R squared for VAR -Estimation by cointegrated least square

[Nadiarock1]

Hi,

I would like to have an advice for getting the R-Squared from a VAR/system - Estimation by Cointegrated least Square.
1. I have estimated a Fm-ols and I saved the coefficients (long run)
2. I use these coefficients in a ECM models to get the short-run parameters. .

But, I got a problem with the variables I specified and I would like to have the R-Square parameter to insure me that my equation system is ok.

How can I get this R squared in a system by cointegrated least squares.

Thank you.
Nadia

[moderator]

What formula for the R-square did you have in mind?

Regards,
Tom Maycock

[Nadiarock1]

Hi,

I would like to use the Ajusted R-squared.

I have modified some of the programs I used to incorporate this code (%RBARSQ ) after linreg and it worked.
The next step is to get a R-squared with the VAR/Cointegrated.

Here is my code for the VAR/cointegrated. If I had a @source, I would go it that file and add the command for R-Squared, but with the command "system" I don't know how to use it. If I add it at the end of the following code, I got only 1 R-squared for the entire VAR/cointegrated system.

system(model=ectmodel)
variables y1 y2 x1
lags nlags
det constant tariff d import gdp
ect producersupport
end(system)
estimate

Thank you very much, Nad

[TomDoan]

Hi,

I would like to use the Ajusted R-squared.

I have modified some of the programs I used to incorporate this code (%RBARSQ ) after linreg and it worked.
The next step is to get a R-squared with the VAR/Cointegrated.

Here is my code for the VAR/cointegrated. If I had a @source, I would go it that file and add the command for R-Squared, but with the command "system" I don't know how to use it. If I add it at the end of the following code, I got only 1 R-squared for the entire VAR/cointegrated system.

system(model=ectmodel)
variables y1 y2 x1
lags nlags
det constant tariff d import gdp
ect producersupport
end(system)
estimate

Thank you very much, Nad

I think you want that to be

Code: Select all

lags 1 to nlags

If you add this after your code above, you'll get a vector of the standard centered R-squareds:

Code: Select all

compute varsigma=%sigma
vcv(center,matrix=basesigma) %regstart() %regend()
# dm dy
compute rsquared=1.-%xdiag(varsigma)./%xdiag(basesigma)

The R-bar squared is computed with:

Code: Select all

compute rsquared=1.-(%nobs-1.0)/(%nobs-%nreg)*%xdiag(varsigma)./%xdiag(basesigma)

Is there are particular reason you wanted this? It's not included because it's almost never reported. With non-stationary data, they're generally numbers like .99999.

[Nadiarock1]

Hello,
I want to get the R-squared from my VAR/Cointegrated because I have a problem with the signifiance of the ECM for the sort run and the Error correction term. Basically, I have unit roots and cointegration for my variables in my database. I estimate a fmols with good significance with expected coefficient signs and a quite good R².

Then, I want to use these long run coefficients previously estimated with the fmols for the long run and use it with the ECM. When I get the VAR/cointegrated of my systems, the sort run as well as the error correction term become non-significant. As you has written, the r-square is 0.99 in my system...hmmm. Also, there is no multicollinearity..

Do I have to difference my exogeneous variables that have unit root? By definition, the endogenous are differenced in the VAR/Cointegrated. I try to find out why I get the results.

Thanks again.

Nadia

[TomDoan]

If you're using SYSTEM with ECT, then no, you don't difference the variables. The inclusion of the proper forms for the endogenous variables is handled automatically when you do an ESTIMATE.