MSSYSREGRESSION—Markov switching systems regression

[TomDoan]

@MSSysRegression is a file with the support procedures for Markov switching multivariate linear regressions (same right hand side variables) with either the full coefficient vector switching or part of the coefficient vector switching and part fixed. The covariance matrices can either be switching or fixed. If you need Hamilton-style switching means, use @MSVARSetup instead.

mssysregression.src

Detailed description

[aloprofundo]

SWITCH=[C]/CH/H
C means that coefficients are switching, H indicates variances are switching.
NFIX=# of explanatory variables which are fixed across regimes [0 or all]
If SWITCH=C or CH, the default is 0, if SWITCH=H, it's all. The fixed coefficients must be placed first in the supplementary card for regressors.

Hello Tom,

I downloaded the RATS program to replicate results from Ehrmann, Ellison, Valla (2003) and tried to run eev_mcmc. The software stops running and I have to restart it whenever I try to set "switch" to be C, however everything works perfectly when "switch" is set to be CH. I would like to estimate the model assuming the covariance matrix to be state-independent, could you please tell me how to do so? Thanks a lot

p.s.
here is where I downloaded the program:
http://ideas.repec.org/c/boc/bocode/rtz00177.html

[TomDoan]

The whole point of the EEV paper is to distinguish variance regimes so everything about the program is aimed at that. The maximum likelihood and EM programs are much easier to modify to allow for H not switching (change sigmav to sigma in the non-linear parameter set). However, the MCMC requires a completely different method of drawing the covariance matrix, and a completely different method of "labeling" the regimes.