Somewhere in the dry heat city of Palo Alto, Stephen Boyd is running an artificial intelligence (AI) lab financed by the world’s largest asset manager. Boyd, who also runs the electrical engineering department at Stanford, has been an adviser to BlackRock’s systematic active equity team for many years.
BlackRock is tight-lipped, but enthusiastic about the experiments cooking in the laboratory.
“At the lab, they’re taking a fresh perspective, separate from the day-to-day workflow,” says Jody Kochansky, BlackRock’s chief engineer and head of its Aladdin Product Group. “When you think about the daily grind in any company, you get very close to whatever problem it is that you’re trying to solve, and it’s sometimes hard to get that ‘unconstrained solutions’ mindset.”
BlackRock, which has $6.3 trillion under management, is delving into the world of artificial intelligence, along with JP Morgan Asset Management, Goldman Sachs Asset Management, Pimco and other big firms, all of them racing to build machines able to exceed the insights of human traders, portfolio managers, risk managers and clients.
In the West Coast lab, the programme extends beyond the work Boyd has done on alpha generation with BlackRock’s active equity team. The remit now is to alter the world of asset management, from risk models to operations to uncovering data anomalies.
Kochansky views artificial intelligence as a natural extension of the data-science efforts that BlackRock has been involved in for years. Greater availability of affordable computing power and storage, along with an abundance of new data that AI can learn from, have created new opportunities for asset managers.
The lab can bring a very different point of view to some of our most vexing problemsJody Kochansky, BlackRock
BlackRock has already had some success in the quantitative part of its asset management business, where it is using AI to trawl through data for drivers of stock and bond returns. Ninety-two per cent of BlackRock’s systematic active equity (SAE) assets, which use technology to improve investment insight and generate alpha, were above benchmark or peer median for the five-year period. Boyd has consulted with the SAE team since 2013 and before joining formally to lead the AI lab this year.
AI is also helping the firm spot potentially dangerous anomalies in its operations, such as miscalculations of risk and trade-settlement issues.
“The lab can bring a very different point of view to some of our most vexing problems,” says Kochansky. “That’s really what the AI lab is designed to do.”
In 1988, the year BlackRock was founded, broker-dealers used mainframe computers to analyse yield curves to sell securities to the buy side. But, after purchasing them, the buy side had to rely on the broker-dealer to periodically run analysis for them on the progress of the securities – giving the broker-dealer an advantage in trading information over the buyer.
So BlackRock built a workstation that could develop analytics at a much lower cost and leveraged those analytics to build better portfolios.
Fast-forward to today and BlackRock’s Aladdin system assesses the risk on $15 trillion, roughly 7%, of the world’s financial assets.
“When I started with the firm in 1992, among the things that you had to do was run analytics at the individual security level. But then, of course, you needed to roll it up into portfolios and benchmarks, and be able to compare them all,” Kochansky says.
“So my job initially was taking a stack of paper off the printer containing the portfolio reports and I would go through every page, one at a time, to detect anomalies,” he adds.
It soon occurred to him that he could write a computer program to sift the data, effectively automating his own job, and that the program would be better at it than he was. “It’s about applying a consistent, repeatable process to the data you have available to draw whatever kinds of conclusions you might want to draw,” Kochansky says.
Years ago, BlackRock used data on mortgage-backed securities to better understand the risk profile of every portfolio. Generating the prepayment models to weigh that risk required large amounts of information, and it still does.
Nowadays, however, BlackRock is harnessing alternative data to predict prepayments. Loan-level data for private-label mortgage securities has been available for 10 or 15 years, but BlackRock is streaming consumer data from credit-scoring agencies such as Equifax and TransUnion to augment the predictive capability of its models. Bond investor Pimco has been doing something similar.
Kochansky says building these models has been fascinating. “You’re never really done, because the moment you’re done, the world changes in a way that you need to refit your model,” he says.
Lately, he has been focusing on accelerating the use of machine learning, natural language processing and other AI tools for BlackRock’s investors. This has meant centralising the firm’s previously scattered data-science efforts into the data-science core team, while preserving their ability to understand the business context of the problem they’re trying to solve.
As Kochansky puts it, the company needed a data community “to share approaches, share data and commiserate on problems”.
It’s likely that whatever Boyd is cooking up in BlackRock’s AI lab will end up in their hands.