Inferring Ethnicity, Quantifying Disparate Impact, and Fair Lending Compliance

Since the Fair Housing Act (FHA) of 1968 and the Equal Credit Opportunity Act (ECOA) of 1974, banks have been required to monitor, report, and remediate credit scoring models and underwriting policies that show evidence of disparate impact.  Disparate impact occurs when protected minority groups receive quantifiably unfavorable credit and pricing decisions vis-a-vis their unprotected counterparts.  In this session we will discuss the methodology and data used to the infer the ethnicity of loan applicants (BISG – Bayesian Improved Surname Geocoding), the metrics that quantify disparate impact, the test statistics that determine the significance of those metrics, and the modeling approaches used to develop less discriminatory alternatives (widely known as LDA’s to Compliance Professionals).

About the Presenter

Keith Shields has been an Analytics leader and practitioner in the Financial Services Industry for over 30 years.  For 14 years at Ford Motor Credit, he led a global team of over 75 mathematics, statistics, and engineering professionals in the development and maintenance of the credit risk models, pricing strategies, and loss forecasts supporting Ford’s $200 billion loan portfolio.  Since then, he has had a variety of analytic leadership and consulting roles at automotive and fintech lenders, most recently leading a credit risk modeling team at LendingClub Bank, where he had responsibility for the risk assessment and underwriting of the bank’s $1 billion auto refinance portfolio.  While at LendingClub, he also developed the Bank’s analytical processes for the development of Less Discriminatory Alternative (LDA) models and disparate impact metrics, ensuring compliance with Fair Lending laws and federal regulations.  Keith has a Bachelor of Arts degree in Mathematics from Wittenberg University and a Master of Science degree in Industrial and Operations Engineering from the University of Michigan.