Dear Johannes,

I have run my DSGE model using Bayesian method in Dynare successfully without any warnings nor errors, however, when I try to run the same model using Maximum Likelihood method, I have received the warning and error messages, 'Matrix is singular, close to singular or badly scaled.'

I am wondering that what are the causes, why Bayesian method runs well without any warnings or errors in Dynare, but maximum likelihood method runs with warnings and errors in Dynare?

Thank you very much!

Jesse

The question is in which function this error appears. A possible explanation is that with ML some tested parameter sets are pathological, while with Bayesian estimation the prior prevents testing these parameters.

Dear Johannes,

Thank you very much for helpful guidance, I am grateful.

Now I can run maximum likelihood estimation, however, I find standard errors are Inf and 0 t values, although I can run Bayesian without any warnings nor errors, does it indicate that data did not add much to the posterior distribution of the parameters?

Thank you!

Jesse

No, that suggests problems with mode-finding. Did you look at the mode_check plots? A prior often helps to smooth out the likelihood function and results in a better behaved problem for maximization. That may be what you experience.