Dynare forums

hi
i would like to know what exactly the difference between mh_jscale and mh_init_scale is.
they seem to be both doing the same thing, ie. adjusting the step size of the M-H steps.
i guess the init_scale comes into play only when we do multiple chains, so that initial values for each chain is different.
so after the 'initialisation' draws are done for each chain, the mh_jscale takes over instead of the init_scale at a later step in the metropolis.m dynare file.

am i right?


cheers
Reuben

Hi Reuben,

You are perfectly right!

We randomly select the initial conditions of the 'parallel' Markov Chains in a gaussian distribution centered on the posterior mode. We need to be sure that we obtain very different initial conditions (otherwise the convergence checks are meaningless). So we increase the variance of the proposal distribution with a scale factor mh_init_scale bigger than mh_jscale.

Best,
Stéphane.

reubenpjacob wrote:hi
i would like to know what exactly the difference between mh_jscale and mh_init_scale is.
they seem to be both doing the same thing, ie. adjusting the step size of the M-H steps.
i guess the init_scale comes into play only when we do multiple chains, so that initial values for each chain is different.
so after the 'initialisation' draws are done for each chain, the mh_jscale takes over instead of the init_scale at a later step in the metropolis.m dynare file.

am i right?


cheers
Reuben