In a discussion with Declan King MRICS, managing director and head of real estate at ValuStrat, RICS proptech analyst Andrew Knight explored the development of AVMs, especially in the Middle East.
While AVMs have become common in the residential market in Europe and the US, especially for remortgaging, they remain a rarity in the Middle East.
King explains that a challenge in the region is a relative lack of trusted transactional data, which is plentiful in countries like the UK: 'Data is at the front of centre of what is needed for us to create trustworthy AVMs'. Dubai itself has reliable data availability but the other six emirates have no official data sets. ValuStrat, an RICS Tech Partner, has set about collecting trusted data from RICS valuers in the region across a wide variety of assets for its own AVM. King points out the importance of real estate businesses getting involved in developing their own AVMs so that the market can be led by valuation experts, not just tech companies.
But for AVMs to be effective, transparency is essential and institutional investors need to be prepared to share transaction data in order to reap the benefits – a significant change from how the market has traditionally operated.
AVMs are best where clients want bulk valuations and to make quick decisions with a lower spend, and where they are bodies that can mitigate against associated risk, banks being the main example given the large volumes of assets that may be involved.
King believes AVMs can open up new markets for bulk valuations that at present are too time-consuming and expensive to carry out with any regularity – AVMs can offer a cost-effective way of assessing value on a more regular basis. This is not to say it will be entirely automated – the valuer will still be needed, applying their expertise to the outputs.
ValuStrat head of research, Haider Tuaima, is a statistician with a software background and explains that their AVM is currently at an early Beta stage; they are road testing their models, reviewing the outputs and rankings to see if they seem accurate and whether algorithms may need to change. It is essential that the market can feel confident in automated models when they come into use. Tuima says that the AVMs will give a price range rather than a single number – they have compared a valuer's judgement to this and found around a 3% error rate for apartment blocks around the centre of this range, which should be acceptable. However, a 10% variance found for individual units mean that the algorithm is not yet ready in this area. Ultimately the AVM valuations may be best for assisting the valuer with comparisons as opposed to coming up with the final valuations themselves.
'An AVM should mirror how a valuer works', says King, meaning it should capture official transaction data, past valuations of an asset, listing prices/asking prices supplemented with indices, opinion capture, and so on. Freehold offices with a value per square metre will be well suited to AVM valuation, but more complex assets such as mixed-use buildings will clearly present a greater challenge.
Inspection will still be required for most buildings, King believes, to compensate for what some have called 'artificial stupidity' – an AI may not take account, in the case of homes, things like awkward-shaped plots, sloped gardens, smells from an industrial plant nearby; all things a valuer would note on an inspection. Valuers can currently deal better with markets in flux; an AI might tend to pick up changes too quickly or slowly, but a valuer can judge market sentiment and whether prices are outliers from their experience. AVMs nonetheless can act as a starting point and timesaver.