problems with log data density

[tess1704]

Hi everybody,

I am estimating a RBC model with Bayesian techniques. I use the ustable version of DYNARE because the standard one gave problems with Brooks and Gelman (1998) diagnostics.

1) I get Log data density [Laplace approximation] = 321.993362 with posterior maximization and after posterior mode calculation it is 322.301353.

Is this value ok?

Because I guess it should be far higher, isn't it?

2) Dynare gives me this warning after calculating the initial value of the log posterior and before starting the optimization procedure:

Initial value of the log posterior (or likelihood): 96.484
Warning: File 'filename_optimal_mh_scale_parameter.mat' not found.
> In <a href="error:C:\dynare\2010-10-22\matlab\dynare_estimation_1.m,394,1">dynare_estimation_1 at 394</a>
In <a href="error:C:\dynare\2010-10-22\matlab\dynare_estimation.m,62,1">dynare_estimation at 62</a>
In <a href="error:C:\Programmi\MATLAB7\work\capital maintenance\bayesian estimation\maint_bayes_newdata.m,263,1">maint_bayes_newdata at 263</a>
In <a href="error:C:\dynare\2010-10-22\matlab\dynare.m,132,1">dynare at 132</a>

What does this mean?
How could I build this file?
Could this be a reason why I get low value for the Log data density?

Thank you

Alice

[jpfeifer]

1) Unfortunately, I am not aware of any good guidelines regarding the Log data density. This value depends on many different factors, with the number of observed data series and the number of data points having a prominent role. Hence, it is hard to judge the value. I would suggest looking at the smoothed estimates to judge the fit of the estimated model.
2) This is a warning (and not an error) that you can usually safely ignore. Is the MH-acceptance rate OK? And which mode_compute are you using?

[tess1704]

Thank you.

The MH-acceptance rate is about 0.26 on average, so I think it's ok. I'm using Sims' mode_compute (4, in dynare language).