The new surveillance system has a been long time in the making and the team first started developing and testing it in 2016. “We obviously had to proceed with diligence, given that exchange operations in general and surveillance in particular is a highly regulated and sensitive area, so one challenge was that we had to allow the project to take time,” Joakim Strid says. He adds that the production phase has also further pushed the Nasdaq Nordic’s team of surveillance analysts to become more diligent in how events are categorised and actions documented, “in order for the [machine] learning process to give the best outcome”.
The new machine learning system was built to complement the existing surveillance system to analyse abnormal market events and their subsequent categorisation by surveillance analysts across the Nordic markets. “This project marks an important milestone in the use of machine learning in the capital markets,” said Tony Sio, head of exchange and regulator surveillance, market technology, Nasdaq. “By closely collaborating with the Nasdaq Nordic surveillance team, we’ve been able to build unique algorithms that have improved the efficiency and effectiveness of monitoring our own markets. At the same time we’re progressing on a broader machine learning strategy and exploring other applications of this technology to strengthen the surveillance process for markets worldwide.”
The new system is also the exchange’s answer to Mifid II and the Market Abuse Regulation (MAR), which came into force at the beginning of this year and summer 2016 respectively and have put increased demands on market operators and market participants. Mifid II overhauled nearly every aspect of trading, attempting to inject more transparency and lessen conflicts of interest, while the market-abuse rules defined attempted market manipulation as an offense, greatly expanding the number and types of potential events that surveillance systems need to capture. At the same time, markets are becoming more complex and attempts of misconduct on the market more sophisticated. Joakim Strid says the new surveillance system enables the team to work more efficiently and save time. “Trade surveillance is very much like searching for needles in haystacks. If we can remove some of the haystack, our analysts will have a better chance at finding needles and they will make better use of their time. Our analysts’ time is our most precious asset,” he explains.
Nasdaq Nordic said that the machine learning capabilities will initially be used to prioritise the surveillance workflow. “The technology predicts the likelihood that the event will lead to an action by an analyst. This will particularly help in situations where workload is high, such as during the opening and closing of the markets,” the exchange said in a statement. “The new prioritisation ranking is then used to complement traditional quality controls in relation to alert handling, which then enables surveillance officers to identify outliers where the actual handling of alerts has differed from the prediction of the algorithm. Lastly, the existing alert designs will be evaluated based upon new relationships or rules revealed by the machine learning technology and redesigned and improved accordingly,” it continued.
Joakim Strid says he believes that eventually machine learning will become a built-in functionality in SMARTS and in other systems. “It will become natural features that we don’t think too much about. The interesting part is when the same type of approach can be taken on other areas and when we can use it for detection and pattern recognition. That will add value on another dimension,” he says.
Nasdaq Nordic has launched a new machine learning surveillance system for the stock exchange that uses algorithms to help analysts categorise alerts of suspicious activity from Nasdaq’s different trading platforms and to provide managerial oversight into analyst behaviour and decision making. The new system that has been used in production for around four months is the brainchild of Nasdaq SMARTS and the Nasdaq Nordic Market Surveillance team. The exchange hopes it can be put to greater use in the future to weed out market abuse. Joakim Strid, head of European surveillance at Nasdaq Nordic, tells FBNW that the increasingly large sets of data available forms an excellent base for machine learning. He explains that one of the benefits of the new system is that it has enhanced Nasdaq Nordic’s method of quality assurance and manager oversight. “In trade surveillance, analysts take regulatory decisions every minute of the day. Needless to say, we trust the judgment of our analysts but we still need to perform oversight to ensure consistency and accuracy. The machine learning tool creates reports of outliers, where the action taken by an analyst has differed from the predicted one, which is a great complement to random or risk-based reviews,” he explains.
The new surveillance system has a been long time in the making and the team first started developing and testing it in 2016. “We obviously had to proceed with diligence, given that exchange operations in general and surveillance in particular is a highly regulated and sensitive area, so one challenge was that we had to allow the project to take time,” Joakim Strid says. He adds that the production phase has also further pushed the Nasdaq Nordic’s team of surveillance analysts to become more diligent in how events are categorised and actions documented, “in order for the [machine] learning processIf you’re new to Tell Media Group, create an account.
Read more about our memberships
