dependent variable in my right hand side variables and I think the best way to deal with the bias is to use the Galvoa Instrumental variable approach.Thanks in advance

[Joao Santos Silva]

Dear OWUSU ANSAH,

You would have to write the code for the estimator. Is your lagged dependent variable the variable of interest? If it is not and it has low correlation with the variable of interest, you can simply try xtqreg.

Best wishes,

Joao

[Mia Liu]

Dear Joao Santos Silva,

We are using xtqreg for a panel fixed-effect regression. The median regression results all look as expected, in terms of the signs and magnitude. However, the standard errors are VERY large. I am wondering whether that's something that people have commonly experienced with the package xtqreg?

Thank you!
Mia

[Joao Santos Silva]

Dear Mia Liu,

No, that is not my experience. How large are your T and N?

Best wishes,

Joao

[Mia Liu]

Thank you Joao Santos Silva! My T and N are both large. T is about 20 and the number of distinct persons is about 30,000.

I am wondering whether our specification is suitable for quantile regression. We are interested in estimating whether an event has any impact on wealth. So I(t>=0) is an indicator variable, for when the event has happened. The left hand side is wealth. We are interested in the coefficient "gamma".

We have used xtreg with fe option, which works well. In the robustness check, we want to try to see whether we could also use xtqreg with id(person) option. That's when we got very large standard errors. Interestingly, the coefficients from the xtqreg are similar to results from xtreg.

Any insights would be highly appreciated.

Thank you,
Mia

[FernandoRios]

Hi Mia
adding my two cents here.
I would suggest you try bootstrapped standard errors with blocked bootstrap (using id as cluster)
from my own understanding on the method is that standard errors for xtqreg work well as long as the iid assumption holds.
if not, I found that the problem you report raises
in that case, I think bootstrap standard errors work better.
hth
fernando

[Mia Liu]

Thank you for your nice suggestion! FernandoRios

[Joao Santos Silva]

Dear Mia Liu,

FernandoRios makes an excellent suggestion; see here an example of how to do it.

Best wishes,

Joao

[Mia Liu]

Thank you Joao Santos Silva and FernandoRios. After play around the code, I discovered that, even without bootstra, if I just restrict my sample to a more balanced panel, by using c>=10, then my standard errors behave well and my coefficients are similar to those from xtreg. So I am guessing, xtqreg works well with a more balanced panel?

Best,
Mia

[Joao Santos Silva]

Dear Mia Liu,

The panel being balanced should not matter, but having large T does, so restricting the sample to T>=10 is a good idea.

Best wishes,

Joao

[Mia Liu]

Thank you Joao Santos Silva!

Mia

[Taruntej Singh]

Dear Prof. Joao Santos Silva

I wish to apply quantile regression on my panel data of 281 firms over a 5 year period.

I have come across these commands: qreg, qreg2, qregpd, xtqreg and mmqreg.

Please let me know which one will be best suited for my analysis.

PS- I have already conducted panel data regression with firm and year fixed effects and standard errors clustered at the firm level (using the xtreg command) as my primary analysis. I need to follow this up with quantile regression.

Regards

Tarun

[FernandoRios]

Hi Tarum
xtqreg and mmqreg do the same. so either will be fine
the qreg and qreg2 do not do panel
and qregpd can also do panel but has a different set of assumptions that you need to understand compared to the method of moments implemented by xtqreg and mmqreg
fernando

[Taruntej Singh]

Dear FernandoRios

Between xtqreg and qregpd, which one should I go for then for my panel?

Regards
Tarun

[Mia Liu]

Dear Joao Santos Silva and FernandoRios:

Thank you both for your advice last time on my questions regarding xtqreg.

I have another question. When I use xtqreg, fe and xtreg, fe, on the same data set, the number of observations on the xtqreg (~90,000) is smaller than that in xtreg (~120,000).

I didn't do any adjustments like T>10. I used bootstrap on xtqreg.

More specifically, my code is
xtreg wealth ${fereg_retire} b(0).d_afterretire, fe
bs, rep(100): wealth ${fereg_retire} b(0).d_afterretire, id(hhidpn_num)

Would you know why that is the case? Thanks a lot!

Jialu