By Mike O’Hara, Special Correspondent.
The ongoing necessity for many trading professionals to work from home has placed an additional burden on compliance teams struggling to ensure their firms play by the rules while staff operate away from designed-for-purpose trading floors. The situation has been a catalyst for change in how surveillance against financial crime and market abuse is conducted.While monitoring of actual transactions is little changed by the work from home phenomenon, greater emphasis is now being placed on voice and e-comms surveillance, as traders are forced to use nonstandard voice, chat and other channels for their pre-trade client communications, while continuing to be subject to regulators’ stringent data capture and archiving requirements.
With trading staff now using their own home PCs and mobile phones to communicate through channels such as Zoom, Google Hangout, Microsoft Teams and WhatsApp, in addition to their standard Bloomberg chat and other trading platforms and applications, the integration of trading/transaction surveillance and e-comms surveillance has become increasingly important.
Within this environment, there is ample scope for market abuse, unless all such channels are monitored appropriately. In fact, in a recent speech, the FCA stated that from a surveillance perspective, office and working from home arrangements should be equivalent. Firms are expected to implement rigorous oversight within this new environment, particularly regarding the risk of staff using privately-owned devices.
So how can firms ensure they are covering all the bases, and what can they do to close any gaps where activities are not being adequately captured and monitored? This topic will be discussed on A-Team’s forthcoming webinar on January 19, ‘Data management for analytics and market surveillance’. Featured speakers include Ilija Zovko, Senior Researcher at Aspect Capital; Nick Maslavets, Global Head of Surveillance Development, Methodologies, & Analytics, at Citadel; and James Corcoran, CTO Enterprise Solutions, at Kx. Registration is free.
Culture & Conduct
In the wake of the financial crisis and the FX market manipulation and Libor scandals, much emphasis in recent years has been placed on improving culture and conduct within financial markets. But with compliance staff now unable to ‘walk the floor’, lacking physical oversight and the ability to speak with people, hear rumours, ask questions, and generally observe what’s going on, conduct risk is arguably at a higher level now than it has been for some time.
That means that as well as monitoring for market abuse, compliance teams increasingly need a digital surveillance capability to monitor employee behaviours, to detect signals indicating that there might be a conduct issue.
There are various technology solutions that can help firms monitor trader activities in a remote working environment (See box, below). However, people need to be mindful of relying too heavily just on technology, says Helen Bevis, Head of Strategic Partnerships at SteelEye, which provides surveillance and records retention capabilities for financial markets participants.
|Surveillance Technology Vendors to Watch
(Best Trade Surveillance Solution for Dodd-Frank Act, RegTech Insight Awards 2020)
www.Eventussystems.comNICE Actimize Surveil-X
(Best Trade Surveillance Solution for MAD/MAR, RegTech Insight Awards 2020 & 2018)
www.niceactimize.com/complianceShield Financial Compliance
(Best Data Visualisation Provider, RegTech Insight Awards 2019)
Other Notable Surveillance Technology Vendors
“We run the risk of removing the human interaction and relationships that managers should have over their teams, because we’re depending too much on surveillance and monitoring techniques within technology solutions, and not raising flags when we see them from a human aspect,” she says. “The two need to work together, there needs to be a balance”.
Under current regulations, such as MiFID II and MAR (Market Abuse Regulation) in Europe and Dodd-Frank in the US, the requirement to capture, monitor and analyse all electronic communications presents an ongoing challenge to financial markets firms. Remote working – with the increased reliance on chat, voice calls over unsecured phone lines and increasingly video conferencing – has only intensified this burden and has introduced additional challenges, around privacy, for example (see White Paper here).
The key to successful e-comms surveillance however, is not just being able to monitor conversations, but to understand what is actually occurring in them (see White Paper here). Vendors are constantly striving to improve their voice recording, transcription and translation mechanisms, across multiple languages, to help firms meet these requirements. Automated transcription of call recordings is now fairly standard practice in the industry, but the quality and accuracy of those transcriptions is still something of a work in progress. It is, however, improving all the time.
SteelEye, for example, provides voice transcription and translation in 54 languages, with configurable alerts for specific words and phrases used in phone conversations. Their solution also contains a lexicon amassed from dozens of Market Abuse cases and court filings.
Future technology advancements may start to offer additional insight into audio, without relying purely on the transcription itself. Elements such as tone, pitch and volume, determining whether a speaker is aggressive or passive for example, together with all the metadata that comes with the audio, could help to provide context and a much richer understanding of conversations.
Another surveillance challenge that firms constantly face is how to integrate the various different types, sources, and structures of their trade-related data across different asset classes, many of which present their own idiosyncrasies. For exchange-traded instruments, trading is generally conducted electronically on standardised platforms, so trade-related messages can be monitored relatively easily. But for areas such as fixed income and structured products, things become more problematic. Some firms may be trading on 20 or 30 different platforms, all of which will have their own protocols and data formats, with the majority of trading negotiated bilaterally between counterparties, using voice or chat as the main trading mechanism. In order to make sense of that data, it needs to be somehow normalised.
“Preferably you get the unstructured data with some kind of meta information”, says Lars-Ivar Sellberg, Executive Chairman of Scila AB. “And that typically differs quite a lot between different sources, so you need to be good at integrating this data and normalising it. Because that’s the key, it’s not just a matter of understanding the formats, you really need to make sense of the data itself.”
To be most effective, trade surveillance needs to happen in real time as trades and quotes occur. But things become more complex when a firm needs to consolidate real-time data with delayed or T+1 data, which is a common requirement, according to Sellberg. So trade surveillance platforms need to be able to accommodate this, by monitoring activity not just from an individual trader/trading desk perspective, or from an instrument/asset class perspective, but also from a timing perspective.
Holistic & Integrated surveillance
The skill sets within compliance departments – and the technologies they use – still tend to be somewhat siloed at larger firms. Trade surveillance and e-comms surveillance are often run by different teams, for example.
‘Holistic surveillance’ is the Holy Grail of bringing the two together, along with other elements such as archival, search and discovery, and the ability to perform full trade reconstruction across all channels when an investigation is necessary. True holistic surveillance should also be cross-asset, cross-market, and cross-geography, which requires an integrated solution from a technical point of view.
Some firms are now looking at how to achieve this. It requires an understanding of where the relationships lie between communications and trades, along with the ability to create timelines as to how and why activities have taken place.
Such an approach could potentially integrate compliance with HR systems and CRM systems too. With the benefit of having all of that data available, there are use cases where other departments could potentially utilise and visualise certain aspects of case studies and investigations, or even look at performance-related data.
Firms could then start to understand if there are signal that indicate staff depression, or potential drug usage for example, where they might want to add support structures around those teams, expanding the role of surveillance from, “We’re watching you to make sure you’re not breaking rules”, to, “We’re watching you to make sure you’re okay”.
AI & machine learning
Many legacy surveillance platforms use simple rule-based or alert-based systems, which tend to be very reactive and offer limited capabilities for tuning. This means that false positives, which are highly time-consuming and resource-intensive from a compliance point of view, are common (see White paper, here).
More modern surveillance solutions, which use aspects of AI and machine learning, go beyond alert-based results, and promise to reduce false positives so that compliance teams can be more proactive with their time. These new approaches are starting to be used to enhance the rule-based surveillance and to focus more on behavioural aspects.
“With the data that we’ve been able to pull together and normalise, we are moving into a more behavioural form of surveillance”, says Paul Clulow-Phillips, Global Head of Capital Markets Surveillance at Societe Generale CIB. “We’re looking at individual clients, individual traders, comparing them with peers, comparing them with historic activity, and looking for anomalies, rather than necessarily searching for a specific pattern using a rule-based approach.”
Beyond the Natural Language Processing (NLP) and Natural Language Understanding (NLU) common in modern surveillance systems, AI and supervised machine learning is also increasingly being used to help classify an alert as serious or not. An operator can start off with a set of generated alerts, classify them, and based on that initial example set, the machine learning algorithm then further classifies newly created alerts in terms of how serious they are. Over time, the algorithm learns more and becomes more and more precise, filtering out false positives as they are generated, finding anomalies and deviations from normal trading, and using those as a starting point for further investigation.
This type of approach is essential if false positives are to be kept at a minimum, according to Shiran Weitzman, CEO of Shield, a technology company that uses a range of AI and machine-learning models for both trade and e-comms surveillance. “We’ve managed to reduce the noise level of false positives dramatically to a rate of 0.1% from the overall interactions, which is a breakthrough in surveillance monitoring. We simply couldn’t accept the 99.9% false positive standard in the market”.
AI can be used in every part of the process, says Weitzman. “In the enrichment of how you build the data set, in generating a smart lexicon so you no longer you need to write and define terms, in semantic analysis, and in understanding intent and context. And the systems are continuously changing and evolving and learning”.
Cloud & as-a-Service deployments
Looking at the market as a whole, from small groups of point & click traders, investment firms and hedge funds, through to multinational investment banks with thousands of traders, they all have very different needs and policies around surveillance and monitoring. So there is no “one size fits all”.
Historically, the larger the firm, the bigger the system. But with the emergence of cloud and as-a-Service offerings, firms (regardless of their size) now have the capability to deploy modern surveillance solutions in a variety of ways. A platform architected around a microservices framework for example, can potentially be deployed on the public cloud, in a private or hybrid cloud, or on premise.
The as-a-Service model offers firms the flexibility to utilise different areas of modern surveillance platforms for different areas of their business, rather than having to buy a costly enterprise license for the entire firm, resulting in faster time-to-market and a lower TCO.
In conclusion, it’s clear that the surveillance technology space is evolving rapidly, and will continue to advance as AI and machine learning becomes more widely adopted. Simple rule-based systems operating in silos will no longer suffice for trade surveillance, and basic recording and capture of phone calls and chats, without the means of understanding those conversations, will be inadequate for e-comms surveillance.
Surveillance technology vendors are continuously upgrading their platforms with new features and functionality, so firms are somewhat spoilt for choice in terms of solutions. But the human aspect of all of this remains a key element. The systems are only as good as the people who operate them.