A Comprehensive System for Selecting and Evaluating DFA Model Parameters

Stochastic scenario generators for assets and liabilities are critica1 components of a robust DFA model. Vital to any stochastic scenario generation system is the selection of the underlying parameters. The process of parameter estimation is second only to model structure in the quest for generating reasonable results. If the model is simple, we can use standard statistical methods such as maximum likelihood to estimate parameters. However, for very complex models, we need to establish criteria for evaluation and find the parameters that are best with respect to those criteria. In this paper, we discuss a parameter estimation system called American Re-lnsurance Company’s Constraint Evaluator System. This system allows modelers to define a multitude of targets and to assign a weight to each target to create a comprehensive objective function. Each target represents a quality that the model should possess with an assigned leve1 of significance (weight). The targets are based on historical analysis or on some rational vision for future relationships. We discuss the analysis involved in setting appropriate targets including the monitoring of relationships between variables in a multi-period environment. Our goal is to minimize the deviation between the user-defined targets and the model output. This is a non-convex optimization problem, which we use a combination of techniques to solve. Finally, we study the robustness of our parameter estimates as it relates to the number of scenarios and the observed model outputs.