Ventana software support forum

Hi,

If I want to use MCMC as a method of payoff sensitivity analysis in the standard Powell optimizer, do I have to manually edit the .voc file to set the options associated with MCMC? I am asking because when using the user interface (after clicking Optimizer and going to Optimization Control panel), there seems to be no option named "PAYOFF MCMC" under the "Sensitivity" drop-down list. The options available are "Off", "Payoff Value", "Payoff Percent", "Param Percent" and "All Constants". Maybe "Payoff Value" is "PAYOFF MCMC" as referred to in MCMC Options of Vensim Reference Manual?

Also, the same panel does not allow me to edit those MCMC options and parameters, if "Powell" is chosen as the Optimizer. (Of course, when "MCMC" is chosen, all the options are there. But then that means using MCMC as a standalone method?)

So, do I need to edit all these in the .voc file rather than through the user interface when I want to use MCMC as a method of payoff sensitivity analysis in the standard Powell optimizer?

Many thanks.

I think it's actually better to keep things separate - do a Powell optimization, then restart from the optimal point and do a separate MCMC. Or, skip Powell altogether and go straight to MCMC.

Hi Tom, thanks for your advice. I have tried to do it in 2 steps. The 1st is a regular optimization using Powell through which I found calibrated parameters at maximum payoff. Then, the 2st starts from this optimum status (by replacing the original parameters with the calibrated parameters in the model), and MCMC is chosen as the optimizer, with all other MCMC-related options and settings being the default. However, after clicking finish, I get the following warning message:

Floating point error during optimization with
------------------
*Shape parameter = 3.87151
*Scale parameter = 19.1889
------------------
WARNING: Too many floating point errors initializing chains; aborting..
The final payoff is -4.090221e+02

Does it mean that I have done something wrong, and the MCMC process has not run at all?

A related question is: the MCMC Output of the reference manual says " (when used as a payoff sensitivity method), confidence bounds on parameters." Since I am taking the 2-step approach, how to get information on confidence bounds on parameters?

Many thanks for your help.

If the optimizer runs into errors starting up the model, it won't run.

I'd plug in these values and see why they cause an error.

In the 2-step approach, the best way to get confidence bounds is to open up the runname_sample.tab file and look at the 2.5% and 97.5% quantiles of the results. Also, scatter plot scale vs. shape, and you can see the joint distribution of the parameters.