Bootstrap modeling: Beyond the basics

Motivation. Bootstrapping is a very versatile model for estimating a distribution of possible outcomes for the unpaid claims, is relatively easy to use and explain to others, and can be readily “generalized” to be more flexible and combined with other related models that can be used to assess risk for a wide variety of enterprise risk management issues. While the CAS literature includes several papers that describe the bootstrap model, all of these papers are limited to the basic calculations of the model or focus on a particular aspect of the model. In contrast, this paper outlines the modifications to the basic algorithm that are required in order to put the bootstrap model into practical everyday use.

Method. This paper will start by pulling all of the issues from different papers into the complete basic bootstrap modeling framework using a standard notation. Then it will describe some of the enhancements required for practical usage and it will show how the output of the model can be easily "extended" to address other risk management issues. It will then expand the basic model and generalize the approach, as well as address many common modeling issues that arise during the diagnostic testing of the model parameters and assumptions. Finally, it will summarize testing of the model using simulated data and suggest possible areas for further research.

Results. The paper will illustrate the practical implementation of the bootstrap modeling framework as a powerful tool for estimating a distribution of unpaid claims.

Conclusions. The paper outlines the full versatility of the bootstrap model for the practicing actuary.

Availability. A set of companion Excel files are available at http://www.casact.org/pubs/forum/10fforum/, which contains the calculations illustrated in this paper as well as serving as a learning tool for the student or practicing actuary.