Home / Wiki / A / Automated Valuation Model (AVM)

Automated Valuation Model (AVM)

Share

An automated valuation model (AVM) is a computer model that estimates the market value of a residential or commercial property from data, without a valuer visiting it. AVMs combine records of past sales, property characteristics such as size and location, listing data and house price trends to produce a value estimate, usually with a measure of how reliable that estimate is. Mortgage lenders, investors in mortgage pools, insurers, estate agents and property portals all use them.

AVMs sit alongside traditional valuations rather than replacing them in every case. A full mortgage valuation or appraisal by a qualified professional remains standard for many purchase loans, while AVMs are used where the risk is lower or speed matters, for example for a remortgage at a modest loan-to-value ratio. Because a poor estimate can lead to over-lending or unfair outcomes, regulators in the United States and the European Union have set expectations for how lenders govern and test them.

01

How automated valuation models work

AVMs use several modelling approaches, often blended together:

  • Hedonic regression. The value is modelled as a function of attributes such as floor area, bedrooms, age and location, in the spirit of a hedonic adjustment.
  • Repeat-sales indexing. The last recorded sale price of the property is rolled forward using a local house price index built from homes that have sold more than once.
  • Comparable sales. The model selects recent nearby sales of similar homes, like the comparables a valuer would choose, and adjusts for differences.
  • Machine learning. Gradient-boosted trees and similar methods learn price patterns from large datasets and can capture interactions that simpler models miss.

Most AVMs return a point estimate, a value range and a confidence score or forecast standard deviation. A high confidence score generally means many recent comparable sales and a fairly uniform housing stock; a low score is common for rural, unusual or high-value homes. Lenders usually accept an AVM result only above a set confidence threshold and within set loan size and loan-to-value limits.

02

Uses in lending and portfolio monitoring

At origination, lenders use AVMs to screen applications, to support lower-risk lending decisions without a physical inspection, and to check the reasonableness of valuer reports. After origination, they revalue whole portfolios to track current loan-to-value ratios, which feeds capital calculations, provisioning and stress tests. Investors in mortgage-backed securities and covered bonds rely on indexed or modelled values in cover pool valuation and loan-level reporting. In the United States, the government-sponsored enterprises Fannie Mae and Freddie Mac use their own models when deciding whether certain loans can proceed without a full appraisal.

03

Governance and validation

Because an AVM is a model, it falls within a lender's model risk management framework. In US banking this framework follows the 2011 supervisory guidance on model risk management issued by the Federal Reserve and the Office of the Comptroller of the Currency. Good practice covers independent validation before use, regular back-testing of estimates against actual sale prices, monitoring of hit rates and error by region and price band, controls over input data, and clear rules on when a human valuation is required instead.

Lenders that buy AVM results from outside vendors remain responsible for them and typically compare several providers, test them on their own sales data and keep a cascade that picks the best-performing model for each area.

04

Regulation in the United States and Europe

In the United States, the Dodd-Frank Act added a requirement for quality control standards for AVMs used in mortgage lending. The OCC, the Federal Reserve, the FDIC, the NCUA, the CFPB and the Federal Housing Finance Agency finalised a joint rule in 2024. It requires institutions to adopt policies and controls designed to ensure a high level of confidence in estimates, protect against manipulation of data, avoid conflicts of interest, require random sample testing and reviews, and comply with nondiscrimination laws. The last factor ties AVM oversight to fair lending concerns that models trained on historical prices may undervalue homes in some neighbourhoods.

In the European Union, the European Banking Authority guidelines on loan origination and monitoring set conditions for using statistical models to value and revalue property collateral, including documented model performance and periodic review by a qualified valuer. These sit alongside broader requirements on collateral valuation for lending under EU banking rules.

05

Limitations

AVMs cannot see the condition of a property, recent renovations, structural defects or local nuisances that a valuer would notice. They lag sharp market turns because they are built on past transactions, and their accuracy drops where sales are thin. Errors can be systematic rather than random, so a model that is accurate on average may still be biased for particular property types or areas. For these reasons lenders treat AVM outputs as one input within limits, rather than as a substitute for professional judgement in higher-risk cases.

Last updated 17 Nov 2026
Share this entry

Ref 1030000844