The Arms Race For Quants Hits Finance Hard

Finance was one of the first industries to recruit quants - and firms in that industry have demonstrated a willingness to pay up.

But the competition for those skills has intensified - and the larger question is whether, if everyone is doing it, how it offers a differentiated edge rather than an undifferentiated risk. JL


Justin Baer reports in the Wall Street Journal:

Traditional stock- and bond-picking firms are paying up to hire mathematical and computer experts. They want to dive into pools of data—and the machine-learning tools that harness that data and other information—in search of trading ideas and blind spots. (But) mindful of the crowds that form around popular ideas, some managers have used data and experts outside of the world of finance for fresh perspectives.“There will be new sources of data and it will help with investment decisions, but the real question is can an asset manager sustain an edge on those kinds of insights.”
Neuberger Berman Group LLC’s data-science team began the year with big plans, if limited resources.
“Just me and an intern,” group head Michael Recce said.
Six months later, the group has added a fifth member to its team of data scientists and engineers as it works through a backlog of projects for the firm’s portfolio managers. Neuberger’s investment managers have already poached some of Mr. Recce’s staff, leading him to recruit their replacements.

“Each town needs one,” he said, paraphrasing Alexander Graham Bell’s famous line predicting the telephone’s widespread use.
Data scientists, who have already helped usher media companies, retailers and other old-line industries into the digital age, first arrived at Wall Street banks decades ago as computing power and the complexity of derivatives surged. They built computer models that traded securities, giving rise to the quantitative-investing field that has changed how markets function.
At asset managers, these scientists aren’t about to outnumber the phones found on the trading floors. But many of these traditional stock- and bond-picking firms are now paying up to hire mathematical and computer experts. They want these recruits to dive into pools of data—and the machine-learning tools that harness that data and other information—in search of trading ideas and blind spots.“Most of it is at an early stage, and I don’t think it’s matured yet,” said Onur Erzan, a senior partner at consultant McKinsey & Co. “There will be new sources of data and it will help with investment decisions, but the real question is can an asset manager sustain an edge on those kinds of insights.”
Mr. Erzan estimates that many managers are spending, on average, tens of millions of dollars to beef up their data-analysis capabilities. The resources arrive at a tenuous period for investment firms as trillions of dollars have flowed into low-cost, index-tracking investments, and out of so-called active managers whose uneven performance could no longer justify higher fees.
Petter Stensland’s foray into data began in late 2012 when, as a junk-bond analyst at AllianceBernstein Holding LP, he needed to analyze the prospects of the many oil-and-gas companies emerging in the nation’s energy boom.
For many of these producers, their value—and their ability to repay their debts—hinged on how much oil or gas they could pull from the land they owned. The problem was there were no reporting standards on energy reserves, and the producers’ own views were subjective, Mr. Stensland said.
Mr. Stensland said he and his team began to compile data from the states’ oil and gas commissions, stitching together decades of history on oil wells’ output in each parcel of land (typically around 640 acres). Building Bertha, the database’s chosen name, took 15,000 man-hours. The analyst traveled to industry towns such as Midland, Texas, and Roswell, N.M., where he could copy older data preserved in oil and gas libraries’ microfilm.
By mid-2014, Bertha had persuaded the high-yield investment team that certain producers were at greater risk of defaulting on their debt. The firm sold the shakiest bonds.
Those insights helped the AB High Income Fund sidestep steep losses that year on junk bonds, a selloff that took down Third Avenue Management LLC’s Focused Credit Fund that December.
In 2016, the firm launched a private-equity fund that uses Bertha to help identify undervalued oil-and-gas companies.
Mindful of the crowds that form around popular ideas, some managers have used data and experts outside of the world of finance for fresh perspectives.
Joe Byrum had been analyzing how new corn and soybean hybrids might grow under hundreds of different climate and soil conditions when he fielded a job offer from an asset manager located near the center of the U.S.’s corn-producing heartland. Mr. Byrum joined Des Moines, Iowa-based Principal in July and now leads the $674 billion investment firm’s data-science efforts.
“What you find is industries do certain things really well,” said Mr. Byrum, who had run the research department at Syngenta, an agricultural company. “There are plenty of opportunities to apply those techniques and approaches” to managing money.
One project that is already under way at Principal: Preserving and cataloging the expertise of the firm’s portfolio managers. It might be how investment staff responded to a market crisis, or just simply preserving the daily routines that at one point, become “like driving a car—you don’t think about it.”
“These great people retire, and that expertise is gone,” he said. “The opportunity is to capture that knowledge.”
Pacific Investment Management Co., one of the world’s largest bond investors, assembled its database of home-loan information more than a decade ago. It is massive, with 50 billion details on some 200 million U.S. mortgages.
Pimco built models that could analyze that information, and predict when borrowers might prepay their loans. But because those models were so labor-intensive, the firm could only sample 2-3% of the mortgages at a time, said Mihir Worah, the firm’s investment chief of asset allocation and real return.
The firm joined with a research university, whose data-science department developed computer-based machine-learning tools that make short-term predictions gleaned from patterns detected in historic information.
Those tools have allowed Pimco managers to sample as much as 20% of its mortgage database—and deliver more accurate forecasts on the groups of borrowers more likely to refinance or sell their homes in the next month, he said.
The information can help them value mortgage bonds the firm has held in various funds.