Payoff values in MCMC with Stochastic option

[hazhir]

Hi,
We are running MCMC on stochastic models, so want to average the payoff (with a pseudo-log-likelihood interpretation) across multiple runs specified by sensitivity analysis controls. From the help file it is not fully clear if the payoff is summed over all sensitivity simulations (the default for optimization but problematic for MCMC, because it inflates the payoff, as if many datasets are available) or averaged over the sensitivity runs (which is more sensible for MCMC but would be inconsistent with optimization). The payoffs reported in mcmc tab files seem to point to the latter (average payoff rather than sum) but I am not sure. Anybody can elaborate which alternative happens in MCMC when stochastic keyword is active?
thanks
hazhir

[tomfid]

It should be summed - I think SO was created with the mindset of policy optimization rather than calibration. A possible response would be to inflate the floor temperature in the MCMC settings to = the number of runs in the ensemble. I'll take a look at the code.