Reigo Investments Uses AI for $100M Bridge Loans Securitization

Reigo Investments, a global investment firm based in Israel, working with Cantor Fitzgerald, announced the closing of a $100 million residential bridge loan securitization.

According to a company release, “The securitization features a 24-month revolving period, after which the transaction will begin to amortize. The pool of loans is business purpose loans, 100 percent senior positions with 6-24 months term. In addition, participation and whole loans will be included in the securitization. Property types include SFRs, multifamily, mixed-use, land, and construction loans.”

“When we started Reigo three-and-a-half years ago, we wanted to take data science and real estate and create value,” Yariv Omer, co-founder and CEO, tells GlobeSt.com.

During the startup phase, the company spoke with many analysts. “We found out they’re using the same Excel models with the same parameters that have never been changed,” Omer says. That would suggest that anything firms could learn about what makes good investments can’t get incorporated.

What Reigo does for investments is use machine learning and big data to analyze past loan decisions to build predictive systems. Working with institutional investors in both Israel and the U.S., the firm identifies hundreds of parameters and tries to correlate them with loans to determine which are likely not to default, which might default but are likely to allow recovery of money, and which would likely default and not return investor funds.

“It’s a classification problem,” Omer says, referring to a classic use of machine learning in putting things into fixed categories.

When examining a new loan, the systems first obtain additional information about the loan, parties involved, and property from historical sources. Then the software tries to match the loan to other historical loans and their performance.

There are inherent potential problems in machine learning, which involves examining previous examples of decisions. The A.I. programs treat the examples used to train the software on what to look for as correct. If those examples don’t provide optimal results, machine learning can provide a faster way to make the  same mistakes.

One approach Reigo uses is to train systems on one portion of previous cases and then test the resulting decision algorithms on another portion.

Source data can also provide problems. “This is one of the most challenging things in real estate and private lending,” says Omer. “There is no one source of truth.” Instead, the company uses multiple sources and compares one set to another.

As a third risk management approach, the software doesn’t provide final decisions. Rather, it compiles sets of recommendations that go to a human investment committee, which makes final decisions and provides feedback to the machine learning system.

“Almost every day we test a new model to see if it’s better or not,” Omer says. At some point in each month, the company reviews the model changes and then makes “small, incremental steps.”

The approach has worked. Non-performance rates for loans that Reigo authorized were much lower, Omer claims, than the overall non-performing rates for the market.