Automated valuation models (AVMs) have been in Australia for 10 years or more, but only now are they starting to move into the portal space. As a free tool, they can offer a rather seductive hook for home-owners or buyers seeking a quick and entertaining estimate. Sites such as Zillow in the US have been very successful – with the ‘Zestimate’ playing a key role in making the site so popular.
“Zestimates were the sparkling new bauble leading the news media and the public through the door and allowing the company’s introduction to a massive audience.”
From a professional real-estate agent or valuer perspective, the results of the AVMs can be easily dismissed as a hit-and-miss affair. However in the case of the US marketplace, this valid criticism has done little to deter the fascination with instant online estimates. The success of sites such as Zillow and Trulia has granted them the capacity to continuously improve the accuracy of these models. Although models and data have improved, they will never be perfect.
Whilst an automated estimate is convenient and certainly popular, the accuracy issues warrant attention from rea estate professionals. Regardless of the popularity and ongoing improvements from services such as Zillow, we still expect AVM estimates to “miss the sale price of homes 25% of the time by more than 20% of the value”. Accuracy can vary significantly depending on location, which is an important point for Australian real estate agents.
These types of results applied to a $500,000 purchase may lead to an error of $100,000 or more. A home-buyer would be foolish to make a purchase decision based on an automated valuation alone. In practice, we expect Australian versions of automated price estimates to be used to promote contact with real estate agents, using the AVM as a starting point. Well-written and boldly highlighted disclaimers will only deal with part of the problem resulting from errors.
No matter what your opinion of the AVM, a basic knowledge of how they work and more importantly, how they don’t work, becomes an important skill for real estate agents in the future. These tools are here to stay.
High errors ultimately create problems with unrealistic expectations. Sellers will readily dismiss low estimates but will not ignore high ones. For an agent trying to win a listing, you now have to be prepared for this potential risk. Dealing with the challenge of other agents who try to buy the listing with unrealistic prices is nothing new, so dealing with a flawed AVM should not be an unsurmountable challenge.
A computer-based model relies on the subject property data, comparables data and a robust model. Get any of these 3 wrong and errors will result. Whilst in theory a complete database of all sales could be created that captures qualitative measures such as construction quality, street appeal and views, this theory is near impossible to apply.
A computer model can only process what it can count. This includes lot size, house size or even how busy a street is. However the AVM will often struggle in markets with houses with significant variation in quality.
Discrediting another agent is much harder than picking the errors in an AVM report. By keeping a file of bad AVM reports in your local area, you can instantly refer to these as part of your listing presentations.
Chances are that the vendor already has one or two AVMs already. As part of your preparation, generate an AVM report so you know what to expect. This service should be standard with your data subscription. This will instantly highlight why a machine can never replace local professional knowledge.
If sales volumes are easing and listings inventory building, chances are the AVM will have less data to work with. When a model has fewer comparables to choose from, it will often reach back further in time and use older sales, select different sized properties and adjust them or seek out sales from further distances. These techniques can be effective, but it does depend on your location. Poorly matched comparables can often change how an AVM performs.
As markets change so too do AVMs. What may have worked really well 6 months ago may not necessarily apply today, especially where sales volumes start to fall or prices start to change significantly. Knowing these facts will help.