Dynare forums

Dear all,

I am encountering some problems with bayesian estimation.

I kept getting the error message "dynare_solve is unable to find steady state".

Also, what does it mean "mode values outside prior bounds. reduce prior trunc".

Grateful for advice and suggestions.

learning wrote:
I kept getting the error message "dynare_solve is unable to find steady state".

this is a warning, not an error message: for some parameter values, Dynare can't find a steady state. In that case, those parameter values are discarded

Also, what does it mean "mode values outside prior bounds. reduce prior trunc".

Some how you are forcing some mode values (maybe with a mode_file=...), but some of these values are outside the prior bounds. By default, in order to avoid computing likelihood in regions where the prior is about 0, Dynare truncates the priors so as to leave out a probablitiy of 10^-32 on both side of the distribution. You can change this value with the prior_trunc option.

Best

Michel

Hi Michel,

thanks for the prompt response. I like to clarify further.

On the inability to find steady state. I have performed a stoch_simul, and there's no problem getting a steady state that mimic the data. I have set the prior mean near to the experimental parameters. The puzzle in my mind is why does this still occur?

On mode. Do you have an example on how a mode file should look like? I would appreciate if you could elaborate further on this. I have looked into the prior_trunc option. Does it mean that I can set it as "prior_trunc=1e-16", and this will effectively narrow the dispersion of the distribution.

Thanks again in advance, learning

learning wrote:Hi Michel,

On the inability to find steady state. I have performed a stoch_simul, and there's no problem getting a steady state that mimic the data. I have set the prior mean near to the experimental parameters. The puzzle in my mind is why does this still occur?

in short because estimation is about finding the value of the estimated parameters and in this process, Dynare tries many different values, some maybe quite far from the prior mean.

On mode. Do you have an example on how a mode file should look like? I would appreciate if you could elaborate further on this. I have looked into the prior_trunc option. Does it mean that I can set it as "prior_trunc=1e-16", and this will effectively narrow the dispersion of the distribution.

Dynare is creating a mode file <filename>_mode.mat after a first estimation. You can reuse it instead of redoing optimization. This is obviously not what you are doing.

If you aren't using a very recent snapshot, please update your version of Dynare. If you still have the problem, please send me the *.mod file and the data.

Best

Michel