Classification of territory risk by generalized linear and generalized linear mixed models

Territory risk analysis is an important study in determining auto insurance rates. To accurately understand the risk in different areas, it is necessary to group similar types of insurance loss level together. Clustering methods help with this grouping process. In this study, we used Forward Sortation Areas (FSAs), which are the first three characters of Canadian postal codes, as the basic units for analysis. These FSAs are nested within cities or towns, and this nesting can affect the insurance costs due to various factors unique to each city or town. For example, limited public transportation may lead to more driving, increasing the risk. Understanding these effects is crucial for estimating the risk associated with each area accurately. Another important question in clustering methods is determining the optimal number of clusters to use for the most accurate results. This must be also addressed in the clustering of territory risk.

In our study, we used a special method called spatially constrained clustering to group together areas with similar insurance loss patterns. Within clustering, we also address the issue of how we determine optimal number of clusters to make the choice of number of clusters statistically sound. This helps us create meaningful territories for rating purposes. We then used mathematical models to calculate the level of risk for each cluster we identified. Within each cluster, all the smaller areas called FSAs have the same level of risk. This helps us better understand and classify the risk associated with different areas.

The findings of this research are important for how auto insurance rates are determined. By using spatially constrained clustering and some mathematical models, insurance companies and regulators can better understand the risk in different areas, in a more logistical way. This helps them set fair insurance rates that match the level of risk in each place. Overall, this research helps insurance companies make better decisions about rates and ensures a fairer insurance system for auto policyholders.

The findings provide important benchmarks that guide how rates are set, making the rate regulation process more informed and effective. Ultimately, this benefits both the insurance companies and policyholders by making the pricing system fairer, more efficient as well as meeting regulation requirements.To learn more, see the full article:

Shengkun Xie & Chong Gan (2023). Classification of territory risk by generalized linear and generalized linear mixed models (external link) . Journal of Management Analytics, 10:2, 223-246. DOI: 10.1080/23270012.2023.2187716