Page 1 of 1

Markov_switching duration parameter explanation

PostPosted: Tue Nov 12, 2013 5:31 pm
by davidkelley
Would someone be able to briefly explain what the option in the markov_swithching command labeled "duration" actually refers to? Are we somehow restricting the state variable so that a state lasts multiple periods before it is allowed to potentially switch states?

If an answer to the above question doesn't imply that a value of 1 is nonsensical for this duration parameter, would someone also be able to explain why giving a duration of 1 gives the following error:

Code: Select all
CreateMarkovStateVariable_Single(): Error in restrictions
Error in MS-SBVAR MEX file.

Error using mexErrCheck (line 41)
Error encountered in: ms_sbvar_create_init_file.
Error in ms_sbvar_setup (line 443)
mexErrCheck('ms_sbvar_create_init_file',err);
Error in ms_estimation (line 40)
    ms_sbvar_setup(options_);
Error in lag4 (line 66)
[options_, oo_] = ms_estimation(M_, options_, oo_);
Error in dynare (line 120)
evalin('base',fname) ;

Thanks in advance. I apologize if I've missed this question already being answered elsewhere.

Re: Markov_switching duration parameter explanation

PostPosted: Wed Nov 13, 2013 9:58 pm
by davidkelley
After spending the requisite time with SWZ, it is apparent to me that the parameter refers to the priors on the Markov chain and not a restriction.

Might I recommend substantial edits to the reference manual, seeing as it doesn't mention that this argument has anything to do with priors? Also, it is giving the syntax incorrectly (using the old command "states" instead of "regimes").

Re: Markov_switching duration parameter explanation

PostPosted: Thu Nov 14, 2013 9:36 am
by jpfeifer
Thanks a lot for pointing this out. We will deal with it.

Re: Markov_switching duration parameter explanation

PostPosted: Thu Nov 14, 2013 5:58 pm
by davidkelley
Thank you.

In the meantime, is there still a way to specify a prior with different durations for each state of a Markov chain? Or was that functionality lost when the 'state' command was dropped?