Phase II - Released in 2024-2025
This paper aims to create a framework to help insurers develop models that are more likely to comply with evolving regulations on unfair discrimination and bias.
This paper aims to explore regulatory perspectives on algorithmic bias, including U.S. state regulator concerns with current insurance pricing practices, perceptions of fairness testing approaches and plans for future activities.
This paper evaluates the potential for telematics or usage-based insurance rating variables to reduce insurers reliance on protected information, (e.g. sex, age), or sensitive information, (e.g. marital status, territory, credit).
This joint report from the CAS and Society of Actuaries offers a comprehensive overview of the latest and emerging regulatory activities in China, the U.S., Canada, and Europe, focusing on the prevention of discriminatory practices in the use of artificial intelligence (AI) within the insurance industry. It has a particular relevance to our international audiences.
Please note: we recommend CAS members and the broader actuarial community read this joint paper in anticipation of the full release of Phase 2 of the CAS Research Paper Series on Race and Insurance Pricing.
This two-part paper serves as a practitioners guide to applying fairness testing and mitigation methods to actuarial pricing data and models. Part 1 provides an overview of bias types in insurance, introduces imputation methods for protected class labels, and presents simple fairness tests based on established statistical fairness criteria, expanding upon the 2022 "Methods for Quantifying..." paper.
This paper explores the possible outcomes of common regulatory solutions aimed at reducing impacts of bias on one protected class dimension, such as ethnicity, as it relates to other protected class dimensions, such as age or gender.
Phase I – Released in 2022
By Roosevelt Mosley, FCAS, CSPA and Radost Wenman, FCAS
As the insurance industry focuses attention on potential racial bias across all practice areas, this paper examines three approaches to defining and measuring fairness in predictive models. It also provides an overview of several bias mitigation techniques that can be performed during the input, modeling, or output phase of a model once a set of fairness criteria has been adopted.
By Members of the 2021 CAS Race and Insurance Research Task Force
This paper examines issues of racial bias in lending practice for mortgages, personal and commercial lending, as well as credit-scoring. It looks at these four areas and describes solutions intended to address any potential bias, which may include government intervention, internal bias testing and monitoring measures, and development of new products to mitigate bias.
By Kudakwashe F. Chibanda, FCAS
This paper defines several terms that are currently being used in discussions around potential discrimination in insurance – protected class, unfair discrimination, proxy discrimination, disparate impact, disparate treatment, and disproportionate impact – and provides historical and practical context for them. It also illustrates the inconsistencies in how different stakeholders define these terms.
Please see errata for revisions made on September 23, 2022.
By Members of the 2021 CAS Race and Insurance Research Task Force
This paper examines four commonly used rating factors in personal lines insurance – credit-based insurance score, geographic location, home ownership, and motor vehicle record – to understand how the data underlying insurance pricing models may be impacted by racially biased policies and practices outside of the system of insurance.
