Thanks a lot for your time and patience! I have estimated the SVAR with the
variable of interest being exogenous and I have the results I was searching
for. I also found other papers which indeed do this exercise instead of
2013/6/5 Sven Schreiber <svetosch(a)gmx.net>
Well yes, that was my first suggestion. That's the difference
"doesn't react" or "doesn't move (at all)".
But for the IRFs it shouldn't make a difference whether you put y2 as
exogenous or if you explicitly estimate this restricted structure with a
univariate y2 equation. Assuming you don't care about the shock in this y2
equation, only about the shocks in y1 and y3.
Where it would make a difference AFAICS is if either you want to do actual
forecasts, or perhaps if you want to get confidence bands with parameter
uncertainty. [Without further thinking I'm not 100% sure about this latter
point, however, but it seems plausible.]
*Gesendet:* Dienstag, 04. Juni 2013 um 20:10 Uhr
*Von:* "Gabriela Nodari" <gabriela.nodari(a)gmail.com>
*An:* "Gretl list" <gretl-users(a)lists.wfu.edu>
*Betreff:* Re: [Gretl-users] SVAR coefficients
Ok, I will try to estimate the SVAR with x_1 as exogenous, I am also
quite sure that the SVAR package allows it. But I am getting a little but
Actually I am trying to follow a paper where the authors do this exercise,
Yt=(y1, y2, y3).
They shock y1, and they check the effects on y2 and y3. As for y3, they
say that there is a direct effect, and an indirect effect (via y2). To
isolate these 2 effects, what they do is:
"to restrict the coefficients of the underlying VAR in such a way as to
force the response of y2 to a shock in y1 to be zero [...] we postulate a
different (restricted) economic structure, i.e. y2 is structurally not
allowed to respond to y1 and y3 shocks [...]. A necessary condition is to
set the impact reaction =0. This plus restricting the AR coefficients on
lagged y1 and y3 in the y2 equation will be sufficient for imposing that y2
does not react to a y1 shock at any horizon"
It seems that your first suggestion was the way to replicate what they do,
isn't it? But I don't know how to do it in gretl!
2013/6/4 Sven Schreiber <svetosch(a)gmx.net>
> Am 04.06.2013 16:20, schrieb Gabriela Nodari:
> > Dear Sven,
> > Thanks for your timely answer! I guess you understood right. But let me
> > try to be more precise.
> > I want to check the effects of a shock within the VAR by restricting the
> > x_1 variable to not move.
> Well, if a system variable is not endogenously responding, we typically
> call it exogenous instead. To me it seems that that's what you
> (implicitly) want to do: create a scenario where x_1 is treated as
> My first suggestion was to implement that by changing the x_1 equation
> to a univariate equation only including lagged values of x_1. But if you
> say "x_1 shouldn't change" that would probably not be the correct
> implementation: Instead you would replace the x_1 equation by some
> totally pre-specified scenario path, presumably a constant value. (If
> you don't do actual forecasting but just IRFs, then I guess you don't
> really have to specify these paths.)
> I'm not up to date on the capabilities of the SVAR package,
> unfortunately -- wait, I'm actually curious myself, so I'll check the
> docs... and in the help file it says that that you can specify a list of
> exogenous variables, so it seems you should be alright.
> Gretl-users mailing list
Gretl-users mailing list