Autoregressive ordered probit in Stan

80 views

Skip to first unread message

Oskar Knapik

unread,

Dec 5, 2015, 8:31:20 AM12/5/15

to Stan users mailing list

Dear all

I am new user of Stan. I wonder if it is possible to implement autoregressive ordered probit model as it is in https://mediatum.ub.tum.de/doc/1097570/file.pdf

Any help will be appreciated.

All the best,

Oskar

Ben Goodrich

unread,

Dec 5, 2015, 10:10:48 AM12/5/15

to Stan users mailing list

On Saturday, December 5, 2015 at 8:31:20 AM UTC-5, Oskar Knapik wrote:

I am new user of Stan. I wonder if it is possible to implement autoregressive ordered probit model as it is in https://mediatum.ub.tum.de/doc/1097570/file.pdf

Yes, such a model can be expressed in the Stan language. However,

It is tedious to specify the constraints on the latent variables, especially when there are more than two categories
Even when you do that, it is very difficult to get Stan to sample efficiently due to the dependence of the latent variables on the unknown common parameters

At least for the application in that paper, I would question whether the ordered probit likelihood is appropriate. Although it is true that the prices can only change by discrete amounts, it is still a ratio-scale variable rather than something that is merely ordinal.