Machine learning will have the biggest impact on asset pricing decision-making, according to 52% of the property professionals polled by EY Australia & New Zealand and MIT Real Estate Innovation Lab during a recent webinar on the future of valuations. And yet 59% of the industry is still hamstrung by “skills and knowledge gaps.”
This came as no surprise to my co-host, Dr. Andrea Chegut. Andrea is co-founder and director of MIT’s Real Estate Innovation Lab, and teaches data science and machine learning at one of the world’s top universities. She has also spent close to two decades developing asset pricing models for commercial real estate, green buildings and digital infrastructure.
After a “sophisticated machine learning process,” MIT has uncovered a host of environmental, social and governance parameters – from daylight and views to the level of street-level greenness – influence financial outperformance.
For example, the Lab has found buildings with high access to views command a 6% rental premium.
MIT’s hard data confirms what many in the industry understand through instinct. Our audience was asked which building features would have the biggest impact on future valuations, with healthy building features coming out on top (33%), followed by flexible, collaborative space (21%), smart building technology (19%), connection with nature (15%) and energy efficiency (12%).
But failing to understand how design features impact long-term valuation can lead to “millions of dollars in lost revenue,” Andrea warned. And she had a clear call to action for real estate professionals who thought they could dodge the data and digital discussion.
How to build a hunting machine
So, what does this data science framework look like?
Dean Hopkins, Chief Operating Officer of Oxford Properties, and Joanna Marsh, Investa’s General Manager for Innovation and Advanced Analytics, have spent the last two years working together to build “hunting machines.” These sophisticated algorithms can be “pointed” at any real estate market to understand and unearth undervalued real estate.
Rather than “trading information over bottles of wine,” as Joanna said, a hunting machine looking for the next build-to-rent project analyses a data set with six million records and 30 years of data. Financial data – rents, leases, valuations and more – is overlaid with demographic and geographic information, from school profiles to development application histories. The machine can spot infinitesimal changes to the data and “the anomalous pieces of dirt or buildings that can be repurposed,” Dean added.
“Our models need to be porous enough to take in human weighting, knowledge and understanding and blend it with the data to augment it,” Joanna noted. Humans are then “supercharged” and can set to work on their highest and best use – which is relationship building and negotiation.