Supervisors Symposium

MODEL BIAS

Model Validation: Bias Concerns (cont’d) Jimenez suggested that insufficient attention is being given to the potential for bias in supervisory algorithms used by regulators. • Financial regulators have used machine learning to detect fishy text in corporate filings, identify money ‐ laundering networks and discover tax cheats. • Bias can lurk in the collected Big Data, however.

21

21

MODEL BIAS

Model Validation: Bias Concerns (cont’d) Jimenez cites the 2017 FDIC National Survey of Unbanked and Underbanked Households: • 14% of Hispanic and 17% of African ‐ American households are unbanked. • In some instances, such households may not have a government issued ID card. • The upshot is that these households may rely heavily on cash and prepaid cards but bank AML algorithms flag frequent prepaid transactions as being suspicious even though there is no problem involved and similarly banks may file a disproportionate number reports to FinCEN for suspicious IDs used as substitutes for official government issued IDs.

22