Machine learning - The future of financial trade surveillance
===

Ali Curi: Markets ConversatION is an ION podcast where we discuss topics of importance to capital market participants with product owners, subject matter experts, and industry leaders. 

Marco Frangi: Gen AI is already the new big thing. Also, in trade surveillance space, a lot of people, a lot of vendors are talking about using Gen AI within trade surveillance, they are looking for use cases and application of such a technology. I still think that a lot of these people are going the wrong direction because they are trying to find an application for Gen AI, while I think it should be really the other way around. So do we have a use case or a pain point currently for which we think that Gen AI could be a game changer or that can help us in solving the issues?

Ali Curi: Hi everyone, and welcome to Markets ConversatION, I'm Ali Curi. On today's episode, Marco Frangi from ION Markets will dive into the impact of machine learning on electronic financial trade surveillance. He reveals how machine learning is transforming the detection of market abuse and how cutting edge algorithms are slashing false positives, boosting operational efficiency, and providing insights into trading patterns.

Marco will help us understand the balance between ML powered analysis and human expertise and tackle the critical question, "Can we trust the machines to safeguard our markets?"

Let's get started. 

Marco Frangi: Thank you for having me, Ali. 

Ali Curi: Thank you for being here. Marco, before we get to our conversation, let's learn a little bit more about you.

Tell us about your background and what is your current role and responsibilities at ION. 

Marco Frangi: I have a quantitative background. I studied math at university. Then I spent one and a half year at a consulting firm, which actually really shaped my approach to work. And then almost five years ago, I joined LIST, which was then acquired by ION.

And I started working as a product specialist on a trade surveillance solution. Now, I oversee product management for the same surveillance solution. And I interact with clients and prospect[s], I build product roadmaps, I interact with sales as a subject matter expert, and all this kind of stuff. 

Ali Curi: All this kind of stuff sounds really good.

Marco, before we get into the nitty gritty of the details. What about you share with us some background on how machine learning and trade surveillance work together. Explain to us the difference between machine learning and let's say AI. 

Marco Frangi: Yeah, I think that's a very good question to start from. I mean, AI is a much broader term, which usually all technologies trying to, I would say, mimic human interaction refers to. While on the other hand, ML is, let's say, just an example of AI. More specifically, ML we are talking about, I would say, a category of AI tools, which allow a machine to learn from past data and actually provide a specific output related to a specific task. So AI is a broader term, ML is just one of the example of what AI can be. 

Ali Curi: Okay, so now we'll dig a little deeper and we have some idea on their difference. What does the electronic financial trade surveillance entail and why is it a vital function with financial institutions? 

Marco Frangi: Trade surveillance is, first of all, a regulatory requirement. So, we have to be clear about this, we are talking about what the law requires financial institution and investment firms to comply with more specifically in this specific case, we are talking about financial institutions need to monitor its own trading activity on financial markets and the activity of their clients in order to detect what could eventually be manipulative activity. So insider trading, market manipulation, all this kind of behavior. The idea behind this is to try to make the market more resilient, robust and fair for all the market participant[s], even though this needs to be clear, even though it is not a revenue generating activity for the firm, still is critical because you can occur with fines by regulators and also you need to avoid a reputational damage.

Ali Curi: Right. So there's various uses, first of all, because it's, it's required, but also the advantages. Let's talk about some of the key advantages. What are some of the key advantages of machine learning systems over traditional systems in, let's say, detecting market abuse? 

Marco Frangi: Yeah, threat surveillance has always been about rule based algos, trying to identify what I described before as manipulative patterns within the trade activity of a financial institution.

Such algos then eventually trigger what we call "alerts." Which then needs to be analyzed by the surveillance team within the bank. They need to go through such alerts and eventually [point out] any real manipulative behavior. So in the last [four or five] years after the pandemic we saw really a dramatic growth in the number of such alerts.

This was due to many different factors markets volatility the increased participation from retail clients in the financial market. So we have different reasons for this, but this led to a dramatic growth in terms of surveillance, especially for big institutions. Such numbers were not manageable by the teams.

And here is where machine learning actually started to become a game changer because it can learn what is actually anomalous among all the various cases that you can have as an output from your rule based algo. And can eventually identify and highlight only what is actually anomalous. So it can be really, it was actually, really a game changer.

ION Ad: This episode is brought to you by ION. At ION, our clear derivatives solutions, automate your complete trade life cycle and deliver actionable insights whenever and wherever you need them. We offer execution and order management, post trade processing, and a complete front to back business solution. To learn more, visit us at iongroup.com/markets, or email us at markets@iongroup.com. 

Ali Curi: Marco, one of the challenges in trade surveillance is the high volume of false positives. How can machine learning help these irrelevant alerts and improve the accuracy of detection? 

Marco Frangi: Yes, Ali, that was exactly what I was talking about before. What is basically a false positive?

A false positive is an alert which I, as an analyst, I would love not to see at all. I don't want to spend a second looking at it. Okay, so ML, as I was saying, can greatly help in reducing such alerts because we have tools, for example, which can learn what is actually an anomalous, irrelevant behavior and then leveraging this knowledge can tag newly generated alert[s] as relevant and not relevant, or even can let the system trigger only relevant alert. 

Ali Curi: Now, given that surveillance teams often operate with limited budgets, how can machine learning contribute to reducing costs, for example? 

Marco Frangi: Yeah, as we just discussed, ML helps reducing not relevant alerts, which means that you can drastically reduce the time spent in the analysis.

As you can imagine, since we are talking about manipulative actions within financial markets, usually surveillance analysts need to be skilled senior resources, which means usually big salaries. Therefore, if you have less work to do, you can do it with fewer resources and less time, which easily leads to significant cost savings, together also with making the team focus it on maybe tasks which could be with more added value for the institutions.

Ali Curi: Now, one of the things that I came across in looking into the topic was something called "explainable AI" or "explainability." What is it? And how does explainability in trade surveillance help machine learning models become more transparent? 

Marco Frangi: We all know that ML and AI tools were often perceived as a "black box," what people call "black box."

Why? Actually, it is 100 percent understandable because usually you don't actually know the criteria they use to produce the output they give you. Okay, so it's understandable why people usually perceive them as black box. Nevertheless, there are some techniques which make such tools more transparent. And this is what explainable AI is about, it is about such techniques. The example that I can [give] you is, if we focus on trade surveillance, the example I was making before, so an ML tool which is able to suggest you if an alert, newly generated, is according to its knowledge, relevant or not relevant. So it's giving you a suggestion you can ask yourself, "Which criteria did this tool leverage to tell me this, to suggest to me that the alert is not relevant?" So we have, for example, in this case, a theory which is called "Shop Value Theory." I won't go into details, but in a nutshell, it simply breaks down the suggestion on all the information of the alert.

So, for example, the trades involved, the client involved. So, all the information which are understandable from the analyst. This theory tells me, okay, I used this and this and this information to provide you with this suggestion. So it's understandable now for the analyst. Which were the feature[s], the characteristic[s] of the alert, the tool used to tell me that I would likely close it as not relevant, or I will likely analyze and consider the alert as actually relevant.

So I have this breakdown on things that I know, which makes AI suggestion, ML suggestion more understandable to me. 

Ali Curi: Great, Marco. And looking ahead, what do you see as the future of machine learning and trade surveillance? Are there any emerging technologies or trends that you think will help shape this field going forward?

Marco Frangi: I mean, nowadays, Gen AI is already the new big thing. Also, in trade surveillance space, a lot of people, a lot of vendors are talking about using Gen AI within trade surveillance. They are looking for use cases and application of such technology. To be honest, actually, I still think that a lot of these people are going the wrong direction because they are trying to find an application for Gen AI, while I think it should be really the other way around.

So do we have a use case or a pain point currently, for which we think that Gen AI could be a game changer or that can help us in solving the issue? So I think that a lot of people are going the wrong direction because I mean, as the best of my knowledge, I still think that nowadays ML is still the best technology to address most of the issues and the pain points in trade surveillance.

Ali Curi: I think there's going to be some exciting things on the frontier pretty soon. Marco, what's the one big thing that you hope listeners would take away from the episode? 

Marco Frangi: I hope that they have now, first of all, maybe a little bit more understanding of what trade surveillance is, how critical it is for financial markets, and also how much leveraging proper technology can make your life easier also, it's also in this area. 

Ali Curi: Great. Well, now we're going to take a little quick sidebar. Let's talk about career advice. What is some advice you wish you had heard earlier in your career? 

Marco Frangi: Maybe I would say not to be afraid of being outside your comfort zone and not to be afraid of, I mean, failing, making mistakes.

I eventually realized after some years that actually this is the easiest way, to be honest, to grow, to learn things, to improve yourself as a professional. 

Ali Curi: I think that's great advice. We learn from our mistakes. Marco Frangi, thank you for joining us today. I hope you visit us again. 

Marco Frangi: Thank you very much, Ali. It has been a pleasure. 

Ali Curi: And that's our episode for today. You can follow ION Markets on Twitter and LinkedIn. Thank you for joining us. 

I'm Ali Curi. Until next time.