Ok, thank you for clarification. I have another question then: Since Gretl has an option to bootstrap the confidence interval, does that mean that the impulse-response functions are valid even though the residuals of each VEC equation are non-normal it their distribution? I know the t- and p-values of the VEC output are unreliable with non-normal residual distribution, but doesn't bootstrapping remedy this, since it relies on the "true probability distribution"? Forgive me if I got it totally wrong, I'm only a beginner at this.

Best regards,

Lars

2017-08-18 15:16 GMT+02:00 Allin Cottrell <cottrell@wfu.edu>:

On Fri, 18 Aug 2017, lasses skola wrote:

Hi, and thank you for your response! I am not using any addons to my
knowledge. I was a big vague maybe. There is an option to calculate
confidence intervals of the IRFs using bootstrapping.
I think I put the question in the wrong way as well. What unit is the shock
in? One standard deviation? Or SE of regression?

You've had a reply from Sven, but I believe he's thinking about the SVAR addon rather than the behavior of the built-in "var" and "vecm" commands.

For the built-in commands, the shock is one standard deviation of the innovations of the variable in question. Here's a trivial example (using "var" rather than "vecm" but that doesn't make a substantive difference):

<hansl>
open data9-7
var 4 UNEMP PRIME --quiet
matrix IRF = irf(1,1)
print IRF
matrix S = $sigma # covariance matrix of innovations
printf "std deviation of UNEMP innovation: %.5f\n", sqrt(S[1,1])
</hansl>

If you run the above you should see that the IRF vector begins:

0.24541
0.29007
0.27650 ...

while the printf on the last line shows

"std deviation of UNEMP innovation: 0.24541"

hence confirming that the shock is one standard deviation (and the response is in the "natural" units of the target variable).

Allin Cottrell

_______________________________________________
Gretl-users mailing list
Gretl-users@lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users