We design and implement statistical studies that test the reliability of automated valuation models (AVMs) for residential lending purposes. Boxwood's AVM validations measure numerous dimensions of AVM performance that guide lenders in understanding the strengths and weaknesses associated with day-to-day use of their valuation models.
Our tools and techniques also satisfy agency guidelines for the conduct of independent validation studies. For instance, we address a full range of statistical measures of AVM accuracy, variability and risk. Also, we ensure that vendor models are "blind" tested (i.e., we employ out-of-sample test methods) by gathering recent benchmark data from our comprehensive database of sales transactions to compare against retroactive AVM valuations.
Boxwood's modeling expertise, arising from development and maintenance of proprietary commercial AVMs, translates into unique insights into residential AVM model construction, confidence scoring and AVM performance. Our firm's 'insider' knowledge has proven extremely valuable to clients in understanding how to capitalize on model strengths while also managing the operating risks.
We have been fortunate over the years to perfom validation studies on behalf of a diverse group of U.S. commercial banks ranging from modest-sized institutions to the nation's largest mortgage lender. Client validation studies are customized to each bank's requirements, from specific geographies and vendor models employed, to client time tables and budgeted expenses. Boxwood's principal and co-founder, Michaell Taylor, conducts the studies. He has performed quantitative, statistical and financial risk modeling work throughout his career; and his Ph.D. involved a concentration in quantitative methods.
A portable overview of our AVM testing services is available here.