I use the moments_varendo command in estimation to compute the posterior theoretical moments of my observable variables. Everything works fine, but when I looked in detail at the theoretical posterior variance of some of my observables, I found that the HPD lower bound is always zero. By computing the lower bound myself using oo_.PosteriorTheoreticalMoments.dsge.covariance.density.y_obs.y_obs (or by just looking at the density of the posterior variance), I realized that even though the mean is positive the variance (of the theoretical variance of my observable variable) is so large that the lower percentiles are negative. Dynare seems to check this and thus in oo_.PosteriorTheoreticalMoments.dsge.covariance.deciles.y_obs.y_obs and oo_.PosteriorTheoreticalMoments.dsge.covariance.HPDinf.y_obs.y_obs these values are automatically set to zero.
My question is how to interpret this finding? Is there something wrong in my estimation which results in the large posterior variances (of the variances) and thus the negative values? My expectation would have been to get a posterior distribution for the variance with a positive support (given that the variance is some squared form of the posterior of the parameters and the structural variances). Why is this not the case?
Hints are greatly appreciated!