Number of MH iterations and number of burning periods

Estimation ML/Bayesian estimation
1

Dear Johannes,

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

I have around 30 parameters to be estimated in my DSGE model using Bayesian estimation and the remaining 5 parameters are calirated, and I ran 1000000 iterations using Metropolis Hastings sampler, is 1000000 iterations too much for only 30 parameters?

And how many burning periods are suitable?

Thank you very much.
Jesse
PhD Candidate

2

There is no general rule on the number of iterations. It very much depends on the autocorrelation in the draws/inefficiency factors. You should always check convergence via convergence diagnostics and trace plots. Usually a burnin of 100,000 draws is sufficient if you start at the true mode.

1 Like