The RATS Software Forum

Is it possible to estimate a VAR with with breaks in the sense that the coefficients on the lagged values of the independent variables are different for different time periods? This would be similar to the Bai Perron estimator for a linear regression with breaks. I was considering writing up each equation as a linear regression equation using the Bai Perron procedure but I thought I would check if there is a more direct way to do this.

If you do Bai-Perron equation by equation, you'll almost certainly get different break points for each, which isn't what you want.

Example lutkp608.rpf from the Lutkepohl textbook replications does full breaks in a VAR. Note that there is almost no chance that any formal test of a full coefficient break in a VAR would be significant just because of the sheer number of added parameters in any but the smaller models.