Free to consumer price estimates are popular tools in North America, but they are not without problems. In the coming year or two we expect to see automated valuation modelling being used more widely in Australia – offering either existing or new portal entrants a way to stand out in a rather tough online marketplace. Assuming these tools will become more widespread, one of the best things an agent can do is learn a little about how they work and equally as important – how they don’t work.
Dozens of white-papers and books are available online on both AVMs and home price index methods. However with the exception of a few books written specifically for appraisers in the US, most of the material is written in statistician speak using complex language and symbols. The authors seem to be more interested in impressing their geeky peers than to explain the topics in simple terms.
A computer generated AVM (automated valuation model) uses sales data and one of 3 main techniques to produce a price estimate. The first is a simple ‘Index’ approach, taking the last sale date and sale price and adjusting its value according to market movement. If the last sale price was ‘right’ and occurred within the last 7 years, it is often one or the more accurate models. However in volatile markets or suburbs with few sales to work with, the index can be less reliable. The same data problems that impact AVMs can also impact the Index itself, making it hard to accurately measure how much a market price has changed. So if you combine a dodgy last sale price and a low-turnover market, the Index method can return poor results.
The next method is often referred to as an ‘hedonic’ or ‘regression’ model. Think of this model as estimating a square metre rate for the lot size and a value per bedroom and bathroom. By fitting a straight line to the data, we expect lower prices for smaller homes and higher prices for larger homes, returning a slope value. Once we know the ‘coefficient’ values for lot size and house size, we can estimate the price of the subject property. In homogenous markets with high volumes of sales, the regression model can often work well. However where the quality of dwellings varies significantly or where some streets are more valuable than others, this model starts to struggle.
The third model seeks to match a handful of properties to the subject property. Many AVM companies around the world refer to this approach as the ‘expert model’, as it endeavours to emulate the process used by a professional valuer or real estate expert. This can include looking at sales from the same street and similar streets, matching similar sized lot sizes and house sizes. This approach can also leverage aspects of the two earlier models, using the index to adjust older sales that enter the final comparables list according to any market movement. Size differences may exist, even after the comparables have been selected based on how well they match the subject property (lot size and house size). For example, if one of the comparables is 50 sqm larger in lot size than the subject, the computer can automatically adjust the sale price by the coefficient value ($n x 50 sqm).
Whilst often superior to the regression model, the expert model also has its limitations. The computer will only ever match sales based on what it can measure; this includes time, distance from the subject property, street type, lot size and house size. But as any real-estate agent would know, these measures can often only account for part of the price. Factors such as views, construction quality and street appeal are just a few of the variables that can determine prices in your area.
Challenges for agents
Agent web sites in the US are great sources for stories about ‘Zestimates’ and how they can often adversely impact a listing or sale.
As one first-time buying couple discovered, the Zillow’s ZEstimated value range on a potential new home was $1,100,000 to $1,200,000. SO, while the listed price on the home was over $1.4 million, these buyers cited the Zillow ZEstimate as the reason for their low-ball offer of $1.2 million. Needless to say, the property was sold to another buyer, and at a price well over the Zillow estimate
The problems most commonly highlighted include buyers or sellers not using the AVM as the starting point in the review process. In the case of buyers – some will gravitate to the AVM that is below the value of the listing to support a low ball offer. In the case of sellers, some are drawn to the AVM result that is higher than the true market value. Both situations have the potential to become tricky for agents.
On the whole, most buyers and sellers are smart enough to know the limitations of the models. However agents really need to be equipped to handle every situation.
- As you prepare for a listing presentation, run one or two AVMs. If included in your data subscription, you won’t need to worry about transactional costs. If the AVM is available for free online via a portal, this should not be too hard to do. Check if the AVM result is accurate or not relative to the price you think is right.
- Know your local market and how AVMs perform in your area. If the models are always ‘low confidence’ the AVM report will usually display a red traffic light. Know how to explain this to buyers and sellers who may not notice or appreciate its meaning.
- Be aware of how unique properties perform via an AVM.
- Keep of file of AVM reports that show high errors. Showing a few dud results will come in handy when explaining the limitations of AVMs to a vendor.
- Consider adding property data and AVM methods to your team training.