Nahyun Kang

Title: Tweedie Evolutionary Credibility for Insurance Pricing with AR(1) Latent Dependence
Date: August 6th 2026
Time: 2:00pm
Location: Zoom
Supervised by: Himchan Jeong

Abstract:
Insurance claim data often exhibit a large proportion of zeros, strong right skewness, heteroscedasticity, and dependence across repeated observations on the same policyholder. To address these features, this thesis develops an evolutionary credibility framework for insurance pricing based on the Tweedie distribution and an AR(1) latent dependence structure. Conditional on a time-varying latent factor, the claim amount is modeled through a Tweedie generalized linear model with observable policy characteristics entering the mean function. The latent factor is specified as a multiplicative random effect whose serial dependence captures persistent unobserved heterogeneity over time.

Within this framework, we derive a linear credibility premium for future claim cost and study its relationship with naive and static premium estimators as special cases under different dependence assumptions. A simulation study is conducted under known parameter values to illustrate the proposed pricing procedure, and out-of-sample validation is used to compare the competing premium formulas.