question about %dlminit

[tclark]

For use by others, I am porting some RATS estimation code of mine to other software. As part of that, I would like to be able to replicate what %psdinit and %dliminit do if the largest root of the coefficient matrix implies a non-stationary system. I am using %psdinit and the companion form of the VAR to compute the steady state variance, which I need for another purpose. In the very occasional draw in which the coefficient matrix has a largest root of 1 or more, %psdinit yields a solution that matches up to the first matrix of %dlminit. I know that solution is intended to be appropriate with the explosive root. Can you point to a textbook or paper reference for the solution that %dlminit(1) and %psdinit are providing in this case? I can't find any such additional information in the forum or the supporting RATS documentation. Thanks much.

Todd

[TomDoan]

That's in

https://estima.com/articles/TP2010-1%20 ... namics.pdf

Note that it depends heavily on the QZ decomposition which, if the target application doesn't include it, is REALLY long. (The C code is about 15000 lines).

[tclark]

Thanks very much