[Cobby Stoneson]

Thank you very much Joao Santos Silva and Jeff Wooldridge . I will try both methods to see how they turn out and if there are any more concerns I will ask for further clarifications. I am grateful for your responses.

[Cobby Stoneson]

Dear

Jeff Wooldridge,

I am not sure if I should open new thread on this, but please permit me to clarify a few things on the quantile regression from the procedure outlined above:

Assuming a balanced panel, the following should work. z1 ... zJ do not change over time, the xk do. I'm showing for the median and with time dummies.
Code:

egen x1bar = mean(x1), by(id) egen x2bar = mean(x2), by(id) ... egen xKbar = mean(xK), by(id) qreg2 y d2 ... dT z1 ... zJ x1 ... xK x1bar ... xKbar, q(.5) cluster(id)

---

1. You stated that the procedure above should work for assuming a balanced panel. Please what happens if I have an unbalanced panel? Are there some modifications that can be made for the procedure to be valid for

unbalanced panel?

2. I was reading 12.10.3 of your book,

"Econometric Analysis of Cross Section and Panel Data," 2010 where you describe the procedure in detail. However, I do not seem to pin down the exact name to call this procedure/method in my paper. Is it Correlated Fixed Effects Pooled Quantile Regression using the Chamberlain-Mundlak Device? Or What is the appropriate formal name that I should call it?

3. My final question is arising out of my result from using a balanced sample from my data. I present below results just for the most important variables of interest (X2, X3 and their interaction). I left out the time averages and everything else. I realize that the mean-effect for X2 is largely above the effects at all the quartiles (in absolute terms). Are such results acceptable where the mean effect is either above the effects at the upper quartile or below

the effects at

the lower quartiles?

 
Results for Total Demand: Quantile Effects






























(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

(13)

(14)

(15)


Mean_Effects

Q_05

Q_15

Q_25

Q_45

Q_50

Q_55

Q_65

Q_75

Q_85

Q_8875

Q_90

Q_95

Q_99

Q_9975

VARIABLES














































X1
58.1145**

21.9807

23.8127

41.3204**

62.0513***

76.3585***

81.1684***

71.7518***

69.6911***

88.0096**

115.1847***

102.9998**

71.5924

-127.9446

-339.5787***


(26.1367)

(17.1387)

(18.4458)

(17.3128)

(21.3057)

(22.7966)

(21.5244)

(21.1384)

(24.4936)

(34.9356)

(43.8180)

(45.3374)

(84.0194)

(197.0906)

(67.3985)
































X2
-63.7874***

-16.0557

-19.9413**

-28.4479***

-32.6784***

-39.5835***

-41.7841***

-32.9478***

-27.3055*

-36.7389

-51.0431***

-43.1305

-24.9020

64.1774

59.2534***


(12.2551)

(9.9122)

(9.0798)

(9.1337)

(10.3506)

(11.8519)

(11.8891)

(12.1729)

(14.3594)

(23.0355)

(18.1450)

(26.5561)

(31.9265)

(122.0253)

(9.1669)
































X2*X3
-19.6117**

-2.5290

-12.6594

-18.2245**

-14.5731

-7.8030

-10.8540

-18.4405

-35.7097***

-48.9821*

-43.7227

-51.5993

-71.5634**

-92.4668

-215.2832***


(9.5994)

(11.6036)

(8.7018)

(8.7483)

(9.4771)

(9.3916)

(10.6235)

(12.7284)

(12.4179)

(26.0121)

(27.5833)

(37.7203)

(33.7521)

(134.6215)

(15.0696)
































X3
2.0538

-8.5343

3.2451

9.5758

5.8199

3.1130

4.7100

4.5319

11.2573

18.7810

22.7945

18.4069

36.1491

-16.1521

41.5128**


(6.4674)

(9.3836)

(6.5352)

(6.7032)

(7.0103)

(7.2752)

(8.2248)

(9.1673)

(9.1687)

(22.2990)

(23.8810)

(24.8503)

(22.3783)

(41.7486)

(18.2451)
































Observations
7,240

7,240

7,240

7,240

7,240

7,240

7,240

7,240

7,240

7,240

7,240

7,240

7,240

7,240

7,240
































R-squared
0.418

0.2094

0.3337

0.3801

0.4122

0.4185

0.4206

0.4184

0.4099

0.3920

0.3839

0.3854

0.3578

0.2707

0.2047
































Number of Individuals
1,448




























































Correlated Fixed Effects
YES

YES

YES

YES

YES

YES

YES

YES

YES

YES

YES

YES

YES

YES

YES
































Robust standard errors in parentheses





























*** p<0.01, ** p<0.05, * p<0.1

Thank you again for your time.

Cobby.

[Mehmet Sari]

I am glad to see the post #62 above. I was reading the same part of Jeff Wooldridge's book yesterday and I had the same exact question. I hope he can answer it.

Thanks.

[Jeff Wooldridge]

Cobby, Mehmet: It's probably not too important what you call it. I think correlated random effects quantile regression is as descriptive as anything.

As far as unbalanced panels, I would use the method I propose in my 2019 Journal of Econometrics paper. The CRE approach can be used, but now you should model the heterogeneity as functions of the number of complete cases available for each i. The easiest way to do this is to define different dummy variables for having two periods, three periods, ..., up to T periods. These T-1 dummies can be added along with the time averages. You can also interact those dummies with the time averages to get more flexibility.

As I discuss in my paper, my proposed method is for using complete cases. Data missing on any variable gets dropped. I define a complete cases indicator ahead of time to reduce the chance of making a mistake.

I hope this helps.
Jeff

[Cobby Stoneson]

Dear Jeff Wooldridge ,

This is absolutely helpful. Straightway I am off to look for and read your 2019 paper on the approach.

And thank you Mehmet Sari for helping "pull" the distinguished professor from his busy schedules.

Thank you all.

Cobby.

[Michael Haylock]

Hi Joao,
Joao Santos Silva
I am running the latest version of the package xtqreg.
Thanks for the package, it is turning out to be useful.

I am having trouble getting the tables of location and scale parameters while using bootstrap:

sysuse auto
xtqreg price weight length i.foreign, i(headroom) quantile(.25) ls
bootstrap, cluster(headroom) idcluster(newid) group(headroom) reps(10): xtqreg price weight length i.foreign, i(headroom) quantile(.25) ls

When using the bootstrap, tables of location and scale parameters are not reported. It is not obvious to me why this is the case... can you please help me here?

I am aware that these are saved and the covariance matrix is as well, but it a bit tedious doing all this by hand when the tables otherwise are automized.

Thanks in advance,
Michael

Stata SE 15.1

[FernandoRios]

Hi Michael,
The reason why the location and scale parameters are not reported, is because, unless otherwise specified, bootstrap only reports results for e(b). xtqreg only stores the total results on e(b). Location and scale are stored in other matrices.
Perhaps this would be useful tho:

sysuse auto
program xtqwrapper, eclass
xtqreg price weight length i.foreign, i(headroom) quantile(.25) ls
matrix b1=e(b)
matrix b2=e(b_scale)
matrix b3=e(b_location)
matrix coleq b1=total
matrix coleq b2=scale
matrix coleq b3=location
matrix b=b1,b2,b3
ereturn post b
end
bootstrap, cluster(headroom) idcluster(newid) group(headroom) reps(10): xtqwrapper

[Michael Haylock]

Dear FernandoRios, thanks very much for the code. That's very helpful.

[Joao Santos Silva]

Thanks for replying to this, FernandoRios.

Joao

[Ryan Banerjee]

Dear Ch Man,

The variable will be dropped if it does not vary over time. The fixed effects can be computed using the expression at the top of page 12 (see the link to the paper in the help file); all the ingredients needed to compute that expression are saved by the command: e(q), e(b_location), and e(b_scale).

Best wishes,

Joao

Dear @Joao Santo Silva
Could you check the code to compute fixed effects following a call to xtqreg? I wasn't 100% sure about the notation in your paper. Also as a 99% R user, I sometimes find the magic of the STATA e() variables confusing!
Thanks
Ryan

** First run xtqreg with say quantile(0.05) 
xtqreg f4infl infl dlrealgdp dlneer dloilprice_domcur, i(countryid) quantile(0.05)

** Collected e() variables
mat b05=e(b)
mat score xb05=b05
mat q1=e(q)
mat score qest05=q1

******************************************************************************
*     back out the fixed effects alpha_i(tau) = alpha_i + q(tau)*delta_i
******************************************************************************

* 1. alpha(i)

mat fe1=e(b_location)
mat fe2=e(b_scale)
mat score feloc=fe1
mat score fescale=fe2

gen alpha1 = f4infl-feloc
sort country obs
by country: egen alphai=mean(alpha1)

* 2. delta_i

gen Rhat    = f4infl-feloc - alphai
gen Rhatabs =  abs(Rhat)
gen delta1  =  Rhatabs - fescale
sort country obs
by country: egen deltai = mean(delta1)

*************************************************************
* Put the pieces together for the quantile fixed effect
*************************************************************
gen quantile05_fe = alphai+(deltai*qest05)

[Joao Santos Silva]

Dear Ryan Banerjee,

Apologies for the late reply. Your code looks fine to me but I suggest you check it with a simple example where you can easily confirm the results.

Best wishes,

Joao

[Ryan Banerjee]

Apologies for the late reply.

No need to apologise! Your reply was blisteringly fast! Thanks again, your function is super helpful.
Best
Ryan

[Joao Santos Silva]

Dear Ryan Banerjee,

I have updated the command so that you have an option to generate the fixed effects. If you want to try it, please do contact me by email.

Best wishes,

Joao

[Jake Ed]

Joao Santos Silva , Hello,I am running into an error with using bootstrap with xtqreg. I would appreciate any suggestions for correcting my mistake, which is likely a very simple one. I have 75 groups over 97 years in my panel.Thanks in advance.

xtset Names1 year

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

generate newid= Names1

bootstrap, cluster(Names1) idcluster(newid) reps(100):xtqreg mentions max careerage xcount mentions_lag i.year, i(Names1) quantile(.05) 

Using the above command, I get the error message:

running xtqreg on estimation sample)

Bootstrap replications (100)
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5
repeated time values within panel
the most likely cause for this error is misspecifying the cluster(), idcluster(), or group() option

[Joao Santos Silva]

Dear Jake Ed,

I believe you need i(newid) instead of i(Names1).

Best wishes,

Joao