Abstract
Accurate and unbiased property value estimates are essential to credit risk management. Along with loan amount, they determine a mortgage’s loan-to-value ratio, which captures the degree of homeowner equity and is a key determinant of borrower credit risk. For home purchases, lenders generally require an independent appraisal, which, in addition to a home’s sales price, is used to calculate a value for the underlying collateral. A number of empirical studies have shown that property appraisals tend to be biased upwards, and over 90 percent of the time, either confirm or exceed the associated contract price. Our data suggest that appraisal bias is particularly pervasive in rural areas where over 25 percent of rural properties are appraised at more than five percent above contract price. Given this significant upward bias, we examine a host of alternate valuation techniques to more accurately estimate rural property values.
Highlights
Accurate property value estimates are an essential component of the mortgage underwriting process
A number of empirical studies have shown that property appraisals tend to be biased upwards, and over 90% of the time, either confirm or exceed the associated contract price
The processing time is calculated including both the time for training and for out-of-sample estimation with the following specifics: 1) as shown in Table 1, our training dataset contains 336,216 observations and our test dataset contains 84,154 observations; 2) each model contains 115 explanatory variables except that for random forest we further break down the data by state and estimate for each state the same specification netting out the 49 state FEs; 3) processing time for all algorithms are based on the same dual-core CPU @ 2.6GHz with 8GB memory and 1600 Max RAM speed
Summary
Accurate property value estimates are an essential component of the mortgage underwriting process. This table reports the out-of-sample performance (measured by R2 and RMSE) for different model specifications employing all records (All) in the test dataset and the subsamples associated with the sales price in the top (P75) and the bottom quartile (P25) of the sales price distribution.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have