Automated Valuation Model

An automated valuation model (AVM) is a computer-based tool that combines statistical modeling with a residential database to generate valuation estimates for specific houses. Generally available online and at little or no cost, AVMs can offer useful preliminary indications of value but should be employed with caution. In addition, they have a controversial history of use in first-lien mortgage underwriting.

Results of AVMs are typically reported in a range, and a confidence score or margin of error may be included. The predominant statistical methodology employed is regression-based hedonic pricing. An AVM database contains a collection of home sales with information on transaction prices and some detail on physical property characteristics. Information may be collected from public sources, such as deeds and local government property records, or from private sources, such as multiple listing services and valuation reports prepared by licensed appraisers.

An AVM begins the valuation process with a search of its database for properties exhibiting key similarities to the subject property being valued. Search characteristics might include size of the home, age of the home, number of rooms, and maximum radial distance from the subject property. Once a set of transactions has been selected, statistical procedures are used to estimate relationships between each of the key variables and prices paid for homes in the subject market. These relationships are then applied to the subject property characteristics to derive an estimate of value for the subject property. This estimate is automated in the sense that human judgment is not employed at any stage of the AVM valuation process.

AVMs were developed to decrease the time and cost associated with supporting residential real estate decision making. Efficiency gains are accomplished by automating the valuation process, thereby eliminating the role of considered human judgment. This trade-off is the source of limitations associated with AVM use, which are widely acknowledged by both providers and users of AVM services. Since physical inspections are costly and time consuming, the actual physical condition and quality of homes in an AVM database, as well as those of properties being valued, will not be considered in an AVM valuation. AVMs are exposed to errors when properties deviate from the central tendency. Unlike the traditional appraisal process with its heavy reliance on human judgment, AVMs cannot detect subtle yet important differences in the relative attractiveness of neighborhoods and in the marketability of individual homes. Because AVMs depend on various data providers to populate their databases, data reliability is also an issue. Information verification is often sacrificed for efficiency, and the abundance of data in terms of comparability, correctness, completeness, timeliness, and market comprehensiveness is a concomitant concern.

Performance limitations associated with the machine-human trade-off have therefore called into question the accuracy of AVMs. AVM results tend to be least accurate in unusual valuation situations requiring the greatest amount of expert human judgment. These situations would include subject properties with extraordinary characteristics likely to escape the attention of automated processes, such as extensive remodeling or attractive views. Likewise, properties in thin markets with little comparable data challenge AVM accuracy, as do houses distinctive within their own neighborhoods. AVMs perform best where there are abundant data in standard valuation situations, such as the valuation of a typical house in a neighborhood of frequently traded homes that are all relatively homogenous with respect to age, quality, condition, style, floor plans, amenities, and location characteristics.

...