Automated valuation model

Automated valuation model (AVM) is a computer-based system that uses mathematical algorithms and statistical modeling to estimate the market value of real property. Zillow launched its Zestimate feature in 2006, becoming the first consumer-oriented AVM to reach widespread public use^[1]. These tools have transformed how lenders, insurers, and investors assess property values across the real estate industry.

Applications in mortgage lending

Mortgage lenders employ AVMs throughout the loan lifecycle. During origination, these models help underwriters assess collateral risk quickly and cost-effectively. Prequalification processes benefit from instant valuations. Appraisal quality control uses AVMs to flag potentially inflated human appraisals.

CoreLogic, the largest AVM vendor, maintains approximately 4.5 billion property records spanning 50 years from nearly all U.S. counties^[2]. Freddie Mac's Home Value Explorer has been used internally for over 15 years for portfolio and risk management. A recent study found loans originated through Freddie Mac's AVM were 9.6 percent less likely to default than similar loans backed by traditional appraisals.

The regulatory environment has grown more favorable. In 2019, the FDIC, Federal Reserve, and Office of the Comptroller of the Currency jointly increased the de minimis threshold from $250,000 to $400,000 for residential transactions not requiring physical inspection appraisals.

How AVMs work

The technology combines multiple data sources and analytical approaches. Public records provide tax assessor values, sales history, and property characteristics. Machine learning algorithms compare target properties against massive datasets of comparable homes.

Two primary methodologies drive most AVMs. Hedonic models use regression analysis to estimate how various property features contribute to value. Square footage, bedroom count, lot size, and location each receive statistical weights. Repeat sales indices track how individual properties or comparable homes have traded over time.

Data inputs typically include:

• Tax assessor records
• Multiple listing service information
• Recorded deed transactions
• Geographic and demographic data
• Historical price fluctuations

Accuracy and confidence scores

Zillow reports its Zestimate achieves a nationwide median error rate of 2.4% for on-market homes^[3]. Geographic variation remains significant. In Nevada, 90% of Zestimates fall within 5% of eventual sale prices. New Hampshire sees that figure drop to 69.9%.

Freddie Mac developed the forecast standard deviation (FSD) metric in the late 1990s. High confidence requires an FSD of 13 or less. Medium confidence falls between 13 and 20. Higher values indicate low confidence. Commercial providers like CoreLogic and Equifax submit to third-party testing from independent raters. Consumer platforms conduct their own reliability testing.

Historical development

AVMs emerged in the early 1990s as computing power expanded and property databases grew more comprehensive. By the early 2000s, lenders had widely adopted the technology. The 2008 financial crisis brought increased scrutiny. Many blamed over-reliance on automated tools for contributing to the housing bubble.

Zillow, founded in Seattle in 2006, made automated valuations accessible to ordinary homeowners. Redfin, Chase, ReMax, and RocketHomes followed with competing products. Today's models incorporate artificial intelligence and digital photo evaluation capabilities.

Limitations and concerns

AVMs cannot assess property condition. A well-maintained home and a neglected one may receive identical valuations if their recorded characteristics match. Major renovations not reflected in public records go undetected. Physical inspections remain irreplaceable for many purposes.

HouseCanary estimates only 40% of homes can be appropriately evaluated solely by an AVM. Data quality presents ongoing challenges. Outdated or inaccurate records produce flawed outputs. Rural areas and unique properties often lack sufficient comparable sales.

Performance inconsistency in communities of color and low-income neighborhoods has drawn regulatory attention. The Brookings Institution published research in 2023 examining how algorithmic bias may perpetuate historical appraisal disparities^[4].

References

• Fannie Mae (2023). Automated Valuation Model Guidelines.
• Freddie Mac (2022). Home Value Explorer Technical Documentation.
• Brookings Institution (2023). Governing the Ascendancy of Automated Valuation Models.
• International Association of Assessing Officers. Standard on Automated Valuation Models.

Footnotes

1. ATTOM Data Solutions. What is an Automated Valuation Model (AVM) in real estate?
2. Brookings Institution (2023). Governing the Ascendancy of Automated Valuation Models.
3. Zillow. How Accurate is the Zestimate?
4. Brookings Institution (2023). Governing the Ascendancy of Automated Valuation Models.