The house price algorithm - a little too simple?
One consequence of the recession is that advertisers are very keen to get business. Here in Exeter, we receive several leaflets each week from local businesses, trying to get our business, especially in the service sector. We could have a different business or individual to clean our house every day for a month, have our garden redesigned every month, have someone do the ironing, clean the stove or shampoo the pets. And the deliveries include several magazines, which combine syndicated articles with local adverts. Today's blog is prompted by a piece in one of these magazines, which used the word "algorithm" in the opening sentence. (How many people read no further?) Suppose you want to buy a house in Exeter. You collect information from the estate agents, either in person or online. You find a house that you like. Let's say it has an asking price of £250,000. Before committing yourself, you log in to a property valuation website - and one is the dominant player in this online business, and find that it sold for £200,000 four years ago, and the property valuation website gives an estimate of £220,000 for it. So what do you think about the price? The piece in the magazine suggested that this is a frequent problem, and is a consequence of too much transparency, supported by a poor algorithm. That price of £200,000 is a single number; it conceals a great deal about the house when it was bought at that price, why it was for sale, what state it was in, and numerous other pieces of information. If it was a new house, the builder may have offered special deals for purchasers at the start of a new development, or at other times to help cash flow. It it had been an older house, what state was it in when it was sold? Did it need a lot of repair, leading to a low asking price? Was it sold to clear a debt? Was it sold in a hurry, accepting a low offer to get the property sold as quickly as possible? So what is to be done about the difference between £250,000 and £220,000? You, as buyer, wonder whether you should offer to buy at a price nearer to £220,000 than the higher one. The vendor feels that the website is misleading; so does the vendor's agent, who proposed the higher price. And it all follows because the property valuation website has used a naive algorithm to achieve the figure of £220,000. Their algorithm is a simple forecasting model based on quite sparse data. They do not have the sales price for a particular property every year. People do not move house that often. So they collect prices for a small neigbourhood, and prices for similar sized properties in a wider area, and generate their expected increase in value as a percentage by a suitable combination of these two datasets. But each house is individual, and the two surrogates (small neighbourhood, similar properties) do not give a very good correspondence to what happens to the house that is for sale. One could forbid such property valuation websites - but that would be pointless. Or the property valuation website could make a very clear statement about the difficulty of forecasting (even to giving a range of values). Or, perhaps most important, improve the algorithm. I wonder where there are other algorithms whose use is potentially unhelpful?