Color and Cost of Misaligned CARES Act
65 Pages Posted: 28 Sep 2020 Last revised: 12 Oct 2021
Date Written: October 2, 2020
I utilize proprietary servicer call transcripts to document the implied cost in the CARES Act mortgage forbearance applications from people not in financial hardship. 13% Government-backed performing loan borrowers applied for forbearance of which only 1.5% were unemployed. I document selective verification of unemployment status by the servicer processing these forbearance applications for Non-Government borrowers. This selective verification has a racial undertone, i.e., African Americans apply at a 22.2% lower rate than White Americans as evidenced by communications initiated by them due to lack of financial sophistication. Among the borrowers who are informed about the CARES Act via communications initiated by the servicer, African Americans are dissuaded from availing forbearance (10.8% less than White Americans) and are offered loan modification or no relief. I also find that this discrimination permeates from the forbearance applications to the actual take-up rate of approved forbearance. I document taste based discrimination directly from these call transcripts using soft information from natural language processing.
Keywords: Selective Verification, Race, Forbearance, Machine learning, NLP, COVID-19
JEL Classification: R38, R28, K32, J15, H12, H81, G01, E61, E63, D80, D61, D12, D18
Suggested Citation: Suggested Citation