The RATS Software Forum

Hi there!

I have one question with regard to the small sample critical values for the OIR restriction test in the SVAR model. I use Rats v 7.2. I estimated a SVAR model and obtained a group of most sensible estimates. However, the results cannot pass the Overidentifying restrictions test. I think maybe it's due to the small sample (129 data points). Could you please tell me if there is any program procedure to do a bootstrapping exercise of the SVAR model to obtain the small sample critical values for the OIR test?

input code, data, and sensible estimates are attached

Best Regards

Sure. Just use the VARBootsetup procedures to bootstrap the VAR, and re-do the estimates of your SVAR inside the draw loop. However, given how strongly your restrictions are rejected, I wouldn't expect that you would get a qualitatively different result.

"Sensible" parameter values aren't particularly important if it's for a restricted model which is rejected.

Hi, Tom!

I tried editing the code to bootstrap the VAR. My original idea is to bootstrap the VAR 1000 times, in every draw I estimate the SVAR model and collect the computed Chi-squared(7) values and store in something like
dim %%responses(draw)(nvar*nshocks,NSTEPS)
ewise %%responses(draw)(i,j)=ix=%vec(%xt(decomp,j)),ix(i)

After I have 1000 such computed Chi-squared(7) values, then I revise procedure @mcgraphirf to let it report the upper and lower 5 or 10 critical values for the Chi-squared(7) values. But I'm completely stuck on how to let the RATs give the computed Chi-squared(7) values from the CVModel instruction and the remaining steps. There is no input options in CVModel instruction to give the computed Chi-squared(7) values so that I can collect them by doing the draws. Could you please help me to edit the remaining codes?

code is attached

Best Regards

That's a rather clumsy way to go about that. Just save the test statistics and use SSTATS to determine how many would reject. This adds the bookkeeping for that to your program. It also saves the original CVMODEL estimates and re-initializes the parameters with each set of draws---as you had it written, the results from one draw were used as the guesses for the next which will cause problems if one of the draws has a hard time fitting the model.

Hi, Tom!
Thank you so much for your help! I really appreciate that!
The code runs and stops until it gives the structural parameter estimates. Because I use RATs v 7.2 in which there is no smoothing option for density instruction. Then I had to delete “smoothing=1.5”. It can give the result “Bootstrap p-value of model 0.02700”. This p-value is much higher than the non-bootstrap p-value. Could you please tell me if there is any way to compensate losing “smoothing=1.5” option in the code bootstraptest.rpf with RATs v 7.2? If you run it with RATs v 8.2, can the code with “smoothing=1.5” go through? If so, could you please tell me what the highest bootstrap p-value you can obtain?

Best Regards

If you've found a reference as to how to do a bootstrap adjustment for a test for an implicit restriction like this, please let me know.

Hi Tom,

There is a minor error in the code bootstraptest.rpf. There is something missing in the following:
"system eqsample
variables resample
lags 1 to lags
det constant
end(system)"

"d2exchr{1 to lags}" should be added in line 4 after "constant". Otherwise, the two systems have different specifications. Is my thought right?

Best Regards

True. Unfortunately, there's no simple fix since d2exchr is endogenous. I have a question about how you're implementing this.

You're defining

set d2exchr = d2*exchr

with

set d2 = t>=2005:7

And you define the system as:

SYSTEM eqbase
variables money intr p_ppi p_rpi rgdp pcom exchr
lags 1 to lags
kfset xxx
det constant d2exchr{1 to lags}
End(system)

With that definition, the effect of the D2 dummy is gradual, with all but lag one zeroed out in 2005:8, all but one and two zeroed out in 2005:9, etc. From the description in the comments, it sounds like you wanted a once-and-for-all break to the specification in 2005:7.

Dear Tom,

Thank you very much for your reply! Ok, I see. Regarding the endogenous issue with "d2*exchr", I hope there will be solutions using Rats program implementing the ideas in the second last line of "step 4" part in p. 236 and endnote 2 in p. 238 from the Book Applied econometric time series (2nd edi.) by Walter Enders.

For dummy variable d2, the effect is gradual since the dummy variable is entered in interactive terms. The six dummies variables (there are 6 lags of exchr) are included to account for the start of a different exchange rate regime in 2005:7, which follows from "Tang, H. C. 2006. The relative importance of monetary policy transmission channels in Malaysia. CAMA Working Papers No. 2006-23, Australian National University." (see p. 9 bottom). The different thing is that he uses the interative dummies to account for the pegging of the ringgit to the US dollar, while I use them to account for the start of a new Renminbi exchange rate regime.