Up until recently, machine intelligence could only be used by large sell-side institutions and sophisticated quantitative trading groups to improve how they operated in the marketplace. But as technology advances continue to push the depth and breadth of machine learning capabilities, it is allowing financial firms of all shapes and sizes to improve their compliance and trading capabilities.
Such is the importance of machine intelligence that it has the capacity to help clients improve their business models by finding new meanings in data, from the front to the back office, and then convert those insights into performance returns.
Machine-driven insights
Mike O’Rourke (pictured) is Vice President, Global Head of Machine Intelligence & Data Services and leads all machine learning initiatives at Nasdaq. There are, he says, three key areas Nasdaq is looking to leverage this technology: “To improve interactions with our users by offering more context-relevant and user friendly applications; to provide insights that they otherwise wouldn’t have; and to automate cognitive processes to create more consistent products.”
Within Nasdaq’s data group, there are two critical questions it is looking to answer: How can it this technology help clients trade more efficiently? And how can it help them improve their investment and trading strategies?
To answer the first question, Nasdaq’s Innovation Lab, which leverages cutting edge, scalable technology to provide Nasdaq’s clients with comprehensive multi-asset class market data solutions, debuted Nasdaq Trading Insights last November. This technology can be used to help clients identify missed trade opportunities due to latency or liquidity issues as well as improve fill rates.
As O’Rourke confirms: “We’ve had heads of dealing desks coming to us saying, ‘With SMARTS, you’re able to identify where we might be spoofing the market. Surely you can take that same technology and tell us when one of our algorithms is not working well, or has been poorly designed? Can you tell us when we are being impacted so that we can change our trading strategy?”
To answer the second question, it created the Nasdaq Analytics Hub.
“On Nasdaq Analytics Hub, we take in third party data sets and intermingle them with our own market data,” explains O’Rourke. “Then we implement proprietary machine learning processes to cleanse the data, apply transformations, and ultimately highlight signals within the data – providing outperformance indicators for clients’ trading strategies. It’s about creating market context, and consistency in delivery of that context.”
It is a way to generate insights from data that might not previously have been possible to the client. This has the potential for Nasdaq’s clients to gain an edge in the marketplace and develop investment ideas that their peers are not capable of generating.
The Analytics Hub uses four inaugural data sets: Nasdaq Dorsey Wright technical analysis data, social sentiment data, macroeconomic indicators such as central bank communications, and retail investor sentiment data.
Such is the enormity of data flow being used today for trading and analytics purposes that human beings will increasingly come to rely on machine learning technology. Without it, they simply will not be equipped to uncover underlying signals or anomalies in myriad data sets, which they can then implement into their trading strategies to compete more effectively in the marketplace.
Tuning out the noise
Whereas Nasdaq Trading Insights and Nasdaq Analytics Hub use machine intelligence as a sword with which to penetrate the markets to generate insights into capital – and are therefore more front office-focused – Nasdaq’s technology advances in AI apply equally to the middle and back office where they act more like a shield.
In this instance, the technology is utilized by Nasdaq’s Risk & Surveillance business division, specifically with its SMARTS surveillance solutions.
SMARTS Trade Surveillance, in brief, automates the detection, investigation and analysis of potentially abusive or disorderly trading, to help improve the overall efficiency of the surveillance organization and reduce cost.
“Today, we’re doing all of that for our clients. And it’s preventing compliance risk as well as making sure money isn’t left on the table,” says O’Rourke.
Market regulation encroaches on all aspects of business, requiring exchanges, regulators and financial institutions – both on the buy- and sell-side – to closely monitor and identify market abuse activity, stay in compliance with regulatory rules that govern their trading conduct, and help ensure the overall transparency of the global marketplace.
With traditional trade surveillance, all alerts are created equal. Nasdaq’s machine learning technology analyses the data points to determine the probability that an alert will turn into something more serious. A combination of machine intelligence and natural language processing enables firms to rank and score alerts, helping tune out a lot of the noise in the market, and allowing clients to spend more time focusing on the highest priority alerts.
“By generating those alerts within SMARTS, we can use machine learning to determine what factors were involved in the run up to an alert being raised, rather than use a strict set of logic-based rules,” explains O’Rourke.
This is where O’Rourke’s earlier point on providing context and consistency comes into play for the middle and back office as well.
From a context perspective, as regulators have brought more asset classes into scope the message flow has increased. Nasdaq’s customers are struggling under the weight of flags and alerts that are being generated; especially as OTC markets move towards centralized clearing.
“Up until now, logic-based alerts have been the cornerstone of how one surveilled the markets,” says Michael O’Brien, Head of Product Development, Risk & Trade Solutions at Nasdaq. “The customer wants to err on the side of caution and receive an alert so as not to miss anything, but then the question becomes, ‘How can we point them in the right direction when they’ve got multiple alerts being raised in the market? How can we bring context to the task and help them evaluate these alerts and flags?’”
SMARTS has already run a successfully for the Market Surveillance function at the Nasdaq Nordic exchange. It is, says O’Brien, taking a process where intuition is applied – is this alert interesting or not? – and seeing whether or not the process by which it is escalated or closed by a team of analysts can be predicted, so as to enable the dynamic scoring of alerts.
“We looked at how the client was getting alerts and escalating things. Using machine intelligence technology we were able to predict, to quite a high degree of accuracy, when the client was likely to close off an alert and take no further action, or escalate an alert.”
Going forward, the aim is that this technology will be able to pre-score any of Nasdaq’s clients’ alerts on the basis of how they had historically handled and escalated those alerts. Unlike logic-based rules, which are fixed and linear, machine intelligence has the ability to adapt to different data sets, bringing entirely new multi-levels of institutionalization and automation to clients’ business models.
Detect and discover
“We are framing our alerts as both a detection mechanism and a discovery mechanism,” continues O’Brien, “but this will involve an internal culture and process change for our buy-side and sell-side clients. The way they’ve done trade surveillance over the last 20 years has been very process-driven: looking at and processing every single alert.” The discovery approach focuses less on the risk trading scenario and more on the person executing trades.
The discovery mechanism takes all the trade order aggregations within SMARTS and looks for anomalies, or clusters of unusual data. Whilst this is not generating an alert, per se, what it does do is bring to the users attention something that is beginning to look unusual and may indicate a change in the risk profile of a particular trader.
“Over the long term, we may be able to map different clusters and anomalies to particular behaviours; for example, it could point towards potential insider trading activity, or that a trader or account acts in a manner which is associated with the behaviours that have been associated with an insider, before it becomes an alert,” says O’Brien.
This ability to use machine intelligence to map trading profiles and behaviours to identify potential market abuse, shows just how powerful technology is becoming.
Used correctly, and judiciously, it could revolutionise the compliance programmes within financial organisations and make them even more transparent in the eyes of regulators and investors. For emerging managers especially, the operational alpha that they can achieve through smarter best practices could directly impact their fund raising capabilities.
“As we bring on more and more data streams it will become more important to leverage this emerging technology. The ability to map data to trader-based behaviour, both in staying compliant and achieving alpha, is central to our ongoing machine intelligence growth strategy,” concludes O’Rourke.
