Zillow’s Prediction of 2017 Homeless Numbers in Line With Actual Counts

A Zillow analysis published earlier this year that examined the relationship between rising rents and homeless populations accurately predicted the recently released 2017 point-in-time (PIT) counts, with a median absolute percent error of 8.3 percent. Predictions of the PIT counts in 17 out of 25 metro areas were within the 99 percent predicted interval.

In two areas – Los Angeles and Sacramento – the figures compiled by the U.S. Department of Housing and Urban Development came in higher than we predicted given how much rents rose in those areas. For example, in Los Angeles, Zillow’s analysis indicated that if rents rose 5 percent, then 1,900 additional people there would experience homelessness. The local count in January 2017 indicated that, in fact, 57,794 people in the Los Angeles metro area experienced homelessness, an almost 11,000-person increase from 2016. The increase was greater than expected, given that rents rose about 4 percent from 2016 to 2017.

In six other metro areas – Atlanta, Charlotte, Detroit, Houston, Miami and Tampa – the local counts were lower than Zillow predicted. Some areas have disrupted the link between rising rents and homelessness. Houston, for example, had one of the higher homeless counts in the country just five years ago. Now it’s dropping fast, so that a 6 percent decrease in rent over the past year translated into 426 fewer people experiencing homelessness in Houston.

The latest results will be incorporated into the dynamic Bayesian hierarchical model that Zillow uses for time-varying homeless count data and will be reflected in our next predictions.