Bayesian of truncated regression

[BinhPham]

Hi Tom,

Could I have some information about Bayesian truncated regression? How do we set priors for this kind of estimator? My case is the dependent is continuous in [0,1] so that truncated or tobit is a good candidate.

Thanks in advance,

[TomDoan]

Prior on what? The usual way of handling this is to have a y*=xb+u latent process, with y=y* truncated to some interval. If that's appropriate, the parameters are the b and the variance of u, both of which would have priors just like they would if you could observe y*. This topic is covered in the Bayesian Econometrics e-course.

If the dependent variance is mapped to [0,1] for some reason other than truncation, you would have to figure out a different way to model it (such as transforming the dependent variable).