Flash Crash Trader Navinder Singh Sarao At Westminster Magistrates' Court...Navinder Singh Sarao, a British trader charged over his role in the 2010 U.S. flash crash, leaves Westminster Magistrates' Court after losing a bid to delay extradition proceedings in London, U.K., on Friday, Aug. 28, 2015. Sarao asked a London judge for more time to prepare an expert report on trading, but the judge rejected the request, saying the issue was irrelevant to the question of extradition. Photographer: Chris Ratcliffe/Bloomberg
Trader Navinder Singh Sarao, who is resisting market manipulation charges, at Westminster Magistrates' Court © Bloomberg

In Robert Harris’s 2011 novel The Fear Index a secretive hedge fund builds a computer capable of making its own trading decisions.

Gobbling up information, the machine starts to confuse its human creators by building huge stakes and making a handsome profit from a market panic. As they assess the outcome, one of the protagonists notes: “The beauty of it is that it was but 0.4 per cent of total market volatility. No one will ever notice, except us.”

As markets increasingly rely on computer algorithms, reality is imitating fiction: artificial intelligence is becoming a bigger part of investing and it is also helping regulators ensure that traders do not get away with bad behaviour.

“Algorithms facilitate the ability to get into and out of the market very quickly. That’s the more recent challenge. You can have efficient strategies that 10 or 20 years ago you’d have needed more people to execute,” says Michael O’Brien, head of product development at Smarts Trade Surveillance, a unit of Nasdaq, the US exchanges operator.

Regulators and policymakers have already responded to the market’s inexorable move over the past decade to electronic trading of assets like futures and equities.

The US Dodd-Frank act of 2010 tightened rules against tactics such as spoofing, where traders put orders into the market without any intention of executing them. The trader creates a false interest in buying or selling and then profits when others are misled into buying or selling away from the real market price. These strategies are not new, but as trading moved to computers, the tactics have become faster to carry out and more anonymous.


Estimated annual spend on trading surveillance technology

Those tighter rules have led to several high-profile legal cases brought by the US authorities, including against 3Red, a Chicago high speed trading company, and individual traders Michael Coscia and Navinder Singh Sarao.

The US Commodity Futures Trading Commission’s case against 3Red is ongoing, while Coscia was convicted of manipulating futures market prices last November. A UK judge ruled in March that Sarao can be extradited to the US to face charges.

“Market manipulation cases were as rare as hen’s teeth. Surveillance technology has improved, regulators can pull apart trades line by line or tick by tick,” says Mr O’Brien.

Indeed, PwC, the professional services company, estimates that £156m will be spent this year alone on surveillance technology. Even so, regulators are still aware that it can take years to bring cases to court. They hope to use technology to speed up the process. Last month Finra, which oversees US dealer-brokers, began to issue “report cards” to brokers over how much potential spoofing or possible market manipulation was occurring through their systems.

Yet for all their power, computers still find it difficult to tell the difference between a cascade of manipulative buy and sell orders, and a trader who simply changes their mind a lot.

Traditional approaches to surveillance based on predetermined thresholds often trigger hundreds of alerts a day, meaning small surveillance teams still face a mountain of suspicious trades to work through.

Now some want to bring artificial intelligence to trade surveillance, marrying traditional methods of detection with new techniques emerging from Silicon Valley to mine vast quantities of data.

Neurensic, a US software company founded last year, has developed technology that gives traders more clarity on whether their trades come close to the legal definitions of spoofing and disruptive trading. Powerful algorithms learn pattern recognition and natural language in ways that mimic the human brain.

“Our objective is not just catching criminals, you need a fair standard to provide clarity,” says David Widerhorn, chief executive of Neurensic. “This will be seen as good or bad behaviour. It’s smart policing of big data. It can now find a needle in a haystack in a couple of hours, not a couple of weeks.”

Others say part of the spending push identified by PwC is due to laggards trying to catch up. “In US equities, people are moving on to a new generation. In futures, it’s more of a late arrival,” says Michael Friedman, general counsel and chief compliance officer at Trillium Trading, a New York trading and technology company.

Already some are anticipating the next stage. Earlier this year Nasdaq joined Goldman Sachs and Credit Suisse Asset Management with an investment in Digital Reasoning, a Silicon Valley start-up aiming to to use “cognitive computing” techniques to spot abuses in capital markets.

That will aim to combine monitoring of trading patterns and electronic messages together in “holistic surveillance”, says Mr O’Brien. Monitoring of electronic messages, like trading patterns, is too imprecise and generates too many “false positives,” he adds. “Electronic trading platforms lend themselves to carrying out sophisticated spoofing but that also lays out a really great order trail.”

Nevertheless for all the possibilities afforded by technology, others point out that they are likely to be hemmed in by legal constraints.

Lawyers could argue certain trading activity is illegal, says Mr Friedman. “Everyone agrees you need some automated approach [to surveillance] but it’s based on enforcement precedents.”

Get alerts on US financial regulation when a new story is published

Copyright The Financial Times Limited 2022. All rights reserved.
Reuse this content (opens in new window) CommentsJump to comments section

Follow the topics in this article