Blog
CAS Annual Meeting 2019: Claims Analytics using Multivariate Survival Analysis
2019-11-09
At the 2019 CAS Annual Meeting, I'll be speaking on Claims Analytics using Multivariate Survival Analysis: A Legal Representation Model. The presentation discusses the challenges of constructing and validating a predictive model in which we have only partial information about the response variable, and proposes methods for validating models using censored data. Here's the full abstract:
Our claims team asked us to identify "pain points" for our customers by determining the points in the life cycle of a claim when claimants are most likely to take on legal representation, and identifying factors that affect the likelihood of legal representation. A key challenge in building this model is that the data are censored: we don't know if an open claim will eventually become represented. Participants will learn how to approach such questions using a Cox proportional hazards model, and how to adapt traditional model validation techniques when test data are censored. The audience will participate in the discussion through interactive polling about the factors influencing legal representation, and brainstorming ideas for actions that the claims team might take in response to these factors. The session will be reinforced by a handout that provides sample code and output, illustrating how to apply these techniques using open source software.
The presentation will occur twice, at 4:15 p.m. on Monday, November 11 and 10:55 a.m. on Tuesday, November 12. You can also download the slides and sample code illustrating the model validation techniques discussed during the presentation.
Keywords: Actuarial Science