Indicators of collateral misreporting

Abstract

Automated valuation models (AVMs) are widely used in the valuation of pools of residential mortgages. In this paper, we provide evidence that high pricing errors in human and automated valuation models as well as rejection of loans with relatively low appraisals may generate the appearance of appraisal bias even when the original appraisals are unbiased. We also present evidence that the estimated frequency of appraisal overstatement is quite sensitive to assumptions about loan denial rates due to low appraisals as well as the size of and the correlation between AVM and original appraisal pricing errors.