The shift towards 24/5 equities trading and the rapid growth of prediction markets are presenting new challenges for trade surveillance. With trading activity increasingly spanning multiple venues, instruments and time zones, identifying potential market abuse is becoming more complex, particularly where suspicious behaviour crosses traditional market boundaries.
For surveillance technology providers, however, these developments don’t necessarily demand a wholesale overhaul of existing systems. According to Robert Cope, Director of Markets and Operations Strategy at Trillium Surveyor, much of the underlying surveillance infrastructure can accommodate these changes, provided detection models are calibrated to reflect different market conditions and trading behaviours.The challenge is becoming particularly apparent as US equity markets extend their operating hours. Nasdaq is pursuing plans for a 23-hour trading day, while changes to clearing and market data infrastructure are helping to support the broader transition towards near-continuous trading.
Adapting Surveillance to Overnight Liquidity
One of the most immediate challenges concerns the different liquidity characteristics of overnight trading sessions. During regular market hours, relatively deep liquidity and high participation can make attempts to influence prices more difficult. Overnight, thinner order books, wider spreads and fewer participants can mean that much smaller trades have a disproportionate market impact.
Surveillance systems must account for these differences without generating excessive false positives or overlooking genuinely suspicious behaviour, explains Cope, in conversation with TradingTech Insight.
“Every market is different, even at different times of day,” he says. “When the main market is open, you have much more liquidity and many more participants. Overnight, the volume and number of participants aren’t there, so parameters have to be adapted to that illiquidity to limit false positives. During the regular session, it can take a lot to manipulate a market. In the overnight session, we’ve seen in our system that someone can manipulate GE with 50 shares. It isn’t reinventing the wheel; the types of manipulation are still the same, you just have to adjust for the different market conditions.”
The operational implications are equally important. Trillium Surveyor’s surveillance platform operates primarily on a T+1 basis, meaning extended trading hours don’t necessarily require compliance analysts to monitor activity continuously. That could change, however, if regulatory expectations move towards more immediate detection and intervention.Prediction Markets Expand the Surveillance Perimeter
The rapid growth of prediction markets introduces another dimension to the surveillance challenge, particularly where event contracts are linked to underlying financial instruments or commodities.
An oil-related contract traded on Kalshi, for example, may have an economic relationship with oil futures traded on CME or ICE. Suspicious activity could potentially span these markets, requiring surveillance systems to identify relationships between instruments traded on different venues, with different market structures and participant populations.
The regulatory implications are already attracting attention. The US Commodity Futures Trading Commission has highlighted risks involving manipulation and the misuse of non-public information in prediction markets, reinforcing the importance of effective monitoring as these products become more widely traded.
For Cope, the underlying techniques needed to identify cross-market behaviour are well established. The difficulty lies in applying them across a substantially larger universe of potentially related instruments.
“For many years we’ve had cross-product and cross-asset alerts between, for example, options and the underlying equity, or commodities. You apply that same infrastructure to prediction markets. The difference is that you have many more opportunities for potential manipulation,” he explains. “We look for behaviours and patterns: if we see a pattern in one market, is there something unusual happening in another market? It’s really about connecting those patterns.”
This also highlights a limitation facing surveillance providers. While regulators may have access to trading records across multiple firms and venues, individual market participants and their technology providers typically have a more restricted view. Identifying suspicious relationships therefore depends on correlating observable trading patterns, with compliance teams responsible for investigating the resulting alerts.
AI and the Explainability Requirement
As the number of instruments and potential cross-market relationships increases, AI offers opportunities to improve the efficiency of surveillance investigations. Machine learning and behavioural analytics are already used in surveillance, while generative and agentic AI could support activities such as investigating alerts, correlating evidence and preparing cases for compliance officers.
Cope nevertheless sees clear limits to the autonomy that can be introduced into these processes. “When it comes to trade surveillance, the big elephant in the room is explainability. The outcome needs to be explained, which is why rules-based approaches will continue to have an important role and agentic AI has to be railed or gated to some extent. Regulators aren’t going to accept, ‘My model told me’, without any explainability. And with any AI model, the quality of the data is critical. If the quality of the data is poor, the outcome from your AI agents will be poor.”
Cope notes that Trillium maintains historical market data extending back to 2013 for some markets, providing a foundation for developing and refining its detection models.
More Data Doesn’t Necessarily Mean Better Surveillance
The growing complexity of surveillance also raises questions about the depth of market data required. While Level 1 data provides much of the information used by Trillium’s surveillance platform, Level 2 order-book data offers additional context that can help distinguish suspicious activity from legitimate trading behaviour and reduce false positives.
Cope argues that Level 3 data, despite offering greater granularity, currently provides insufficient additional value for T+1 surveillance to justify the associated data-processing costs.
That distinction becomes increasingly relevant as firms consider the infrastructure required to monitor more instruments across longer trading sessions. Greater data volumes can introduce significant processing and storage overhead without necessarily improving detection outcomes proportionately.
The priority, in Cope’s view, is obtaining sufficiently detailed, high-quality information to identify relevant behaviours and relationships across markets.
Towards One Continuous Market
Extended-hours equities trading, prediction markets and tokenisation are developing along different paths, but their combined effect is to make financial markets increasingly interconnected.
For surveillance teams, this means looking beyond individual venues, asset classes and trading sessions to understand how activity in one market might relate to behaviour elsewhere.
“From an operational standpoint, the financial world is becoming one market,” says Cope. “When we have 24/5, 23/5 and 24/7 trading, along with everything from tokenisation onwards, it’s making the world into one big market. I think the US is leading that because we have the deepest, most well-structured market and everyone wants to participate in it. But ultimately I see it becoming one continuous market.”
Subscribe to our newsletter
