The ML model also gauged the reasons for selling a property, including changes in household size or increases in the neighborhood’s population, or simply a desire to upgrade or downsize a home.
Five US counties – Marion, Brevard, Ventura, Arapaho and Clark – were chosen for the modeling exercise due to having “the cleanest data” and the highest likelihood of accuracy.
Sifting through the data, Tavant checked a total of 113 attributes – of those, about 10 were deemed the most important, as they determined the likelihood that someone was going to list and sell their home.
Sivert added: “Age is a huge factor; where your children are, the valuation of your home versus what you bought it for – not necessarily how much equity you earned but what the perceived value is.”
Sivert said it was now possible to predict when a home was going to sell with 50% accuracy, adding that predictive modeling could look as far as 180 days ahead.
