Dear professors, I estimated a DSGE model in dynare, then the dynare plotted the convergence diagnostic . And my question is that convergence diagnostic is necessary to judge the confidence of the estimation results ? If multivariate convergence diagnostic seems not convergence but the acceptance rate is in 25%-33% , should I believe the estimation results?
No, if the chain has not yet converged to the ergodic distribution, the results are meaningless, regardless of the acceptance rate.
Thanks professor for your reply. The convergence of chain seems like playing tricks, is there any good way to get reliable results?
What do the trace_plot say? Is there a drift in the chain?
Here are my mod file and data file, and trace plot.
The data I used, are took log then used 1st HP filter detrend.
logtest.mod (3.7 KB)
gdata.m (7.3 KB)
block number 1.eps (700.7 KB)
block number 2.eps (699.1 KB)
logtest_mdiag.eps (32.8 KB)
pictures.zip (620.7 KB)
This mostly looks like you may need a lot more draws.
I increased the number of draws to 200000. It seems still not converge.
Here are plots.
mucodiag.eps (33.1 KB)
block 2.eps (278.1 KB)
block 1.eps (278.2 KB)
Now it doesn’t look too bad. I would run chain with 1 million draws.