The impact of climate change on automobile insurance: How to define a bad winter?

In a context of climate change, the severity of winters is tending to increase on the North American continent: Wave of intense cold, violent snow and ice storms. These winter weather conditions, including extreme events, are not currently considered in automobile insurance actuarial models. It is therefore impossible to know the influence of meteorology on financial losses and even less to anticipate them in a context of climate change. While extreme events are relatively well studied, the impact of winter conditions is less well known. These are small individual losses but have a high cumulative impact. This observation leads us to question the meteorological variables that characterize a "bad" winter in terms of automobile insurance.

The objective of the study is to define the proportion of winter weather variations in the number of observed automobile claims and associated losses, as well as to predict future losses in the context of climate change. We also provide valuable insight into uncertainties of the developed forecasts for claim severities with respect to different climate model and greenhouse emission scenarios. The results of our study are a first step for more accurate short- and long-term cost-benefit assessment of climate adaptation in the automobile insurance sector.

The strategy we employ includes an analysis of nonlinear dependencies between weather and claims data; detection of critical thresholds, or tipping points leading to an increased number of claims; and joint frequency-severity predictive modeling of weather-related daily losses. In this study, we use Canadian automobile insurance claim data from Co-operators, local weather station data, gridded instrumental data products, and regional climate model data. The weather observations we use in our analysis are provided by Environment Canada in its Digital Archive of Canadian Climatological Data. Daily climatological data are obtained for each city, from which we use minimum/maximum temperature, total precipitation (snow and rain), maximum speed gust and snow on ground.

Climate Model Output Data from several climate model experiments are derived from the CanRCM4 version of the Canadian Regional Climate Model run by the Canadian Center for Climate Modeling and Analysis. To evaluate uncertainty due to various scenarios of global warming, we use available data at 25 kilometers spatial resolution from two CanRCM4 experiments run with the second-generation Canadian Earth System Model, CanESM2, as input data. We consider two runs from 2006 to 2100, forced with greenhouse gas scenarios known as RCP 4.5 and RCP 8.5. These Representative Con- centration Pathways (RCPs) are commissioned by the IPCC and have been used widely in different efforts to study potential impacts of climate change.

To develop an attribution analysis of current weather-related losses and to assess dynamics of future claim severities, we combine the approaches to modeling several weather-related claims, with a collective risk model (CRM) simulation algorithm.

In this study, we propose a method for joint frequency- severity predictive modeling of weather-related claim severities. The first step of our methodology is to find critical thresholds for weather variables after which the severity of daily claims starts to increase. The study proposed here to explain the critical thresholds and methods to derive them that could be used by the insurance companies to mitigate risks and minimize their losses. The selected thresholds and estimated model parameters further play a role in assessing future trends in the number of automobile insurance claims.

Although insurance policies are updated regularly and can therefore be modified to reflect actual conditions, the sporadic nature of many weather-related perils and relatively rapid changes result in a gap between the observed event statistics and the actual probability of an event that increases in frequency. We need to consider this gap. These fundamental issues related to the management of insurance risks in the face of climate change can only be addressed if the gaps between actuarial theory and practice, statistics and climate science are addressed. This work will require multidisciplinary efforts by industry.