AI-driven bond trading systems were turned off during March's wild selloff, as humans stepped in
Here's yet another example of AI's inability to handle unfamiliar circumstances, and how human oversight remains essential
Humans Top Bots in Covid Crisis Test of Electronic Bond Trading
Algorithms failed in volatile markets, then people stepped in
Market knowledge still needed despite push for coding skills
By Molly Smith and Matthew Leising
It turns out the corporate bond market still needs traders.
The algorithms that dealers use to buy and sell bonds with their customers failed in March at the height of extreme volatility from the coronavirus pandemic, according to investors and price data. The nimble analysis of flesh-and-blood traders was suddenly needed to price bonds, edging out machines that normally can trade large portions of the market without any human input.
The bond market has been one of the last corners of finance to move into the digital age, slowly modernizing from the rise of electronic trading to new venues that will remove much of the interpersonal communication from the process of selling company debt. Yet even as digital platforms set record volumes in the first quarter, market watchers said the bots failed a test when Treasuries and credit spreads were so disorderly.
“Good old-fashioned blocking and tackling is still very much a part of the business,”said Chris Coccoluto, head of investment-grade bond trading at Manulife Investment Management. “It was almost nice to see in a way you had to rely on your relationships.”
The virus-related disruptions were profound. At the height of the volatility, Treasuries rallied to record low yields and corporate bond spreads gapped out to levels last seen in 2009. The securities are linked as the vast majority of investment-grade bonds trade at a premium to Treasuries, adding an extra layer of complexity when neither market was fully functional.
While humans were able to spot the new patterns quickly, the bots couldn’t adapt because algorithms are built on historical data, said Chris White, founder of advisory firm ViableMkts LLC. “This is a reminder to a lot of people who may have been vilifying human interaction in the bond-buying process,” said White, a former fixed-income executive at Goldman Sachs Group Inc. “When things get really volatile, people become extremely valuable to the process.”
For decades, banks have stepped back when prices are unpredictable and buying too much from a customer could trigger multimillion-dollar losses in just days. At the end of March, dealers cut risk-taking mainly by shutting off algorithms that were spitting out incorrect prices left and right.
With Wall Street pulling back, asset managers stepped up to fill the void, according to MarketAxess Holdings Inc. data. Voice trades -- in which counterparties agree to a price over the phone, but process and hedge digitally -- rose to a record, according to Tradeweb Markets Inc.
Roughly 70% of the investment-grade corporate bond market still trades with some element of human interaction, especially larger transactions over $2 million. The record credit trading volumes handled on MarketAxess, Tradeweb and Trumid Financial LLC show how traders took advantage of platforms to execute smaller, simpler transactions electronically, which in turn allowed them to focus their attention on more difficult deals that require complex analysis, said Chris Bruner, head of U.S. credit at Tradeweb. “Taking out any of the steps of manual work flow was really important in March so people could focus on risk,” Bruner said. “They were less worried about exact price, and more worried about moving an entire risk profile.”
That also led to record volumes in portfolio trading, a relatively new practice that can price and sell a bundle of hundreds of bonds in minutes. Barely a concept three years ago, it is now a fast-growing part of the market. New data-analysis tools that allow prices to be fully automated are part of the reason that traders have seen their ranks greatly thinned in recent years.
The pace of tech innovation and disruption won’t be slowed by the events of March -- in fact, it will continue to gain speed, said James Switzer, global head of fixed-income trading at AllianceBernstein Holding LP. In March and April, his firm boosted by 500% its usage of MarketAxess’ all-to-all protocol, which allows for anonymous trading and can match investors with each other, as well as banks, on the other side of a transaction. “That’s what’s going to come out of this, a desire by the buy side to embrace all-to-all trading because we can’t always depend on dealer balance sheets,” Switzer said. “If we have to find the other side of the trade in an anonymous all-to-all fashion, we’ll do it.”
Electronic powerhouses like AllianceBernstein and BlackRock Inc. have forged ahead to embrace technology. While both value having traders with market experience, they’ve also expanded recruiting criteria to include coding skills. Others like Mike Nappi, head of investment-grade credit trading at Eaton Vance, still like the old school way of getting the job done. “We don’t have a fully automated trading desk for a reason -- I want our traders to have a sense of what’s going on in the market,” Nappi said. “We’ll never live long enough to see who the winner is. But through our careers, there are going to be humans needed, just maybe not as many.”