Impact of the COVID-19 Pandemic - Modelling and Forecasting using Affectedness Variables

When setting rates in P&C insurance, actuaries should ensure that the used data is appropriate to the subject of the respective study. Data suitability with regard to the objective must be ensured. Usually, actuaries have a data set that covers several years. Since this data set may consist of several exposure years, its characteristics are subject to change over time and thus have an effect on the target variables of the models. In general, it is not a problem if the distributions over the characteristics of tariff features change, as long as the fundamental properties are unaffected. However, this becomes more difficult when changes to the target variable occur that are not inherent to the model but rather are induced by an additional external effect, such as the COVID-19 pandemic. In this talk, we propose the introduction of new variables, so-called affectedness variables, into the modeling that can be used to characterize the changes in the relevant time periods. Affectedness variables are contextualized variables that are assumed to have an impact on a temporary change in behavior. This makes data that is affected by a temporary change usable for modeling and forecasting.

This talk is based on the report “Impact of the COVID-19 pandemic - modeling and forecasting using affectedness variables” of the Committee on non-life insurance of the German Association of Actuaries.

Find the Q&A here: Q&A on 'External Challenges for the General Insurance Industry'