Insurance Premium Prediction via Gradient Tree-Boosted Tweedie Compound Poisson Models

摘要

The Tweedie GLM is a widely used method for predicting insurance premiums.However, the structure of the logarithmic mean is restricted to a linear formin the Tweedie GLM, which can be too rigid for many applications. As a betteralternative, we propose a gradient tree-boosting algorithm and apply it toTweedie compound Poisson models for pure premiums. We use a profile likelihoodapproach to estimate the index and dispersion parameters. Our method is capableof fitting a flexible nonlinear Tweedie model and capturing complexinteractions among predictors. A simulation study confirms the excellentprediction performance of our method. As an application, we apply our method toan auto insurance claim data and show that the new method is superior to theexisting methods in the sense that it generates more accurate premiumpredictions, thus helping solve the adverse selection issue. We haveimplemented our method in a user-friendly R package that also includes a nicevisualization tool for interpreting the fitted model.