SVAR Montesvar command

[tahir]

Dear All
After estimation of structural VAR in the following way
dec frml[rect] afrml
nonlin g25 g34 g42 ...
frml afrml = ||1.0 ...
....
compute g25=g34=g42=........=0.0001
cvmodel(a=afrml,...... ) %sigma

Query 1. Is the following command for the computation of Structural Variance decomposition
and structural IRF.
errors(impulses,model=modelname,factor=afactor) 4 24 ;because factor=afactor is being used?

Query 2. Is the following command for Structural Impulse Response graphs.
@VARIRF(factor=afactor,model=brazilmodel,steps=24,varlabels=implabel,page=byshocks) because we use factor=afactor (of course without C.I bands)

Query 3. when I use MONTESVAR.RPF for Impulse Responses, despite required changes it gives me the shock window with 7 rows but 3 columns where I need 7 rows 7 columns. To my understanding I changed dim %%responses(draw) ...command replacing 3 with 7 but again "shock window" gives 7 rows 3 columns? where I should make changes.

Thanking you in advance.

[TomDoan]

Assuming that you have the option FACTOR=AFACTOR on your CVMODEL instruction, the first two are correct.

Regarding MONTESVAR, the program as written is only focusing on three shocks:

Code: Select all

   dim %%responses(draw)(nvar*3,nstep)
   compute weights(draw)=weight
   ewise %%responses(draw)(i,j)=xt=%xt(impulses,j),ix=%vec(%xcol(xt,3)~%xcol(xt,5)~%xcol(xt,6)),ix(i)

To get the whole set of shocks, you would need

Code: Select all

   dim %%responses(draw)(nvar*nvar,nstep)
   compute weights(draw)=weight
   ewise %%responses(draw)(i,j)=ix=%vec(%xt(impulses,j)),ix(i)


[tahir]

Dear Sir

Thank you very much for the response, it gives 7x7 graph now.
But I have an other Query
Montesvar command always gives me the same(identical) Impulse responses (IRs) even if I change the identification scheme. For example, estimating two identification schemes, one with nfree=13 the other with nfree=18, The following command
@varirf(factor=afactor, ......) gives different IRs (as it should) but Montesvar always gives the same IRs whatever the identification scheme I use.
I also tried typing the whole command myself instead of copying it but the results do not change.
Query: Why Montesvar command gives me the same IRs for any identification scheme, and how to get distinct IRs for each identification scheme?

Thanking you in advance.

[TomDoan]

You would have to post your program. That shouldn't happen if the program is adapted correctly.

[tahir]

Dear Sir
I adopt the following procedure,
First I estimate the coefficients with CV model, then I put that identification in the montesvar model.
The first case gives me coefficeints, loglikelihood, chisquare and significance level, the second case does not give significance level.
Please see the attachments;
The file IRF vs Montesvar contains IRs and you can see in case of montesvar they are identical.
The second file named montesvar 2 contains the adopted montesvar programme.

Query 1. How to change montesvar to get different IRs for each different Indetification scheme.
Query 2. Why the montesvar does not give significance level (is it the reason that it gives convergence to function value not convergence of coefficients)?

Thanking you in advance.

[TomDoan]

The Monte Carlo graphs are definitely not the same. They're similar (which you would hope would be the case), but not identical. The VARIRF graphs both look wrong (neither look compatible with the MC graphs), so you're probably doing something wrong with the use of that.

[tahir]

Dear Sir

Thank you very much for your response.
Would you please just confirm the procedure that I adopt: First I estimate the coefficients with CV model, then I put that identification scheme in the montesvar model.

Query; Is this procedure right one?

Thank you very much in advance.

[TomDoan]

That's correct. But if you do VARIRF, it should be done right after the CVMODEL instruction. Otherwise, it will use the coefficients off one of the draws rather than the original OLS coefficients.

[tahir]

Dear Sir

Thank you very much for the response.

My query is the Confidence Interval Bands that MONTESVAR command gives, Is it 95 Percent C.I.?

Thanking you in advance.

[TomDoan]

Dear Sir

Thank you very much for the response.

My query is the Confidence Interval Bands that MONTESVAR command gives, Is it 95 Percent C.I.?

Thanking you in advance.

No. The default is 16% to 84% (roughly one standard deviation if the spread were normal). If you want something different, you'll need to change the options on the MCGRAPHIRF.

[tahir]

Dear Sir
Thanks for your reply I have some other queries
I would like to know your opinion about lag order. I have monthly data ( 132 observations), estimating SVAR (7 variables; 7x7), likelihood ration test recommends 6 lags, Final Predictor Errors test (FPE) and Akaike Information Criterion (AIC) suggest 3 lags, whereas Schwartz Information Criterion (SIC) and Hannan-Quinn Information Cretion (HQ) propose 2 lags.
Query 1: what criterion should I follow? If I use four lags (that I tend to) and check the structural residuals, if they are normal, will it be sufficient to claim that results are robust?
Query 2: The shock in RATS (by default) is one standard deviation shock?

Thanking you in advance.

[TomDoan]

Dear Sir
Thanks for your reply I have some other queries
I would like to know your opinion about lag order. I have monthly data ( 132 observations), estimating SVAR (7 variables; 7x7), likelihood ration test recommends 6 lags, Final Predictor Errors test (FPE) and Akaike Information Criterion (AIC) suggest 3 lags, whereas Schwartz Information Criterion (SIC) and Hannan-Quinn Information Cretion (HQ) propose 2 lags.
Query 1: what criterion should I follow? If I use four lags (that I tend to) and check the structural residuals, if they are normal, will it be sufficient to claim that results are robust?

6 lags is too many with that much data. 4 is probably OK. Normally distributed garbage is still garbage---the point of lag choice is not to reduce the residuals to a normal distribution, but to eliminate (correctable) serial correlation. The results will be robust to lag choice if they are .... robust to lag choice. If you get similar results with 2, 3 and 4, then you can say they are robust to lag choice.

Query 2: The shock in RATS (by default) is one standard deviation shock?

Yes. The default shock size in IMPULSE is one standard deviation.

[tahir]

Dear Sir
If the distance of upper and lower band of confidence interval from the impulse response is not same (spread is not normal) what could be the reason and how to solve the problem. (case under discussion is SVAR).

Thanking you in advance.

[TomDoan]

There's nothing wrong with that. In fact, that's expected, because the IRF's are a highly non-linear function of the coefficients. So even if the VAR coefficients are Normal, the IRF's won't be. See Sims and Zha(1999), "Error Bands for Impulse Responses", Econometrica, vol 67, no. 5, pp 1113-1156.

[FRegensburg]

Dear Mr Doan,

I seem to understand the MONTESVAR procedure is designed for over-identified vars- can it be employed for just-identified vars as well?

If so, how would I go about adapting a model such as the one attached (which is a cvmodel with an A and B matrix rather than just an A matrix, as in the example program)? when I add bfrml at the end of the cvmodel line in the example montesvar program I get a sintax error message, but clearly the rest of the code needs to be modified as well for the B matrix, the question is how...

Secondly, if I wanted to use the MONTEVAR procedure instead, does it suffice to run it after I estimate the cv model (as in the second attachment), I'm asking because when I do so, the point estimates of the IRFs are clearly wrong (they obviously tell a completely different story than the ones I get from using errors and then varif).

I look forward to hear from you.