Technology

Data gathering for AVMs

December 2007

We are entering an interesting period in the property market. I will resist using a headline grabbing-expression to describe what may or may not lie ahead, but there is common agreement that uncertainty beyond the usual level of volatility is about to characterise the property market. Even though the overall market may seem stable some property segments may experience individual adjustments and stress. This is in contrast to the recent past of almost universally steady growth.

One of the common complaints of AVM users (and non-users) is that an AVM valuation is never fully reflective of the marketplace right now. Despite the complex methods that are used to arrive at valuations, it is often - quite rightly - said that one of the building blocks of the AVM, the 'concrete' sold price figure from the Land Registry, is typically three to five months out of date.

A fair observation? Let's assume for a moment that it is. What does this mean for a market that may have localised adjustments, or indeed is displaying a level of volatility that it is difficult to predict?

Data

AVMs use a number of data points to produce their output. It has been said, very fairly, that the accuracy of an AVM is a direct product of the data fed into it, and the algorithm used to produce the data.

Given that the algorithm is man-made, it is a fair assumption that each model has an algorithm of similar quality. Indeed, the movement of personnel between organisations in the AVM world almost ensures this. If we discount the commercially vital roles of marketing, relationship management and PR, the only technical factor that can with certainty impact the differential performance of an AVM is the data that feeds into it.

What is required is lots of high-quality and unique data - as much as possible - with the caveat that we need to know where it is from, and when it was produced, ideally as recently as possible. In an ideal world (an ideal valuation world anyway), the value of every property, today, would be known. How could this be collected? An impossible dream? Probably; someone could call all the homeowners and ask them, but they wouldn't really know. We could train an army of surveyors and allocate them a couple of streets each to report on each day, but this is clearly impractical. Unfortunately - or probably fortunately for a number of us and our own mortgage payments - some compromise is required.

Anyone with the desire to obtain it can receive Land Registry data, three to five months after the event. Data from surveyors is a highly prized asset in the world of AVMs, ironically since to an extent all AVM suppliers really do have the goal of replacing work traditionally carried out by surveyors - but that is another article altogether. There are some AVM suppliers who are fortunate enough to have robust long-term agreements in place with surveyors for data supply. However, it could still be argued that these values are also not always fully reflective of the absolute up-to-date market conditions, and depending on the delivery mechanism, can be a couple of months out of date. So how does an AVM keep right up to date with the latest valuation data and highly granular asking-price trends?

Estate agents

I have heard stories of estate agents filling in 'surveys', usually the work experience chap. I have come across tales of local newspapers being scanned for prices, and other more technical methods such as scraping websites for information. These are all stories without sources, so please take them with a pinch of salt and don't call your solicitors just yet. However, just as these stories have little value without sources, in the same way property value information (or data) has little value without the source being known.

Clearly, the best sources of information for the value of property today are those that are selling the property. Estate agents are at the front line of property sales and thus values. Before you leap to point out that asking prices are not selling prices, please wait.

Just a reminder of the definition of 'value' (in the property world): 'the estimated amount for which a property should exchange on the date of valuation, between a willing buyer and willing seller, in an arm's-length transaction, after proper marketing wherein the parties had each acted knowledgeably, prudently and without compulsion'.

As a wise old estate agent once said to me, there is no such thing as a house that won't sell, just a house that is priced incorrectly. Now say what you will about estate agents, but they are motivated by the sale of houses. No sale, no commission. Despite stories of inflated asking prices, I have to say that estate agents have a considerable incentive to ensure that a house is priced accurately. I suggest that the majority of inflated sale prices are driven by sellers' instructions rather than by estate agents raising expectations. I digress: my point is that with proper interpretation and a high volume of data, asking prices provide a very sensitive and predictive assessment of current market values; an accurate haircut based on local data can be applied to translate an asking price into a selling price,. Having statistically sound and comprehensive access to asking prices can provide a vital ingredient in arriving at a superior AVM performance in the market conditions that prevail right now.

I believe that the estate agent is actually the best friend of those wanting to value a property. How might one sensibly gather all the estate agent's data together? Certainly not by 'surveys' or any other time-consuming non-value-added (for the agent anyway) task.

There is however somewhere that some 90 per cent of the UK's estate agents do submit their current asking prices, along with pictures, floor plans, maps, measurements and descriptions on a daily basis to a central source. Property website Rightmove receives up to date, detailed property value assessments on almost every property put on to the market on a daily basis. Useful information, undoubtedly, but even more useful when one considers that for one or another reason, 20 per cent of these properties never develop into a transaction and are thus not seen by the Land Registry.

Would one be happy taking this as the only source of data for an AVM? Probably not, but fortunately that is not a choice that has to be made.

Property index

It is undoubtedly useful to know exactly what a property sold for at a point in the past. The majority of valuation suppliers use an indexation coefficient to forecast forward values to today's date. Rightmove is no exception, using all the current surveyor and asking prices to create a uniquely detailed index to bring the value up to the current day (see Figure 1).

Before I continue, a short comment about property indexes. It is important not to confuse the property indexes that we read about in the press from Nationwide, Halifax and Rightmove with the indexes that are used to forecast forward prices in AVMs. The press-friendly individual indexes generally refer to the whole of the UK market, and whilst a good indicator of overall price movement, would be of little use in indexing the price of an individual property. The indexes used in AVMs drill into individual postcodes, then property types, and can be further subdivided into value quartiles, giving incredibly detailed, specific indexes that can be applied to an individual property with real meaning. The index is enhanced by having access to up-to-date prices of property being marketed now.

Rightmove is not in the AVM business by accident. Being able to perform a comparable property data search from 90 per cent of the UK's estate agents is a compelling data proposition. This makes the range and number of comparable properties used within the algorithm very high. Selecting up to 100 comparable properties, based on location, property type, number of bedrooms and a host of further features, allows the selection of the 10 closest matches to the target property being valued for comparable purposes.

Combined data

When the asking-price data is statistically combined with a model that uses Land Registry data and surveyor's data, we have a very robust, current model with traceable integrity. This is then further combined with the information gleaned from the comparable property details.

Not only price and value details, but pictures, floor plans, measurements and descriptions are all available to add real value to the valuation - and don't forget that a substantial percentage of these properties will not have made it onto the Land Registry records.

Rightmove lists almost a million properties at any one time, which means that with the Department of Communities and Local Government suggesting that there are around 25 million residential properties in the UK, Rightmove has access to sales prices, details and particulars of almost one in 25 of all UK properties on its website. Given the volume and churn of property sales since Rightmove started marketing properties (almost 10 years ago), there is an incredibly rich repository of information available.

Rightmove is fortunate enough to have agreements with some of the UK's largest surveyors to provide their valuation information (from full surveys only, not desktop or drive-by valuations), providing a slightly more timely, very accurate level of information.

The clear advantage of all this information is a rich source of data for an AVM. The integrity and accuracy of the data is of paramount importance, and the model uses all the sources in parallel to ensure the highest levels of integrity. It is able to track the whole transaction lifecycle, from a property being placed on the market, to the surveyors' valuation, to the sale and registration at the Land Registry. An example of the sort of output that backs this up can be seen in Figure 2.

To complete the circle and answer the question posed at the start of this article, it is the rich, timely feed of data from estate agents which allows the Rightmove model to respond so quickly, and potentially spot and calculate accurate trends in property prices.

Again, there are a number of circumstances when we would not advocate the use of an AVM (another article...), however, the current market movement and volatility should not be one of them.

Consider the important factors that make up an AVM - the quality and quantity of data input into it and when the data point was measured.

Alan Gearing is property services related director at Rightmove