Market-Based Model Selection with an Application in Commercial Auto Ratemaking

It is often difficult to know which models to use when setting rates for auto insurance. We develop a market-based model selection procedure which incorporates the goals of the business. Additionally, it is easier to interpret the results and better understand the models and data. As an application, we compare many different models (variants of GLMs, GLMMs, regularized regression, random forest, and spike and slab models) on a robust dataset of 70,000 commercial auto policies. We first compare the models using standard metrics (MAPE, MSPE, and runtime), leading to mixed results and interpretation. The market-based model comparison method shows that the random forest model far outperforms the other models in terms of both loss ratio and market share, likely compensating for the increased computational cost.