Algorithms used to determine stores’ futures

Back in February, J.C. Penney said it planned to close down about 140 struggling stores. That might be just the beginning.

Analysts at System2 LLC, a big-data startup, have identified another 197 stores at risk. The outlets — from San Bernardino, Calif. to a suburb of Omaha, Neb. — have a more than 64 percent likelihood of closing.

How does System2 know? Its computers say so. Founder Matei Zatreanu threw aside a hedge fund career to test a simple theory: that crunching mobile phone pings, demographic information, creditcard bills and other unconventional data yields a better way to invest in real estate and other areas. Using algorithms and machine learning, the company says it can determine which stores have a future and which ones will die — and do it more accurately than investors who rely on conventional information. So far, it has only studied J.C. Penney Co., because its plans to shutter stores have been in the news, but it’s ready to look at more retailers, Zatreanu said.

“There’s new data that’s out there,” he said. “But then what we try to focus on is how do we use this data in a smarter way.” Zatreanu last year left King Street Capital Management, which oversaw $19 billion as of January.

Daphne Avila, a spokesman for Plano, Texas-based J.C. Penney, declined to comment, citing a policy against commenting on market or industry speculation.

System2 is trying to sell commercial-mortgage bond investors on a kind of analysis that is foreign to many of them: scrutinizing every single retailer tied to mortgages in their securities, and figuring out how each company’s fortunes tie into the health of a mall or mall-backed bonds. The edge that big data and artificial intelligence have given to money managers in equities may help commercialmortgage bond investors too, according to Zatreanu.

Looking closely at indi-

vidual loans backing a bond is more common among residential mortgage-security buyers. For commercial mortgages, less data has traditionally been available, and there can be more variables in any transaction because of the wide range of tenants and customers for the property, making forecasting much more difficult.

“There are so many idiosyncratic type events in CMBS [commercial mortgage-backed] securities,” said Kin Lee, a money manager at Angel Oak Capital Advisors. “I don’t think there’s a good way to necessarily model that out,” he said.

David Tawil, president and co-founder of Maglan Capital, a hedge fund that invests in distressed companies, said more sophisticated information can

be a helpful tool for money managers.

“Big data will help us either weed out a bunch of potential investments down to a very select few that we’ll then do research or, to the extent that we found one that we’re very interested in and have a very solid thesis on, it will help us either confirm or debunk the thesis,” Tawil said. He hasn’t looked at System2’s offering.

Traders have already bet a fortune on whether retailers can survive a stampede out of their stores and toward online shopping. J.C. Penney’s stock is trading near an all-time low and wagers against it climbed to 40 percent of shares outstanding in August, Markit data show. Meanwhile short positions on some of the riskiest slices of commercial mortgage-backed securities rose to more than $5 billion earlier this year.

About $760 million of mall

loans that support these bonds have entered a form of default or near default known as special servicing since March, according to Wells Fargo & Co. research. Real estate investment trusts are also in the line of fire, with more than 20 percent of space in malls run by CBL & Associates Properties Inc. and Washington Prime Group Inc. exposed to distressed or shrinking retailers, according to a Moody’s Investors Service report.

“It really does boil down to location, location, location,” said Gary Greenberg, a money manager at Payden & Rygel. “You want to avoid the tertiary mall or the mall that’s the weaker mall within an area that’s still struggling economically.”

Zatreanu by his own admission isn’t reinventing the wheel; instead his team looks at the kinds of metrics that owners of J.C. Penney stock or the

CBL REIT would consider to compile their list of the walking dead. That includes the number of people who visit each store, what a store’s competition looks like, where it is located and whether there are vacant shops nearby.

But Zatreanu’s edge is what he says are better ways to get data faster, and to interpret it. Investors don’t have to just wait for data from filings about spending at malls — they can get aggregated location pings from mobile phones to create a dynamic picture of which stores are being visited within a particular area or mall, and where those shoppers travel from. Average household incomes and home prices for that area can then be layered into this matrix, building up a profile of an individual outlet’s customers.

System2 gathered this kind of information on existing and closed J.C. Penney outlets. Using machine learning, software was trained to determine what it was about the dead stores that led to their demise. When System2 then ran this program for a set of open J.C. Penney stores, it showed nearly 200 shops faced an almost twothirds chance of dying. The model updates as new data sets become available, which can be useful for investors in commercial mortgage-backed securities.

“There’s no objective way of measuring that correlation, until now,” Zatreanu said. “Once we know this kind of information about the specific malls in there, we can start modeling these CMBS tranches much better. This is just the tip of the iceberg.”