Actuarial reserve analyses typically rely on a number of different estimation methods to develop indicated ultimate loss.  The paid and incurred (i.e., paid plus case) chain ladder methods are surely the most common.  Many other methods exist as well, but are used less frequently.  Oftentimes, these methods diverge significantly, and actuarial judgment is used by many actuaries in selecting ultimate loss.  A need exists for empirical evidence to support the use of particular methods over others.

In this paper/presentation, thirty actuarial reserving methods are evaluated empirically against an extensive database of triangular data, from Schedule P of United States Property & Casualty Annual Statements, for the years 1996 through 2010.  The metric of method skill (as defined by Jing, Lebens and Lowe in “Claim Reserving:  Performance Testing and the Control Cycle,” published in the 2009 issue of Variance) is used to evaluate the historical performance of the methods.  Results are provided by company size and line of business.  The effect of correlation on the usefulness of additional methods is considered.  Results suggest the use of several methods not common in actuarial practice, as well as a refinement of weights typically assigned to the more common methods.

A general outline of the paper/presentation is as follows:

• Overview of the analysis, including the data available as well as a discussion of the metric.
• Results of the analysis, including results by company size and line of business.
• The effect of correlation between methods on the results and the practical implications of this correlation.
• Additional discussion on the approach to the analysis, and, in particular, the metric selected. 
• Lastly, some conclusive remarks.