Dynamic online prediction model and its application to automobile claim frequency data

TL;DR

This paper introduces a dynamic Poisson state space model for predicting automobile claim frequency that updates continuously with new data, improving accuracy over existing methods.

Contribution

The paper develops a novel DPSS model with time-varying coefficients and smoothing splines, enabling real-time updates and better prediction accuracy in insurance claim modeling.

Findings

Significantly higher prediction accuracy than existing methods.

Effective modeling of time-varying and invariant coefficients.

Successful application to six years of real-world automobile data.

Abstract

Prediction modelling of claim frequency is an important task for pricing and risk management in non-life insurance and needed to be updated frequently with the changes in the insured population, regulatory legislation and technology. Existing methods are either done in an ad hoc fashion, such as parametric model calibration, or less so for the purpose of prediction. In this paper, we develop a Dynamic Poisson state space (DPSS) model which can continuously update the parameters whenever new claim information becomes available. DPSS model allows for both time-varying and time-invariant coefficients. To account for smoothness trends of time-varying coefficients over time, smoothing splines are used to model time-varying coefficients. The smoothing parameters are objectively chosen by maximum likelihood. The model is updated using batch data accumulated at pre-specified time intervals, which allows for a better approximation of the underlying Poisson density function. The proposed method can be also extended to the distributional assumption of zero-inflated Poisson and negative binomial. In the simulation, we show that the new model has significantly higher prediction accuracy compared to existing methods. We apply this methodology to a real-world automobile insurance claim data set in China over a period of six years and demonstrate its superiority by comparing it with the results of competing models from the literature.