Why the posterior moments of the parameters are different (using the same code) in Dynare 4.4.3 compared to Dynare 4.5.0 and 4.5.1?
data.mat (14.9 KB)
example_stoch_est.mod (834 Bytes)
Dear Fabio,
Can you post the log files obtained with 4.5.1 and 4.4.3. Dynare, on each run, generates one main log file which keep track of all the informations printed in the Matlab command window. This log file is called <NAME_OF_THE_MODE_FILE>.log. If a metropolis is run, there is another log file describing the progress of the Metropolis-Hastings algorithm. This file is saved in a subfolder with the name of the mod file: NAME_OF_THE_MOD_FILE/metropolis/metropolis.log
Best,
Stéphane.
The reason is that you consistently ignored the warning
SOLVE: maxit has been reached
in Dynare 4.4.3.
In Dynare 4.5 we increased the default maxit. As a consequence, the steady state can be computed for all parameter draws and for example the computed mode between versions is very different. In 4.4.3 you discarded those parameters because the steady state could not be computed.
Also we changed the algorithm used to solve for the steady state (see the release notes for 4.5.0 here). Something still worries me, the optimization problem with 4.4.x is more constrained, the contraint is not failing to find the steady state, hence I would have expected to see a smaller posterior density (at the estimated mode) with 4.4.x than with 4.5.x… And it’s not the case…
Best,
Stéphane.
@stepan-a What do you mean? With 4.4.3, we have
Objective function at mode: 369.035595
with 4.5
Final value of minus the log posterior (or likelihood):351.794900
As we are looking at minus the posterior, the smaller the better, which is the case.
Ah… Sorry I misread the numbers. Then it’s okay.
Best,
Stéphane.
Ok, thanks for your helping.