Machine learning, bias correction and plug-in estimators for an accurate microlevel reserving.

Summary Thanks to nonparametric estimators coming from machine learning, microlevel claim reserving has become more and more popular in actuarial sciences. Recent research has focused on how to integrate the whole information one can have on claims to predict individual reserves, with varying success due to incomplete observations. In this paper, we introduce three extensions to comparable existing works: how to deal with censoring and truncation present in such type of data, how to cope with inflation when the inflation factor is unknown, and how to implement an adequate strategy leading to robust personalized reserve estimates. Using independent test sets, our results-on guarantees with typical long development times-indicate the importance of using the total claim development time to predict the reserves with acceptable accuracy. To remain close to reality, our applications are based on two open portfolios based on real-life datasets.