Challenging the Conventional Wisdom: Do Declining Markets Negatively Impact AVM Accuracy?

Introduction and Overview

Recent changes in residential property values have prompted the question: How well does an automated valuation model (AVM) function in turning or declining markets? For several years, real estate prices rose steadily in most areas and very rapidly in some. In the last twelve months, many markets have turned downward, raising the issue of how well an AVM functions in an environment of declining prices. The present study is designed to answer that question by carrying out a due-diligence analysis on the ValuePoint®4 with Dual-Core Technology (VP4) automated valuation model (AVM).

AVMs do allow for the passage of time, using an index or some other methodology to adjust recent comparable sales prices and estimate what they would have been at the time of valuation, applying that information to the subject property valuation. In principle, there should be no insuperable asymmetry between rising and declining markets – between making a positive or negative adjustment over time. AVMs should not break down merely because a market is trending down instead of up.

The present study attempts to empirically and statistically verify if this principle holds true for VP4. In this paper, the performance of the VP4 AVM during the second and third quarters of 2007 is examined in detail, with particular attention devoted to comparing the diagnostics with its performance during the second and third quarters of 2006. To increase the study’s mathematical rigor and the reliability of its results, this study was characterized by an exhaustive approach, examining a national population of residential property sales drawn from the extensive First American CoreLogic property database, rather than looking at a small sample.

We know and expect that any AVM would face greater uncertainty in a rapidly rising or falling local market in contrast to an environment of modest and stable price moves because the underlying prices in its data set will have greater spread or divergence from each other. We also expect that any AVM would face difficulty in markets that quickly turn from rising to falling or the other way around, until its database of comparable sales and its index or other adjustment mechanism caught up and recognized the new direction and trend.

When these two known and expected effects are recognized and allowed for, any residual difficulty that VP4 faces in declining markets will be shown to be modest. More specifically, this study will show that in fact VP4 has very moderate bias in declining markets when it is measured empirically and when measured statistically the bias is proven insignificant or said another way indistinguishable from zero.