Understanding Insurance Profitability using Bayesian Hierarchical Models

Chris & Cynon will present: Understanding Insurance Profitability using Bayesian Hierarchical Models

We produced models for calculating claim notification rate, actual claims, and average claim severity for multi-class, multi-year analysis using BRMS (and Stan).

Our models account for negative incurred but not reported claims (IBNR), small volumes of data, and long tailed lines. We used the lognormal-pareto distribution as well as combined binomial-poisson models, and compression to solve some of the industry standard issues.

We will take you through our process for model building as well as some example models.

In addition we will share some war stories of what it takes to build these kinds of models in large insurance companies

Background
Chris H: Insurance Data Scientist for 6+ years, worked on pricing / marketing / claims problems. Currently working in R&D at Markel. I have only been building commercial Bayesian models for 1 year, and will talk about what it took to get these models up and how we are rolling out these models internally.
Cynon S: Bayesian Actuary for 20+ years :), coach, mentor