ERICA JIANG, WILL SHUO LIU & YEONJOON LEE | MARCH 15, 2022
To obtain loans in most consumer credit markets, households interact with loan officers. The discretion allowed to loan officers may significantly affect households’ credit access. For example, both the U.S. Department of Justice and the Federal Trade Commission have filed enforcement actions against the abuse of loan officer pricing discretion. The newly issued loan originator compensation rules under Regulation Z impose restrictions on payments of compensation to loan originators to address concerns about their lending practices. Meanwhile, artificial intelligence modeling has been developed to eliminate human biases and has been widely adopted in the U.S. consumer credit market (figure 1). Despite the extensive debate on these issues, there is limited research on loan officers’ discretion and how the use of a machine changes the role of loan officers in determining credit outcomes.
In our recent working paper, we answer these questions by taking advantage of a new disclosure requirement after the financial crisis. The Dodd-Frank Act requires individual loan officers to be licensed or registered with the Nationwide Licensing System and Registry (NMLSR). We obtained individual loan officer registration records from NMLS Consumer Access and infer the race of each loan officer using both the surname and the first name. We then merge loan officer race data with each individual loan application record for which the loan officer is primarily responsible. This merged data set provides, for the first time, loan-level mortgage application records linked to each registered mortgage loan officer in the United States.
Figure 1: Regional Distribution of Automated Underwriting System (AUS). The AUS adoption is measured by the share of loan applications that went through an AUS out of the total loan applications submitted from borrowers in a county. Data source: 2019 Home Mortgage Disclosure Act (HMDA).
With this novel data set, we document a significant underrepresentation of minority mortgage loan officers, which is common across financial institutions, especially in small banks and shadow banks (figure 2). The minority loan officer share turns out to be negatively associated with the well-documented racial mortgage rejection gap in the United States (figure 3). The adoption of algorithmic underwriting exhibits a similar negative correlation with the racial gap in mortgage approval and weakens the association between the share of minority loan officers and the racial gap (figure 4). These aggregate facts seem to suggest that.