36410 — Automated Valuation Models (AVMs)
Overview
Node 36410 addresses automated valuation models (AVMs) — statistical or algorithmic tools that generate an estimate of property value from historical transaction data, property attributes, and market indices, without the direct involvement of a valuer in each individual assessment. AVMs are widely used in mortgage lending, portfolio monitoring, and preliminary desktop assessment, though their output is distinct in nature and reliability from an individual valuer’s professional opinion of value.
Most AVMs in the Australian market apply a form of hedonic regression, modelling the relationship between historical sale prices and property characteristics such as location, land size, dwelling attributes, and recent comparable transactions, to generate an estimated value together with an indication of confidence or expected error. AVM outputs are generally most reliable for property types with deep, homogeneous transaction evidence, such as established residential dwellings in active markets, and less reliable for unique, specialised, or thinly-traded property.
This node is intended for lenders, portfolio managers, and other parties considering the use of AVM outputs, applying the general methodology addressed in node 36100 and the professional standards addressed in node 36200 to the automated valuation context specifically.
Scope & Boundaries
This node addresses the methodology, use cases, and limitations of automated valuation models within Australian valuation practice. It does not address general valuation methodology as applied by an individual valuer, addressed in the 36100 sub-series, or the professional standards and reporting framework applicable to conventional valuation, addressed in the 36200 sub-series.
It does not address the underlying data sources feeding automated valuation models in detail, addressed in node 36420, or the broader suite of property technology tools used in valuation workflow, addressed in node 36430.
This node describes the general application and limitations of automated valuation models as at the time of writing. The appropriate use case, confidence threshold, and reliance placed on an AVM output for a given purpose should be confirmed against the model provider’s documentation and applicable lending or professional standards.
Key Concepts
Regulatory & Legal Framework
-
National Consumer Credit Protection Act 2009 (Cth)Australian Securities and Investments CommissionGoverns responsible lending obligations relevant to a lender’s reliance on AVM output in place of a full valuation.
-
Privacy Act 1988 (Cth)Office of the Australian Information CommissionerGoverns the collection and use of property and personal data used to train and operate automated valuation models.
-
ANZVGN guidance notes on automated valuation modelsAustralian Property Institute & Property Institute of New ZealandProvide guidance on the appropriate use and disclosure of AVM output.
Practical Application
Reliance on an AVM output is calibrated to the confidence score of the estimate and the risk associated with the purpose it supports.
| Scenario | Application | Outcome |
|---|---|---|
| A lender processes a low-risk residential refinance application | The lender relies on an AVM output with a high confidence score in place of a full valuation | The refinance is processed without the cost and delay of a full valuation, within the lender’s defined risk tolerance |
| An APRA-regulated ADI monitors mortgage portfolio impairment provisioning | The ADI applies AVMs across its loan book to estimate current security values | The ADI obtains a scalable, ongoing view of portfolio-wide loan-to-value ratios |
| A managed fund conducts a preliminary desktop check before commissioning a valuation | The fund reviews an AVM output and confidence score to determine whether a full valuation is warranted | The fund allocates valuation spend to assets where a full valuation is most warranted |
Cross-Series References
Findings are presented on the basis of data and evidence alone.
