Modelling the occurrence dynamics of these claims traditionally starts from an identified severe weather event, such as a wind- or hailstorm, after which the claims caused by that event are analysed. In practice, however, claims do not only arise on the handful of days with severe weather. Moderate conditions generate fewer claims per day, but there are many such days, and together they account for a substantial share of the total. Additionally, the claims handling process runs continuously. An insurer therefore needs insight into the occurrence and reporting of claims on every calendar day, not only on the days that make the news. Such insight serves several actuarial purposes: estimating the number of outstanding claims for reserving, planning the capacity of claim handling teams, and, given reliable forecasts, warning policyholders ahead of severe weather so that some claims are prevented.
Two complications arise when all calendar days are modelled jointly. First, claims that occurred recently have only been partially reported, so the most recent observations are incomplete. Second, the rare days with severe weather are pooled with the many moderate days, while the dynamics on those days differ substantially. Using machine learning techniques, we model the occurrence and reporting of weather-related home insurance claims per calendar day, accounting for the claims not yet reported and allowing the dynamics to differ across weather conditions.
Lees dit artikel verder onder Download.