Switiching Model Workbook, Example 11.3

[tflavin71]

Dear Tom

I am using a MS-VAR, estimated by MCMC, as in Example 11.3 of the workbook. A referee has asked for a time-varying transition probability matrix where the p's depend on some variable(s). Is it possible to do adapt this example to include this possibility?

Thank you in advance for any insight / help you can offer.

Best wishes.

[TomDoan]

Sure. The Filardo example (or the recreation of the Filardo example) uses TVTP. Note, however, that the tendency has been for people to use transition variables which are not really exogenous, but in fact very closely related to the endogenous variables in the model. (Filardo, for instance, used leading indicators). When you do that, it's really easy for MS-TVTP to simply pick up some other type of neglected non-linearity in the model.

[tflavin71]

Thanks Tom for your quick reply and advice.

I found the Filardo replication example but I am trying to figure out how to code it in the MCMC estimation. Just wondering if I can add some code to example 11.3 or if I have to modify the procedure (MSSETUP.src) in some way?

[TomDoan]

The part that needs to change is drawing the transition probabilities given the sampled regimes. Instead of using @MSDRAWP to draw from a Dirichlet (beta for two regimes), you're drawing the slope parameters on a logit. Interestingly, with the TVTP, EM is simpler if you do a logit rather than a probit for the transition function and MCMC is easier if you do a probit rather than a logit. (There's no real difference in direct maximum likelihood).

[tflavin71]

Thanks very much. I appreciate your insights.