It is indeed necessary that (1-mu)*mu > sigma^2 otherwise the natural parameters of the beta are negative and the density doesn't exist. I added the necessary test.1. When using a beta prior with mean 0.2 and variance 0.5 as in the attached mod-file Dynare warns that the prior has two modes. Nevertheless it continues but then is unable to compute the prior distribution (everything is NaN). Shouldn’t we abort with an error message in this case?