A lesson drawn from an unexpected source has reshaped how BGO approaches one of commercial real estate’s most stubborn challenges: predicting cap rates.
More than a decade ago, Netflix improved its recommendation system through ensemble modeling — a method that blends multiple predictive techniques to generate stronger results. That same strategy is now revolutionizing how his team forecasts cap rate movements, according to BGO Chief Economist Ryan Severino.
“If it’s good enough to suggest your next binge session, maybe it can help project where going-in cap rates will go next,” Severino told GlobeSt. By pairing traditional statistical modeling with machine learning, BGO has begun producing sharper forecasts over the past six to nine months.
Severino’s tests spanned time frames and geographies — from Asia to Europe to North America — dating back to the early 2000s. The results impressed him.
“It’s impressive how good these computers are getting,” he said. “They’re better at finding patterns than human beings.”
The advantage lies in scale and speed. Early experiments ran 10,000 variables, each time-lagged by one to three quarters. That figure has since grown to 50,000.
“This is an insane amount of data correlations and interactions to track,” Severino noted. “We didn’t evolve to do that. We evolved to make important decisions, like keeping alive on the African veldt. You give the computer a bunch of data and it’s just better than human beings at finding the correlations and patterns.”
The first surprise, he said, was discovering that cap rates are probably more predictable than most analysts assume — if large enough data sets are applied. The process isn’t easy, but as he described — it's straightforward. The system’s power is its ability to test vast combinations of potential influences, even unlikely ones. “Would the price of hot dogs matter?” Severino joked. “Honestly, it very well could.”
Another advantage of AI-assisted modeling is the use of instrumental variables, which can serve as stand-ins when direct data are unavailable. For those familiar with statistics, Severino explained, the concept relates to confounding factors — instances where two variables appear correlated but are both tied to a hidden third factor. Instrumental variables help isolate more meaningful relationships.
Still, Severino cautioned, technology alone isn’t enough. “Subject matter expertise is important to see if there is a real connection or not,” he said.
Perhaps the most encouraging outcome of BGO’s AI-enhanced approach isn’t just the sophistication of the models but the direction of the forecasts.
“Maybe the important takeaway is we’re coming out of a place where valuations have been beaten down so much and cap rates have gone up so much that most of the paths forward are positive,” Severino said. “The overwhelming majority of paths going forward have cap rates improving, valuations improving, and total returns improving.”
© Arc, All Rights Reserved. Request academic re-use from www.copyright.com. All other uses, submit a request to TMSalesOperations@arc-network.com. For more information visit Asset & Logo Licensing.