Let’s say we need to use 1 regression equation (Linear regression) and 1 classification (probit classification). Is it possible to correlated errors of these 2 different models. I was thinking on a multivariate normal, but it might not be the best idea.
This WIP notebook may give you other ideas: Bayesian copula estimation example notebook by drbenvincent · Pull Request #257 · pymc-devs/pymc-examples · GitHub
1 Like
Without commenting on the method itself, nothing stops you from stacking the means of two GLMs, adding a covariance matrix, and estimating them jointly as a Multivariate Normal. Seems like in the worst case you will just get a diagonal covariance matrix (ok ok the real worst case is infinity divergences)
2 Likes