Question regarding posterior estimation results

[np2333]

Hi,

I am estimating a medium scale dsge model. I get the following when the estimation is over:

"
ESTIMATION RESULTS

Log data density is -2302.991052.
posterior_moments: There are not enough draws computes to compute HPD Intervals. Skipping their computation.
posterior_moments: There are not enough draws computes to compute deciles. Skipping their computation.
"
I am wondering what the issue might be. u run 2 MCMC chains and the acceptance rates are 50 % and 47% respectively.

[np2333]

and the number of draws in each chain in 100,000

[jpfeifer]

Please provide all files necessary to replicate the problem.

[np2333]

HI Johannes,

thanks for your response. The mode file is attached and a link to the data file is below.
https://dl.dropboxusercontent.com/u/144 ... ep22_1.mat

thanks

[np2333]

sorry i meant the "mod" file is attached. incase its helpful, heres a link to the mode file:
https://dl.dropboxusercontent.com/u/144 ... 3_mode.mat

[jpfeifer]

This is a bug in 4.4.3:

When computing posterior moments, Dynare falsely displays that moment computations are skipped, although the computation was performed correctly.

See http://www.dynare.org/DynareWiki/KnownBugs

[np2333]

Hi Johannes,


thanks. I just want to be sure that its the bug and not something else, because the posterior standard deviation and 90percent confidence intervals for some parameters are extremely tight (see below), and for some reason the number 0.7902 makes an appearance too often which makes me wonder whether these computations are made.


parameters
prior mean post. mean 90% HPD interval prior pstdev

theta_H 0.500 0.7902 0.7902 0.7902 beta 0.0500
theta_F 0.500 0.1919 0.1270 0.2545 beta 0.1000
theta_Hstar 0.500 0.9529 0.9528 0.9529 beta 0.1000
theta_Fstar 0.500 0.7542 0.7233 0.7902 beta 0.0500

[jpfeifer]

I am pretty sure there is no Dynare bug here at work. It rather looks as if some of your parameters, particularly the theta_H did not mix at all in the MCMC chain. This suggests that there is still a problem somewhere.

[np2333]

can you explain what you mean by "mix"? thanks

[np2333]

Even before starting the MCMC chains, the posterior mode standard deviation (with mode compute=9) is 0.0000 for a couple of parameters (including theta_H). Can this help me identify where the problem is?

[jpfeifer]

That is the problem. Your proposal density is too small for these parameter so that they do not move. Load the _results-file from your last run and type

Code: Select all

trace_plot(options_,M_,estim_params_,'DeepParameter',1,'theta_H')

to see this. The syntax for trace_plot is documented in Pfeifer (2014): An Introduction to Graphs in Dynare at https://sites.google.com/site/pfeiferecon/dynare