Comparison of Stochastic Claims Reserving Models in Insurance

TL;DR

This paper compares various stochastic claims reserving models in insurance to evaluate their effectiveness using forecast scoring techniques, aiming to improve reserve estimation accuracy.

Contribution

It introduces a comparative analysis of multiple stochastic reserving methods using forecast scoring, offering insights into their relative appropriateness.

Findings

Certain models outperform others in forecast accuracy

Forecast scoring provides more informative evaluation than traditional measures

The comparison highlights the strengths and weaknesses of different distributional assumptions

Abstract

The appropriate estimation of incurred but not reported (IBNR) reserves is traditionally one of the most important task of actuaries working in casualty and property insurance. As certain claims are reported many years after their occurrence, the amount and appropriateness of the reserves has a strong effect on the results of the institution. In recent years, stochastic reserving methods had become increasingly widespread. The topic has a wide actuarial literature, describing development models and evaluation techniques. We only mention the summarizing article \cite{EV2002} and book \cite{MV2008}. The cardinal aim of our present work is the comparison of appropriateness of several stochastic estimation methods, supposing different distributional development models. We view stochastic reserving as a stochastic forecast, so using the comparison techniques developed for stochastic forecasts, namely scores, is natural, see \cite{GT2007}, for instance. We expect that in some cases this attitude will be more informative than the classical mean square error of prediction measure.