Local bias adjustment, duration-weighted probabilities, and automatic construction of tariff cells
Seminar in Insurance and Economics
SPEAKER: Filip Lindskog (Stockholm University).
TITLE: Local bias adjustment, duration-weighted probabilities, and automatic construction of tariff cells
ABSTRACT: We study non-life insurance pricing and present a general distribution-free procedure for constructing a locally unbiased predictor of the risk premium based on any initially suggested predictor. The resulting locally bias adjusted predictor is piecewise constant, corresponding to a partition of the covariate space. By construction the locally bias adjusted predictor is auto-calibrated. Allowing for possible early termination of contracts means that contract duration should be seen as a random, given the information available when a new contract is written. Requiring that the expected earned premium equals the expected claim cost, given contract features, implies that the actuarially fair premium can be expressed as a conditional expectation with respect to a duration-weighted probability measure. The effects of local bias adjustment and randomly varying durations are illustrated based on simulated data and real insurance data. Automatic approaches to local bias adjustment of an initially suggested predictor are shown to produce tariffs with a rather small number of tariff cells while maintaining or improving the predictive performance compared to the initial predictor.
The talk is based on joint work with Mathias Lindholm and Johan Palmquist.