Bias Awareness in Model Governance
Background
Insurance companies want to avoid both intentional and unintentional unfair bias in predictive analytics and other modeling exercises. The SOA Research Institute has published several papers that address approaches for mitigating the potential for bias in modeling, and there is more work to be done.
Research Objective and Deliverables
The research would articulate in a report principle to consider when assessing a model for potential bias, including:
- Under the premise that it may not be possible to eliminate the risk of bias, the principles and factors to consider when an insurer is defining its appetite for bias risk.
- Criteria and processes for implementing and monitoring bias risk assessment.
- Processes and considerations for reviewing and potentially revising bias risk appetite and subsequently reviewing and implementing model updates accordingly.
Target Audience and Impact
Actuaries and other insurance professionals involved in model development and governance as well as regulators.
Estimated Cost
$50,000
Estimated Timeline
Approximately 9-12 months