Two questions on convergence diagnostics

Dear all,

Two questions:

  1. Is there any way to run a Geweke convergence diagnostic test after a bayesian estimation with multiple chains (on each single chain)?

2)How should I interpret the MCMC inneficiecy factor results?

Thank you!

  1. Unfortunately, there is no easy way to do that.
  2. The ideal case would be having independent draws from the posterior. But the MCMC only delivers you correlated draws. The inefficiency factor tells you which multiple of the number of draws of the iid case you need to have the same information content in your correlated draws. Think about it this way. If the correlation between draws is 1, you always get the same draw and there is no new information you get about the posterior. In this case, the inefficiency factor would be infinity.

Thank you so much, very clear explanation.

Is there any rule of thumb one should follow to assess whether the inefficiency factor is too high or to set the number of MCMC replications according to the inefficiency factor?

You should look at the convergence diagnostics for guidance on the number of draws.

Hi.

I think that if values of inefficiency factors in Dynare output be smaller is better than higher values. This indicates lower correlation between draws.

Important question is that, how Dynare calculates this inefficiency factor per block and for each of the parameters in the model ?

For example we have 800000 sample in RWMH algorithm and we discard 50 percent of it. Therefore we have 500000 draws per parameter in each block. How Dynare calculates the inefficiency factor for each parameter ?

See the header of

Thanks.