AVMs versus Appraisals in Refinances: Evaluating Anchoring Bias through Relative Risk Identification
50 Pages Posted: 11 Jun 2020 Last revised: 14 Sep 2021
Date Written: May 6, 2021
Extensive research has confirmed that there is significant anchoring bias in purchase mortgage appraisals, with appraisers setting the value at or above the contract price the majority of the time. However, appraisal bias in refinance mortgages has been significantly more difficult to measure due to the absence of a sales price, with most research relying on noisy valuations from AVMs or mark-to-marketing prior transactions to provide a baseline of value. To circumvent this problem, this paper utilizes a novel dataset of 9 million recent refinance appraisals from 2013-2017 that contain the initial estimate of value provided by the homeowner/loan officer in the loan application, a value that is not permitted to be shared with the appraiser. We find a limited amount of anchoring bias on this initial estimate of home value, with around 7% of appraisals "hitting the mark" exactly, concentrated among certain appraisers and at high LTVs. In the absence of a reliable baseline of value, we evaluate this anchoring through its impact on risk rank-ordering, using a novel Shapley-style methodology and find that an AVM-based LTV performs roughly the same as the appraisal-based LTV even for these biased loans. We then find the break-even level of confidence where AVM LTVs begin outperforming appraised LTVs.
Keywords: Appraisal Bias, AVM, Refinancing, Mortgage Delinquency
JEL Classification: G21, G28, G51, L85, R31
Suggested Citation: Suggested Citation