Regarding Marginal data densities

Dear all,
I have a question regarding marginal data densities of DSGE model. When using Dynare_v3, sometimes I end up with positive log(P(Y|model_i)) and sometimes negtive log(p(Y|model)). I am wondering if the positive log marginal data densities is always the one we prefer and is it possible to compare a positive log(p(Y|model)) to the negtive log(p(Y|model)? Thanks!

Keqiang

hi i think we should understand marginal densities in a relative sense. the number itself does not mean much, unlike the R^2 in OLS.
you can use it to compare 2 competing models that use the same data series.

thanks for your replay!