Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

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

Re: Re: st: Negative restrictions in svar

From   Nick Cox <>
Subject   Re: Re: st: Negative restrictions in svar
Date   Tue, 1 Feb 2011 14:22:43 +0000

This question appeared twice. A metaquestion is why no-one replied the
first time. I read it and thought "How can we advise well on this
little information?" and deleted.

It seems to me that the choices are yours, including but not only

1. You have a hypothesis about coefficients. You can leave the model
unconstrained and see if the data agree.

2. Your variables are yoked together by definition so that there is
one fewer genuine variable. (An analogue would be % males and %
females, yoked so that their total is 100.) In that case, it is not
clear that asking for separate impulse response functions makes sense
or, more positively, is needed at all. Perhaps it does, but you should
explain why.

3. Working with the difference y - x does not rule out working with x
and y separately in other problems.

4. There is some science that will answer your question, but even
people in your science can't say what it is, as the question is shorn
of context.

There seems to be a widespread expectation that people who answer
questions on Statalist are supergurus who should be able to tell you
what to do in almost any statistical project. Not so; typically they
just know a bit more than average about Stata,


On Tue, Feb 1, 2011 at 1:33 PM, Sebastian Gomez
<> wrote:
> The problem is that I will later analyze Impulse Response Functions,
> so I need the original variables. Besides, they also appear in some
> other equations.
> On 1 February 2011 09:01, Maarten buis <> wrote:
>> --- On Tue, 1/2/11, Sebastian Gomez wrote:
>>> I will estimate a svar, with short-run constraints, using
>>> acns(). I want to impose the restriction to one coefficient
>>> so it's the negative of other, for example
>>> d1*x - d1*y + d2*z = e
>> I don't know -svar-, but often you can impose such a constraint
>> by creating a new variable:
>> d1*x - d1*y + d2*z = e
>> d1*(x-y) + d2*z = e
>> So if you create a new variable that is x -y and add that
>> new variable instead of x and y. The coefficient of that new
>> variable is the d1 you are looking for.
*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index