EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

2220. A beta binomial approach for the estimate of surrender rates by GAMLSS

Invited abstract in session MB-63: Insurance Risk Management, stream OR in Banking, Finance and Insurance: New Tools for Risk Management.

Monday, 10:30-12:00
Room: S14 (building: 101)

Authors (first author is the speaker)

1. Davide Biancalana
Sapienza University of Rome
2. Fabio Baione
Università La Sapienza

Abstract

This study examines the estimate of surrender probabilities with explanatory variables using a Beta Binomial Generalized Linear Model for Location, Scale, and Shape (BBGAMLSS) and introduces an algorithm for the update of surrender probabilities by future observations. The Binomial Generalized linear Model is widely used in actuarial practice and literature to predict surrender probabilities per policy count depending on the policy, policyholder, and economic variables. We propose an alternative regressive model with a response variable assumed to be Beta Binomial. In contrast to a binomial, a discrete random variable called a Beta Binomial has a beta distributed probability of success at each of a certain number of trials rather than a fixed probability. The surrender is a binomial phenomenon where the probability of success is not fixed, because it depends on many variables that cannot be considered simultaneously in a single regressive model, without introducing overparametrization. Lastly, the Beta Binomial random variable has a Bayesian interpretation, because the beta is a conjugate prior to the Beta Binomial. Then, it is possible to define an algorithm that allows updating the estimates of conditional surrender probabilities by the future observations of surrender rates avoiding the estimate of new parameters every year.

Keywords

Status: accepted

Back to the list of papers