[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: instrumental variable for quantile regression

From   "alessia matano" <>
Subject   Re: st: instrumental variable for quantile regression
Date   Fri, 23 May 2008 12:14:10 +0200

Dear Austin and Brian.

Thanks first for your answers. Austin I probably missunderstood the
answer of Brian in that previous message suggesting such a procedure,
I thought it was to solve this kind of problem. Anyway I should be
more explicity about the data I use. I have individual panel data
(with wages and a set of individual characteristics) and I matched
them with provincial variables whose I would like to estimate the
impact on differnet points of workers wage distribution.

Hence I have two problems I'd like to solve.
- The first is that because of likely endogeneity I would like to
instrument my provincial variable and perform a quantile regression.
This is the reason why of my post. I thought that it could have been
possible to run a first stage and then correcting the standard errors
performing the qreg estimation. But it is not correct. Thank for this

- The second. I would like to correct for individual unoberved
heterogeneity and hence to see if it is possible to estimate  a
quantile fixed effects estimation. I know that also this question is
not easy to solve. I red another stata faq that suggests to use
xtdata, fe and then apply qreg clustering the standard errors with a
general sandwich formula (?!?). The last point I did not well
understood, and also qreg does not allow clustering options. If you
have something to advice also on this topic, many thanks. Below the
link of the related stata faq.

thank you again

2008/5/22 Austin Nichols <>:
> alessia matano <>:
> The ref cited elliptically in
> is
> Takeshi Amemiya
> "Two Stage Least Absolute Deviations Estimators"
> Econometrica, Vol. 50, No. 3 (May, 1982), pp. 689-711
> which proves consistency of that model (2SLAD) for the structural
> parameter \beta which determines the conditional mean of y = X\beta.
> That's the conditional mean, not median. I.e. not the usual
> interpretation of LAD in terms of quantile regression; see e.g.
> Koenker and Hallock 2001:
> If you have survey data, read Stas K.'s refs in
> So I guess I need more info from you as to what data you've got and
> what you want to estimate before I make any claims about
> consistency...
> On Thu, May 22, 2008 at 12:17 PM, alessia matano <> wrote:
>> Dear Austin,
>> first thanks for your answer. I also found some of these articles to
>> read that could be useful, and I will do that. of course. I also found
>> out an answer in an old stata faq about the same problem where a guy
>> was suggesting the procedure below. What do you think about it?
>> sysuse auto, clear
>>        program bootit
>>                version 8.0
>>                // Stage 1
>>                regress price foreign weight length
>>                predict double phat, xb
>>                // Stage 2
>>                qreg mpg foreign phat
>>        end
>>        bootstrap "bootit" _b, reps(1000) dots
>> thank you
>> alessia
> *
> *   For searches and help try:
> *
> *
> *
*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index