The RATS Software Forum

Is VAR analysis with 30 data points possible. Specifically, sign-restriction analysis with 7 variables, 6 of which are restricted?

With a one-lag VAR, you could do it. Two lags would come pretty close to running you out of degrees of freedom for the covariance matrix.

Tom,

To get more observations I switched from quarterly to monthly data. However, running the sign-restricted VAR with monthly data leads to very large percent changes; instead of a .04 change, now I see changes of 20 or 30. Any idea what could be driving this? Is it just of function of rescaling my data? I wouldn't think that should be an issue.

FYI - I'm using the code from Uhling 2005 to look at the effect of monetary policy on employment.

Thanks.

How did you "rescale the data"? Yes, changing the scale of the data changes the scale of the responses.

Thanks. The scaling of the GDP deflator varied substantially.

I'm confused. The scaling of a variable should be up to you. How does it "vary considerably"?

With the quarterly data I used data on GDP and the GDP deflator from FRED, but with the monthly data I found a monthly GDP measure and I computed the GDP delfator as (Nominal/Real)*100. For the monthly data I had values like 5 and 6, but for the quarterly data the measures were 100, 102, etc. So I scrapped the GDP measure that I computed and just used monthly CPI data from FRED.

If you're going to convert those to logs anyway (you are, aren't you??), it doesn't matter---the scale of a price index is arbitrary.

Yes, I did convert them to logs.