Modes differ depending on mode_compute

[Michael Paetz]

Hello Everyone,

I realized today, that using the standard mode computation leads to different results than using, for example, mode_compute = 6 (the MC-based routine). Hence, also the means after the MCMC differ. And this difference is not trivial: For example, a shock contributing round about 2 percent to output volatility in the in the case of using mode_compute = 6 contributes more than 20 percent, when using the standard routine.

I didn't attach a file, since the model is rather big and complicated, but I believe everyone can try this with any model-code.

For my model, the standard routine works better (in terms of fitting the data), but is this a general rule? Is the standard routine better than using the MCMC-based routine? And why do these routines differ at all?

Best,
Michael

[Michael Paetz]

Here's an update:

I used a code that worked fine without any problems for version 4.2.1 in 4.2.2 and the standard mode routine fails. Is this a bug in 4.2.1, which is fixed in 4.2.2 or is this a bug in 4.2.1 ? Again, the results of the standard routine in 4.2.1 gives the best results in terms of fitting the data.

[Michael Paetz]

In 4.2.0 I can use the standard mode computation and get the same results as under 4.2.1.

If I use 4.1.3 the standard computation works, but the results differ from the ones of 4.2.0 and 4.2.1.

I am confused...

[Michael Paetz]

Sorry, I found an installation mistake. Before installing a new Dynare version, I need to delete the old dynare folder...

Now the modes differ only to a very small extent and the same is true for the variance decomposition.

I thought I found a bug in dynare, but the only bug was me. Sorry.