Description: In the insurance industry, machine learning models are increasingly used to make decisions that affect people's lives. In order to ensure that models are fair, they should be audited for bias. There are no hard and fast rules for bias audit; metrics that can be used to evaluate fairness vary depending on the context in which the model is deployed. In this session, we present a flexible framework that highlights key considerations in bias evaluation that can be adjusted to suit your process. In this session, we will learn about different metrics to assess fairness, discuss open source tools that can be used for bias audit, and walk through a case study.

Description: From a session at the 2001 Valuation Actuary Symposium, held in Lake Buena Vista, Florida, November 29-30, 2001 Teaching session covering the development of realistic liability models for use in asset adequacy analysis, asset/liability management, pricing/repricing, financial reporting or other corporate purposes. The primary focus is modeling issues for variable product features, such as minimum living or death benefit guarantees on both variable life and annuities, annuity death benefit enhancements, and variable immediate annuities. Topics include: • Analysis of the special risks posed by these products • Selecting realistic dynamic assumptions • Techniques used in modeling these products • Scenario testing and result presentation to best communicate risks