Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, is already up and running.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: xtabond with constraints

From   Nick Cox <>
Subject   Re: st: xtabond with constraints
Date   Tue, 9 Aug 2011 07:06:57 +0100

Sorry, nothing from me on that; that is a question for people who do
this kind of thing, but there are plenty of them on this list.


On Tue, Aug 9, 2011 at 1:07 AM, Pedro Ferreira
<> wrote:
> Thanks Nick,
> I have included that approach in my previous message. However, there
> is a dynamic panel nature to what I am trying to do. Using your
> variable names below, in my case y = x(t+1) and I have x(t) on the
> right hand side. I am wondering if you have any insight about what is
> best: xtabond with y_2 as a function of x(t) or reg3 with the
> constraint that the coefficient on x(t-1) should be -1 (using the
> other equations in reg3 to instrument x(t) and x(t-1) with lags)?
> Thanks a lot!
> Pedro
> On Mon, Aug 8, 2011 at 12:51 PM, Nick Cox <> wrote:
>> If the coefficient in x[t-1] must be identically -1 then you can add
>> that variable to the response variable before you fit a model in terms
>> of other  predictors.
>> That is
>> y = <some stuff> -x[t-1] ...
>> is equivalent to
>> y + x[t-1] = <some stuff>
>> so model y_2 = y + x[t-1]
>> Nick
>> On Mon, Aug 8, 2011 at 4:37 PM, Pedro Ferreira
>> <> wrote:
>>> Dear All,
>>> I am running a model of x(t+1) on x(t) and x(t-1) plus other controls.
>>> I need to restrict the coefficient on x(t-1) to be -1. Any suggestions
>>> for how to do this? My understanding is that xtabond2 does not allow
>>> for constraints. I can add x(t+1) with x(t-1) and make this my new
>>> dependent variable and instrument x(t) gmm style with deep lags of x,
>>> namely deeper than 4. Is this the best strategy? Another approach I
>>> have considered is to use 3SLS, first equation is just x(t+1) on x(t)
>>> and x(t-1) and I use the other equations to run regressions of x(t)
>>> and x(t-1) on deep lags. Using reg3 allows me to force the coefficient
>>> on x(t-1) to be equal to -1 in the first equation. Unfortunately, reg3
>>> does not give me the AR tests for how good the lags are as
>>> instruments. Any ideas, thoughts would be highly appreciated. Thanks,

*   For searches and help try:

© Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index