Automated Valuation Models: A property market perspective

The property market is increasingly adopting automated valuation models with speed, cost and the removal of 'human errors' among the driving factors, but adoption also comes with challenges.

Many people read the term Automated Valuation Models (AVM) and picture a fully, 100% automated process, where a valuation is produced by a computer with little or no human intervention. However, the reality is a much broader spectrum of hybrids involving varying degrees of automation, digital data sources, and different levels of human involvement and interventions.

Driving factors

Reasons for the development and increased adoption of AVM-type approaches are relatively straightforward and mainly client driven. They can be summed up as speed, cost, scale, and consistency. Speed and cost are clear benefits from a client perspective. In many cases, AVMs are also seen by the market as providing a level of consistency when compared with manual valuations and in removing ‘human errors’ from the valuation process.

An additional, and unforeseen factor has been the current pandemic, which has accelerated many existing trends, and for physical property assets, has placed significant constraints on physical inspections and travel.

Where AVMs work best

AVMs work best with widely traded, homogenous assets. Their performance degrades with increasingly heterogenous assets or assets that are thinly traded.

For real estate assets, it is unsurprising that residential property – particularly classes of property that are traded reasonably frequently and have similar characteristics – has led the adoption of AVMs in mature markets. As an asset class, residential property has good data availability for lending, mass appraisal and consumer facing valuations.

Even within residential property, there is a risk in applying AVMs where properties don’t fulfil the basic criteria of homogeneity and a sufficiently liquid and transparent market.

Consumers may use some form of AVM to get an idea of value when considering entering a property transaction. Alternatively, a broker may employ an AVM to calculate value when marketing a property. For both re-mortgage and origination purposes, a lender may use an AVM to support their <strong>underwriting<strong> process in addition to other considerations such as affordability. In all these cases, consumers are owed clarity and transparency regarding how these values are being calculated.

In the Instant Buyer (or ‘iBuyer’) business model, companies purchase residential properties directly from private sellers and eventually re-sell them. With the emergence of this model, there is a strong need for consumer protection and education to ensure fairness on prices offered versus resale values. However, with the US operator Zillow withdrawing from this market in November 2021, citing market volatility, the long-term sustainability of this business model is unclear.

It is also worth highlighting that institutional build-to-rent, or ‘multi-family’, to use the US term, is increasingly being valued using AVMs in many markets.

The nature of effective AVMs in commercial real estate (CRE) and the data sources required are very different from residential property. This is because the various classes of CRE represent much more heterogenous and thinly traded assets than residential. However, there are many instances of the development and application of AVMs in markets where assets are being traded with similar characteristics and a sufficiently deep enough data set of property attributes and market data. Some of these AVMs are producing capital valuations on the assumption of vacant possession, with others focusing on forecasting market rents.

Another active role for automation is in the creation of various market indices around the movement and forecasting of market metrics such as rents, capital values, and yields at sector and subsector level. At portfolio level, we are already seeing the implementation of AVM approaches for CRE, albeit with the caveat that the efficacy and accuracy will tend to degrade as you drill down into segments, subsegments, and individual assets themselves.

Data requirements

One of the oldest phrases in computing is ‘garbage in, garbage out’. While the same principle applies to the data underpinning a valuation produced without the use of an AVM, the efficacy of any AVM is underpinned by the data used to develop and operate it.

At a simple level, we need data that is of a high quality since no model can overcome a lack of data or data that is erroneous. But what does it mean in practice?

Quality in the context of AVMs means considering the following factors, with an emphasis on transparency around every aspect of the data sources being used: recency; availability; security, privacy, ownership, and ethics; provenance and lineage; assurance; consistency; collection methodology; scale and range.

Another consideration on data sources for AVMs, and indeed for non-automated valuations, is the increasing variety of data being used as part of the valuation process. Some of these data points can be seen as proxies for more traditional attributes, such as using crime rates to assess the attractiveness or otherwise of the location and region, Tripadvisor/Airbnb ratings, air quality or broadband availability. In the case of CRE, there is increasing use of location type characteristics, such as local amenities and the quality of transport links, in a way that is already well embedded for residential properties. 

With the rise of environmental, social, and governance (ESG) measures as a driver of value, both AVMs and non-automated valuations need increasing access to various data points around energy performance certification and accreditation schemes. These include, for example, Energy Performance Certificates (EPCs) and the Building Research Establishment's Environmental Assessment Method (BREEAM). Ideally this data should include actual energy performance in addition to certificates of theoretical performance. The need for ESG data is becoming critical across all forms of valuation to firmly establish the correlation and causation of links between ESG and value for both residential and CRE properties.

Caution needs to be taken when adding additional data sources during the use of an AVM, since the effect on the models needs to be measured and the models recalibrated against external reference points.

Resolving risks

AVMs provide many opportunities for the marketplace, but they also have risks. At the highest level there is a potential danger that we increase the use of AVMs without fully understanding these risks and putting in place the appropriate management and mitigation. Risks include, but are not limited to:

• Bias - a general risk with many algorithms and AI approaches;
• AVM outputs being used as indicative numbers to drive perceptions of value, particularly for consumers; and
• AVMs developed in jurisdictions and/or for asset classes where data and model maturity is not sufficient for robust and accurate outputs.

An overarching due diligence framework can help reduce AVM-related risks. Market participants such as valuers, users of valuations, regulators, professional bodies, insurers, and AVM providers should work to a common understanding to ensure transparency, consistency, risk management, and to protect market confidence in asset valuations. 

For a more in-depth analysis of AVMs see Automated Valuation Models (AVMs).

RICS published its AVM Roadmap in July 2021 and subsequently engaged with stakeholders across all the major world markets to understand the current levels of adoption of AVMs and to help shape its response. From July through to October it conducted forums, had individual conversations, and received written contributions from over 120 individuals, representing RICS members and firms (large and small), AVM providers, government agencies, lenders, insurers, and other regulators, standard setting, and professional bodies. Contributors came from North America, UK, Europe, the Middle East, and across the whole of the Asia-Pacific region.