2025 CAS Virtual Race and Insurance Pricing Seminar

Session 1: CAS Research on Regulatory Perspectives on Algorithmic Bias and Unfair Discrimination

This session will share key learnings from one of the new reports in the CAS Research Paper Series on Race and Insurance Pricing. We will begin by sharing highlights of a survey of state insurance regulators, aiming to understand activity related to algorithmic bias and what issues are most likely to be addressed in the coming years. The survey also evaluates regulators' views on the responsibility of the insurer in addressing potential bias, and their perceptions of specific private passenger auto insurance rating factors.

Learning Objectives:

1. Understand the current regulatory environment with regards to potential algorithmic bias in insurance.
2. Explore the potential for future regulatory actions impacting the insurance industry and actuarial models along with considerations for actuaries.
3. Explore the current regulatory perspectives on certain PPA rating variables.

Speaker:

Taylor Davis, Actuarial Consultant, Risk & Regulatory Consulting

Session 2: ASOP 12 and Other Bias ASOPs

Description TBD.

Speaker:

Bob Miccolis, Managing Principal, Miccolis Consulting LLC

Session 3: A Practical Guide to Navigating Fairness in Insurance

Regulations are evolving to address potential bias and discriminatory impacts resulting from AI, machine learning and other predictive models. What are some practical ways you can be prepared? This session will first compare new and emerging bias-related regulations, and then look at practical ways insurers can be prepared in a rapidly evolving regulatory environment.

Learning Objectives:

1. Understand the evolving regulations around unfair discrimination in insurance.
2. Engage with knowledge with other leaders in their own organization to bring their own perspective of unfair discrimination.
3. Take practical steps to consider fairness in their analytical work.

Speakers:

Jessica Leong, CEO, Octagram

Richard Moncher, Consultant, Sigma Actuarial Consulting Group

Session 4: Practical Applications of Bias Measurement and Mitigation Techniques in Insurance Pricing

In this session, researchers from the CAS's Race and Insurance Pricing Task Force will discuss a suite of tools and methodologies covered in their research paper, Practical Applications of Bias Measurement and Mitigation Techniques in Insurance Pricing. This session begin by illustrating the Bayesian Improved (First Name) Surname Geocoding method for imputing information on race, explain its limitations and challenges, and discuss how the outputs of this method can be used in fairness analyses. Next, our speakers will cover the extension of fairness criteria (Independence, Separation, and Sufficiency) to insurance premiums and loss ratios, and provide an overview of more complex fairness testing methods including Conditional Demographic Parity, the Proxy or Control Variable Test, and Nonparametric Matching. Finally, the session will introduce a selection of statistical bias mitigation approaches that can be applied to at the input, modeling, or output stages of insurance pricing models.

Learning Objectives

1. Utilize the BIFSG method to impute race and ethnicity as part of a model fairness analysis.
2. Analyze an insurance pricing model using various fairness testing techniques and evaluate tradeoffs between different techniques.
3. Apply bias mitigation approaches to insurance pricing models.

Speakers:

Eric Krafcheck, Principal & Consulting Actuary, Milliman

Craig Sloss, Technical Consultant and Lead Data Scientist, Definity Financial Corporation

Gary Wang, Senior Consulting Actuary, Pinnacle Actuarial Resources, Inc.

Mike Woods, Actuarial Director, Allstate Insurance

Session 5: Potential Unintended Impact of Bias Mitigation

This presentation will investigate the balance between maintaining accurate risk differentiation and ensuring equitable treatment among various classes of interest when certain rating variables are restricted or banned through regulation. Using a collection of foundational synthetic examples we have investigated a spectrum of scenarios with imposed regulatory actions.

Within these simulated scenarios we review models reflecting traditional regulatory constraints such as:

1.Limiting the range a rating variable is allowed to differentiate (capping), reflecting policies such as limiting territorial differentiation in pricing

2.Prohibiting the use of a certain rating variable, reflecting policies such as prohibiting use of credit-based insurance scores

We also review other modeling approaches intended to improve parity within a particular class of interest, such as;

1. Utilizing the class of interest as a control variable in a rating model
2. Utilizing the class of interest to create residualized rating variables to be used in a rating model

Learning Objectives

1. Examine how the competitive market reacts to regulatory constraints when multiple companies have different portfolio mixes relative to the market population.
2. Explore how the competitive market reacts when two classes of interest have opposite correlations with the regulated rating variable under consideration.
3. Investigate how mitigation approaches that proactively utilize the Main Class of Interest information compare to traditional regulation approaches.

Speakers:

Michael Chen, Senior Consulting Actuary, Pinnacle Actuarial Resources, Inc.

Gary Wang, Senior Consulting Actuary, Pinnacle Actuarial Resources, Inc.