I have a question regarding the number reported as log data density
in DYNARE’s Bayesian estimation. Can we interpret this number as
marginal likelihood?
If so, shouldn’t the number of log data density be negative, rather
than positive? DYNARE always reports the positive number of
log data density and I was not sure how to interpret this number.
I very much appreciate your guidance in this regard. Thank you.
If you are replicating the study of Adolfson and your getting positive values, you are still doing something wrong. However, if you are estimating a different model or you are using a different data set or sample, the marginal likelihood will be different and may even be positive. The reason is that the marginal likelihood may be larger than 1 so that its log is positive (Think of a normal distribution with infinitesimally small variance. In this case the density around the mean will go to infinity. Only for discrete distributions is the marginal density always negative.)