CS-22 - Evaluating and Selecting a Bayesian MCMC Model

This session will walk through the process of evaluating and selecting a Bayesian MCMC model from amongst a set of models. The models were created to develop loss reserve estimates. Sometimes, one can have multiple models which produce plausible results and how to select between those models to find the one that is the best model or choosing to use model averaging may not be clear the first time one uses Bayesian MCMC. Occasionally, a model will generate warning messages indicating that there were numerical problems in the iterative process used to solve for parameters which would make using a given model unsafe and there will be an example to illustrate what that looks like and the diagnostic tools available. The intent of the session is to offer examples to those new to using Bayesian MCMC to illustrate how to use the tools at hand to help with the decision process. There will be some comments during the presentation to point out how the current set of tools has helped make working in the Bayesian MCMC modeling environment more practical in recent years. The session will not cover the theory underlying the tools that are demonstrated.