| Illustration by Aleksandar Savic. |
For better or worse — and there are arguments on both sides — computers rule the trading world. Algorithms and automated-execution techniques that began to boom in the 1990s became ubiquitous in the 2000s. Today a majority of stock trades are executed electronically, many of them tied to quantitative strategies. Virtually all transactions are automated at some stage, and the trend is spreading to other asset classes. With automation came the need for speed — which kicked off an arms race to build faster and faster computing and communications capabilities to minimize the delay, or latency, in filling orders.
Now ultralow latencies are commonly measured in millionths of a second. That may be about as far as the speed race can go. In a maturing marketplace where incremental advantages are harder to come by and fleeting when achieved, market players are looking for the next technological wave — and they are converging on artificial intelligence.
Speaking to Institutional Investor early this year, KCG Holdings chief technology officer Mike Blum observed that high-frequency trading had evolved in stages, from “conquering complexity” in market structure to maximizing speed. “The focus now is on alpha,” said Blum, whose firm was acquired in July by HFT industry leader Virtu Financial. “Now it’s about getting smarter.”
If “all the gunslingers are equally fast,” says Gerald Hanweck Jr., CEO of derivatives risk analytics provider Hanweck, then the difference maker is “having better and smarter algorithms.”
Research firm Greenwich Associates describes the potential in a recent report: “Algos still work by following simple rules written by the algo developer. . . . The next generation of trading algos will incorporate artificial intelligence, enabling them to learn from the trading logs of millions of historical orders and figure out the best way to execute new orders entered into the system.”
Trading is but one of several possible uses for AI in the institutional investment life cycle identified by Greenwich vice president Richard Johnson. Also mentioned are research reports and idea generation, sales support, and compliance.
Others see AI making an impact in terms of cost reduction, efficiency, and quality assurance on everything from order management to transaction cost analysis to pre- and post-trade risk management. Johnson writes that investment banks can meet cost-cutting goals through robotic process automation, a variant of AI.
Henri Waelbroeck, director of research at execution management systems company Portware, sees a “fight for intelligence” shaping up that could resemble that for speed, but it involves a very different set of tools and techniques, many of them not yet out of beta trials.
“Speed will always be important — it’s just that fewer people are fighting over it,” says veteran capital markets technology analyst Larry Tabb, founder and research chairman of Tabb Group.
Indeed, only a few firms are still going full-tilt in the speed game; their high levels of specialization and capital investment during a period of low market volatility served to discourage less-committed competitors. Recognized leaders include Citadel Securities, Susquehanna International Group, and Two Sigma, which all figure to be formidable in the intelligence race too.
Getco, an electronic market maker founded in Chicago in 1999, has come to represent HFT consolidation: In July 2013 it acquired Knight Capital Group, which had suffered a high-profile operational breakdown the previous August. Virtu bought the successor company, KCG Holdings, on July 20 for $1.4 billion.
The “smarts,” as Tabb and others refer to the promise of AI and other advanced data science, are grounded in the computational capacity to bring massive quantities of information to bear on portfolio and trading decisions. In addition to conventional market data, both historical and real-time, analysts and strategists are in hot pursuit of ways to digest news and sentiment indicators and “machine-translate” them for input into alpha-seeking trading models and algos. Meanwhile, alternative data sets such as cargo movements and consumer purchasing patterns are seen as ways to get a jump on mainstream media and market reports.
Intercontinental Exchange’s ICE Data Services, for example, recently began distributing alerts from AI information- discovery company Dataminr to give energy market participants a leg up on breaking news. Accern, which competes with Dataminr, scans 300 million web, blog, and social media sites to distill information relevant to equity researchers and quant traders.
“Only a computer can operate at that scale,” notes Accern CEO Kumesh Aroomoogan. “A quantitative hedge fund will execute trades, real-time and automatically,” based on the signals.
Smarts include the predictive analytics, derived in part from large transaction databases, that many organizations are hoping to capitalize on. Hazem Dawani, chief product officer of Vela Trading Technologies, says “being fast enough and smarter” thus can trump being “fastest to trade.”
The raw processing power that makes such data aggregation and analytics possible is related to that which fueled the speed race, and it is increasingly economical and accessible. Relying on cloud computing, “we process terabytes of data in real time,” says Accern’s Aroomoogan. “That wouldn’t have been possible ten years ago.”
On the hardware front, beyond the availability and continuous performance gains of CPUs supplied by Intel Corp. and others, the financial services industry now widely deploys specialized field-programmable gate arrays, a type of superpowered computer processor. Australia-based FPGA vendor Metamako last November announced a market-data-filtering device with a record-low round-trip exchange latency of 69 billionths of a second.
In 2009, Hanweck began to adapt graphics processing units, originally developed for video games, for the outsize data-crunching requirements of the options market — to the tune of millions of analytical calculations per second. CEO Hanweck, a former JPMorgan Chase & Co. chief equity derivatives strategist who founded his eponymous real-time-analytics firm in 2003, sees machine intelligence as a route not only to enhancing algo quality, but also to building on progress already evident in such areas as robo-advising and smart beta strategies.
Intelligence is increasingly evident in offerings to the buy side. Portware, a FactSet Research Systems subsidiary since 2015, has been providing AI- assisted trading technology to major asset management firms since 2012, when it acquired research director Waelbroeck’s team from Aritas Group.
Bloomberg, which makes sizable R&D investments in machine learning and natural-language processing, in March introduced — through its Tradebook unit — OPTX, a best-execution optimizer that adapts in real time to changing market conditions.
Thomson Reuters significantly bulked up its trading business and buy-side clientele with the January acquisition of REDI Holdings. “It’s more a data challenge than a speed challenge,” says co-head of trading Michael Chin about the objective of connecting clients with the tools they need to hone their competitive edge.
A machine-learning algorithm is at the heart of LOXM, a JPMorgan Chase execution engine due for a global release in the fourth quarter, the Financial Times reported on August 1.
Might the AI explosion follow the progression of speed trading and other breakthroughs to where the business is concentrated and the technology commoditized? “I could see it whittling down to a few big players,” says Portware’s Waelbroeck.
But not before some possibly revolutionary changes take hold, to the benefit of large numbers. Ben Polidore, head of algorithmic trading at agency brokerage Investment Technology Group, anticipates major analytics-driven improvements across the board — from how orders are teed up before a trade to streamlined work flows throughout. A better-performing buy side will attract more assets and deliver more alpha.
“It may seem like we are in a mature business, but a lot of change is coming,” Polidore says. “We’ll be looking back on the present time as quaint.”
