Wrong variance in dynare 4?


I noticed that in a model of mine (here attached) the variance of the variables is different from that obtained with Dynare 3 (this being correct)

In particular you can see that the variance of the gov_shock_us ( computed as std_gov_shock_us^2/(1-rho_eps_g_s^2)) is very different from the one returned by Dynare_v4.

This is not always the case (in a smaller model with only 4 shocks things worked ok).

Does anybody know of a bug in the variances of Dynare 4?
(or does anybody spot a mistake I made)


two_country_bond_imp_problem.mod (24.2 KB)


It seems that Dynare_v4 returns messy results when there are redundant stochastic processes. In the model I posted earlier, there are a few innovations (and shocks) that (in that specification) are not active.

If you eliminate those shocks the problem disappears.

Importantly, if you run a stochastic simulation in Dynare_v3, the sample moments are also messy if you leave the redundant innovations.

Obviously it would be nice if this didn’t happen. Meanwhile… stay clean and avoid redundant stochastic processes.