Historical and smoothed variables

Hi Johannes,

I am learning how to use Dynare to do Bayesian estimation with a manually linearized symmetric two-country model. I followed your paper and specified four observables and four observation equations accordingly. During my estimation process, I run into several problems.

  1. With mh_replic=0, Dynare reports the “Historical and smoothed variables” for all four observables. However, only 3 of them are perfectly matched (figure attached). The last one seems to be poorly matched (the same problem arises with mh_replic>0). But there is no measurement errors specified in the model. This is the case for both version 4.4.3 and unstable version.

  2. When using the unstable version, I also got the following warning:

Warning: Your prior allows for correlations between measurement errors larger than ±1 and will not
integrate to 1 due to truncation. Please change your prior

In initial_estimation_checks at 110
In dynare_estimation_1 at 157
In dynare_estimation at 105
In example at 547
In dynare at 223

This warning did not arise in version 4.4.3. Again, I did not specify any measurement error. Anything going wrong?

  1. A regular “identification” command following the reference manual cannot go through. Dynare will report an error.

  2. If I instead use the following estimation command


Is the following field a correct place to check the variance decomposition for an endogenous variable, say TY?


All related model and data files are attached. Many thanks for your time! I appreciate that.

example.zip (21.2 KB)

  1. That usually means there is stochastic singularity in your model.
  2. That is a bug in the warning message. Thanks for pointing it out. It should say

This is definitely a problem when you try to do model_comparison. Otherwise, you can ignore it.
3. Identification unfortunately does not support correlations at this point. However, looking at the mode_check plots, there seems to be an issue with d12 and potentially for d21
4. Yes, that is the correct field.

You are right, Johannes. I finally figured out where the stochastic singularity comes from in my model. Now the fitted variables look reasonable and I managed to obtain the global mode. Many thanks for your suggestions.