Endogenous prior

Dear Johannes,

With code “endogenous prior” in Bayesian estimation, the model’s second moments can match the data better. However, I find that in terms of model comparison, with or without “endogenous prior” can derive quite different log data density and the model comparison results can even turn over ( model A is better than B with “endogenous prior”, while model B is better than A without “endogenous prior”). Could I ask if this phenomenon makes sense?

Many thanks in advance.


Yes, see

Many thanks!