ION’s Marco Frangi discusses machine learning in financial trade surveillance

Key Takeaways

• ML beats rules-based algos in identifying genuinely suspicious activities
• Generative AI users should identify pain points to address
• Explainable AI is a crucial aspect of AI-driven trade surveillance

In a recent episode of the Markets ConversatION podcast, Marco Frangi from ION Markets discussed the transformative impact of machine learning on electronic financial trade surveillance. The conversation delved into how cutting-edge algorithms are revolutionizing the detection of market abuse, reducing false positives, and enhancing operational efficiency.

“Gen AI is already the new big thing,” Frangi stated, highlighting the growing interest in applying generative AI within trade surveillance. However, many are approaching it from the wrong angle. “They are trying to find an application for Gen AI, while I think it should be really the other way around.” Instead, Frangi emphasized the importance of identifying existing pain points that generative AI could address effectively.

Frangi, who has a quantitative background and oversees product management for a trade surveillance solution at ION, shared insights into the critical role of trade surveillance in financial institutions. “Trade surveillance is, first of all, a regulatory requirement,” he noted. Financial institutions must monitor trading activities to detect manipulative behaviors such as insider trading and market manipulation. This regulatory necessity aims to make markets more resilient, robust, and fair for all participants.

The discussion also touched on the advantages of machine learning systems over traditional rule-based algorithms in detecting market abuse. Frangi pointed out that the pandemic led to a dramatic increase in the number of alerts generated by traditional systems, making them unmanageable for surveillance teams. “Machine learning actually started to become a game changer because it can learn what is actually anomalous among all the various cases,” he said, highlighting its ability to identify and highlight only genuinely suspicious activities.

Machine learning reduces false positives

One of the significant challenges in trade surveillance is the high volume of false positives.

Frangi’s colleague Mirko Marcadella, chief product and marketing officer at ION-owned LIST, wrote recently that compliance teams are facing more alerts and false positives due to stricter regulations and market volatility. Compounding the issue, there’s a struggle to find skilled professionals who can handle these challenges.

In market abuse detection systems, algorithms’ parameters defining an unusual trading activity are set within certain ranges. Surpassing these limits prompts the system to raise an alarm. This alarm is then assessed by a person in charge of reporting verified market abuse instances to regulators. Firms must calibrate the parameter thresholds to detect real abuses while minimizing false alarms. Typically, the approach is conservative, leading to the generation of an excessive number of alerts that, upon review, are found to be harmless. This burdens compliance officers with a largely unnecessary workload and undermines confidence in monitoring tools.

Frangi said that machine learning helps reduce these irrelevant alerts, thereby improving the accuracy of detection. “A false positive is an alert which I, as an analyst, would love not to see at all.” By tagging newly generated alerts as relevant or not, machine learning significantly reduces the time analysts spend on non-issues, leading to cost savings and allowing teams to focus on more valuable tasks.

Making AI more transparent

The concept of “explainable AI” or “explainability” was another key topic. Frangi acknowledged that AI tools are often perceived as a “black box” because users don’t understand the criteria used to produce outputs. However, as the research team at LIST said in a blog published earlier this year, techniques like SHAP values can make these tools more transparent by breaking down the suggestions based on understandable information. “It makes AI suggestions more understandable to me,” Frangi said, emphasizing the importance of transparency in AI-driven trade surveillance.

Looking ahead, Frangi sees generative AI as a significant trend but maintains that machine learning remains the best technology to address most issues in trade surveillance. For market participants, “leveraging proper technology can make their life easier.”