I have tried both stable (4.4.3) and unstable (2016-04-30) version of Dynare for the same model and code, however, their results are different in terms of the mode of some parameters, variance decomposition and historical shock decomposition. So in this case, which version is reliable and which one I should use?
Thanks a lot!
If the mode is different, everything else will be different as well. The reported mode being different means that the two version found different (local) modes. You should discard the one with the lower posterior density and take the one with the higher posterior density. In general, I would try to start a different (global) mode finder at the highest mode just to test whether there are other modes nearby and whether you most likely found the global one.