Loss Reserving Using Claim-Level Data

While the actuarial literature devoted to stochastic loss reserving has been developing at an impressive rate, much of this literature has been devoted to the statistical analysis of summarized loss triangles. This restriction limits the benefits that modern statistical techniques can bring to the subject of loss reserving. This paper will sketch one possible framework for estimating future claims payments using claim-level data. The first part of the paper will discuss the use of covariates (or "predictive variables") to improve one's estimates of future payments, especially in cases where the mix of business being analyzed has changed over time. The second part of the paper will describe how the bootstrapping technique can be applied to claim-level data to estimate reserve variability.