[Simone Alim]

Dear Joao Santos Silva

Thanks for offering your insights! If I simply have to add an interaction term, is xtqreg or qreg better?

I'm also wondering if mdqr or xtmdqr have random effects (https://www.stata.com/meeting/switze...22_Melly.pdf)? Note: my data does not have repeated observations.

Any advise is appreciated.

Warmly,
Simone

[Joao Santos Silva]

Dear Simone Alim,

If you do not need the fixed effects, just use qreg2 (to be able to cluster standard errors); otherwise you need xtqreg. I am afraid I am not familiar with the other commands.

Best wishes,

Joao

[Elias Windscheid]

Dear @Joao Santos Silva
I have used xtqreg for a panel dataset with n=6 (countries) and T>100 (quarter).
When plotting the predicted values for just one of my countries there is periods where the predicted values for Q5 are above the predicted values for Q95, this should not happen right? Whats the mistake here?

I use the following model for estimation (this was the best model for h=4 when comparing Tick loss of different variable combinations) :
xtqreg F4_RHP CLIFS RHP dSRI, quantile(0.05)
predict q5_predicted
xtqreg F4_RHP CLIFS RHP dSRI, quantile(0.95)
predict q95_predicted
keep if country_id == 1


RHP: first difference of log of RHP (index)
CLIFS : country level indicator of financial stress
dSRI: first difference of domestic systematic risk indicator

I included graphs for the predicted values (Q5 and Q95) for Germany as well as RHP series for Germany and all countries.

Thank you in advance!

[Joao Santos Silva]

Dear Elias Windscheid,

Indeed, that should not happen. Are you getting the predictions only for the estimation period?

If you can, please send me the data and the code by email so that I can investigate. I'll report my findings here.

Best wishes,

Joao

[tetter bart]

Dear Mr. Santos Silva,
I am currently working on a panel quantile regression analysis with fixed effects and intend to use your Stata command xtqreg. However, I have a few questions regarding the functionality and interpretation of the command:

1. Which fixed effects are included in the estimation? Specifically, does the model account for country-specific fixed effects, time-fixed effects, or both?
2. Why does the output not display constants or pseudo R2 values? Are these metrics deemed less relevant in this context, and if so, could you kindly explain the reasoning?
3. Are there any preliminary tests or diagnostics you would recommend to assess whether xtqreg is appropriate for my analysis?

I would greatly appreciate your guidance on these matters, as it would help me better understand and apply the methodology.

Thank you in advance for your time and assistance.

Best regards,
Bart

[tetter bart]

Dear Joao Santos Silva,
I am currently working on a panel quantile regression analysis with fixed effects and intend to use your Stata command xtqreg. However, I have a few questions regarding the functionality and interpretation of the command:

1. Which fixed effects are included in the estimation? Specifically, does the model account for country-specific fixed effects, time-fixed effects, or both?
2. Why does the output not display constants or pseudo R2 values? Are these metrics deemed less relevant in this context, and if so, could you kindly explain the reasoning?
3. Are there any preliminary tests or diagnostics you would recommend to assess whether xtqreg is appropriate for my analysis?

I would greatly appreciate your guidance on these matters, as it would help me better understand and apply the methodology.

Thank you in advance for your time and assistance.

Best regards,
Bart

[FernandoRios]

Hi Tetter
1. It only accounts for Panel FE (country-specific in your case). If you would like to add year FE, just like with Panel data, you need to do it explicitly (i.year)
2. Because of the absorbed fixed effects, the Constant is not identified. You could calculate one (that is what xtreg or reghdfe do), but it wont be identifying anything.
R2, for qregression is not very informative.
3. Not sure about pre-tests. One thing to look for tho. If there are many "negative variance prediction" problems, then the results will not be reliable.

You should read the original paper by Machado and Santos Silva (2019) its very clear explaining the method and its interpretation.
F

[Joao Santos Silva]

Dear tetter bart,

FernandoRios already answered your questions and I just want to add that his mmqreg command implements the same estimator but has other options that you may find helpful.

Best wishes,

Joao

[Atlas Dogan]

Dear Professor Silva,
If I may ask your guidance on some matters, in the presence of the cross-sectional dependency, are the results valid?
I have used mmqreg with a robust option because slope coefficients are heterogeneous. (mmqreg target var1 var2 var3, abs(Company_ID) q(10...90) robust)

[Joao Santos Silva]

Strictly speaking, cross-sectional independence is assumed; I do not know the consequences of relaxing that assumption.

[Afonso Moura]

Hi Joao Santos Silva and FernandoRios!
What about temporal dependence? I want to implement a block bootstrap to resample rows of data from the temporal dimension of each country, keeping unchanged the cross-sectional structure of the panel. This consists of resampling “blocks” formed of contiguous rows of data. How can I do this with xtqreg or mmqreg?

Many thanks
Afonso

[Joao Santos Silva]

Dear Afonso Moura,

Again, strictly speaking, temporal independence is assumed in the paper, but more for simplicity, so I think it is OK to do what you say, as long as T is large enough.

Best wishes,

Joao