RINA WANG
Title: The Application of Categorical Embedding and Spatial-Constraint Clustering Methods in Nested GLM Model
Date: Monday, December 18th, 2023
Time: 2:30pm
Location: Hybrid 鈥� LIB 2020 & Zoom
Supervised by: Dr. Jiguo Cao
Abstract:
Generalized linear model (GLM) is a popular choice for pricing non-life insurance policies. However, the high-cardinality categorical rating variables in non-life insurance pose significant challenges for GLM. Furthermore, insurance regulators often require the location related categorical variable to be transformed into a lower-dimension contiguous Territory variable. Motivated by these challenges, this paper proposes a nested GLM framework for non-life insurance ratemaking applications. In this framework, a neural network model with categorical embedding layers effectively translates categorical variables into meaningful numerical representations. Of these, the ones corresponding to the location-related variable are further converted into a contiguous Territory rating variable via spatial-constraint clustering models. Incorporating outcomes from these models, the nested GLM satisfies regulatory requirements, and enhances the model鈥檚 predictive power, while maintaining the interpretability from the (generalized) linear form. The construction and performance of a nested Poisson GLM are demonstrated using a real-life Brazil auto insurance data to model claim frequency.