Autocalibration and Tweedie-dominance for insurance pricing with machine learning

Does autocalibration improve goodness of lift?

“…The authors are grateful to Professor Mario Wüthrich for having pointed out that increasingness of the regression function was not needed in Property 5.1 in Denuit et al (2021).…”

“…Wüthrich, 2019Wüthrich, , 2020. To solve this problem, Denuit et al (2021) proposed a new strategy based on the concept of autocalibration (see, e.g., Kruger and Ziegel, 2021) to restore global balance as well as local equilibrium in the spirit of the original method of marginal totals dating back to the 1960s. Specifically, after the analysis has been performed with a method that does not necessarily respect marginal totals, a local constant GLM fit is achieved.…”

Model selection with Gini indices under auto-calibration

“…A bigger Gini index can only be achieved by a larger information set than the σ-algebra generated by X. Note that Property 5.1 of Denuit et al [2] gives a method of restoring auto-calibration for continuously increasing regression functions.…”

“…Intuitively, this tells us that the Gini index is a rank-based score that is not calibration-sensitive. The missing piece to make the Gini index a consistent scoring rule is to restrict it to the class of auto-calibrated regression models, this is proved in Theorem 3.5, below; for auto-calibration we refer to Krüger-Ziegel [9] and Denuit et al [2].…”

Churn modeling of life insurance policies via statistical and machine learning methods -- Analysis of important features

“…Lately, machine learning techniques have become more and more important in the context of modeling insurance data, see e.g. Makariou et al (2021); Devriendt et al (2021); Denuit et al (2021); Gan (2013). In this work, some of these methods shall be used for a better understanding of the individual contract cancellation behavior of a large German life insurer's clients.…”