Page 1 of 1
Probable bug with Markov switching transition matrix
Posted: Thu May 19, 2016 10:09 am
by startz
There may be a bug when transition probabilities in a Markov switching model are restricted to one. In the attached file, equation ChinaAR essentially "doesn't work" with transition probabilities for state 2 restricted to 0,1. If the restriction is changed to 0.001,0.999 it works fine.
Re: Probable bug with Markov switching transition matrix
Posted: Fri May 20, 2016 11:49 am
by EViews Glenn
Not yet convinced it's a bug :) I get singularity for these data with 1e-10 and 1-1e-10.
Re: Probable bug with Markov switching transition matrix
Posted: Fri May 20, 2016 11:59 am
by startz
That's reasonable evidence. I still wonder whether there's an overflow or something internally that might be catchable.
Re: Probable bug with Markov switching transition matrix
Posted: Fri May 20, 2016 5:05 pm
by EViews Glenn
Worth taking a look when I get a chance.