With a default rate less than 1% on municipal bonds, the odds are greater of betting on the right number in roulette than an issuer failing to pay interest or principal payments on its debt. That's why it was surprising that
Given its rarity, predicting muni bond defaults is a complex and difficult task.
"One of the challenges in predicting municipal defaults is that they occur relatively infrequently and often stem from unique local circumstances rather than consistent, repeatable patterns," said Jennifer Fredericks, Baker Tilly director of business development.
Though the problem is convoluted, some believe AI can provide a simple solution.
Munibonds.ai, which specializes in implementing AI into muni market analysis, believes its capabilities have advanced to the point where it can offer predictive insights into whether a muni issue will default.
Robert Kane, founder and CEO of Munibonds.ai, is working on a product he says can warn consumers about a default one to two years before it occurs. The product aggregates and analyzes issuance records, media coverage, and other publicly available data to spot warning signs of a municipal bond default.
"We're not fortune-telling," Kane said. "This is AI-based surveillance — real-time bond monitoring that catches the cracks before they widen."
MuniPro is also experimenting with predictive AI for muni defaults.
"In predictive pricing, market participants care about the max error, or the worst-case scenario of a model predicting future prices," the firm's CEO William Kim said.
Therefore, market participants need to first build a fundamental credit model, per Kim.
"We rely more on our internal ratings model built from the ground up," as many deals are unrated and rating agencies alone cannot be relied upon to predict defaults, he said.
Despite predictive AI's potential, some industry experts remain hesitant to embrace the concept.
"AI can be a valuable analytical tool, but its outputs still require human judgment," Baker Tilly's Fredericks said. "Like any model, it may identify correlations that warrant further evaluation before conclusions are drawn."
She voiced concern about the potential downfall of relying solely on AI research.
"The danger is, if you have investors that are considering buying and they're utilizing AI, that's only one piece of it," she said. "The quality and consistency of available data remain important considerations for any AI model. Because municipalities vary in how they report and disclose information, differences in underlying data can influence the conclusions generated by AI tools."
Some muni investors also have reservations about predictive AI.
"We do not have any specific knowledge or awareness of any AI models that have demonstrated the ability to accurately predict municipal bond defaults or directional credit trends for that matter," Andrew Clinton, CEO and founder of Clinton Investment Management, wrote in an email.
Research teams and more traditional analytics work better, he said, "We continue to believe the municipal credit research function ... most appropriately resides with experienced credit research and portfolio management teams with deep experience."
Jeff Timlin, lead portfolio manager for Sage Advisory's municipal bond strategies, expressed skepticism about the value of predictive muni default AI.
"I don't know how much better it would be, or even if it would change anything in the market," he said.
Timlin mentioned the tool could possibly benefit smaller issuers that are more susceptible to bankruptcies, however, information about those issuers might be harder to collect.
Addressing concerns, Kane pointed to his company's performance.
"We've gotten great feedback from customers," he said. "We haven't had clients come back to us to say, 'Hey, this is wrong.'"
And MuniPro, though it hasn't released a product, wants to make sure people have accepted the methodology, Kim said.
If municipal AI companies deliver, the market may start to alter its outlook on AI research and predictive analytics, he said.
Timlin said if predictive AI can accurately predict default before any other source, "the proof would be in the pudding."
— Jessica Lerner contributed to this report
