The recent explosion in AI has heightened concerns about how the technology might lead to new unforeseen problems. Aside from the existential risks that seem to drive the most hype, there are the more practical and immediate concerns around bias, trust, and using AI to justify bad judgment. (We recently covered one take on considerations for operational risk surveillance for AI.)
The evolution of surveillance technology to reduce market abuse violations in the financial industry provides additional insight into these problems. There are many forms of market abuse, such as insider trading, collusion, and misconduct. Financial regulators have increasingly been levying eye-watering fines in some of the most egregious examples. This has included $2 billion in 2012, $3.1 billion in 2014, and $5.7 billion in 2015. This has spurred innovation in new tools and processes to help financial institutions to detect and act on various kinds of market abuse more effectively.
Deloitte Risk & Financial Advisory Senior Manager Niv Bodor says several key events have helped shape the regulatory environment for financial market abuse over the last fifteen years:
The financial market crisis of 2008 and various bank rate scandals, including the LIBOR and Forex probes of the early to middle 2010s, ultimately led to the emergence of stringent regulations globally, with the common goals of strengthening financial market oversight and enhancing standards of conduct.
In the US, this led to new anti-manipulation and anti-fraud rules via the passage of the US Dodd-Frank Act in 2010 and expanded the scope of the EU's Market Abuse Regulation (MAR). It has also encouraged private sector efforts to improve conduct standards through the globally led Financial Markets Standards Board. This has motivated financial institutions to enhance their supervision and surveillance processes significantly.
Bodor says there are some essential differences in surveillance regulations in different regions. The US approach is more principles-based, while the EU is more prescriptive. Additionally, some surveillance regulations include added requirements and emphasis on cross-market and/or cross-product manipulation, whereas others have yet to evolve rule-making to govern more complex forms of market misconduct.
Enhancing surveillance efforts
EY Americas Financial Services Risk Technology Leader Robert Mara says the surveillance tools have evolved to keep pace with the growing complexity of market misconduct. Earlier approaches would monitor for when traders exceeded certain thresholds or used certain combinations of words included in a curated lexicon. Newer tools perform market analytics to detect less blatant problems.
Internal surveillance teams must also manage high alert volumes due to false positives and perform periodic effectiveness reviews. The latest surveillance tools are better at calibrating parameters and tuning the lexicon used to detect potential problems. In addition, these tools increasingly use AI and machine learning capabilities to learn from changing risk and behavior patterns.
Mara says that recent regulatory scrutiny on the use of unapproved communication channels has driven the adoption of tools for data capture, retention, and monitoring capabilities across multiple communication channels. Newer tools offer an integrated cloud-based solution with these capabilities across over 150 channels, which will continue to grow as newer methods of communication continue to spawn.
More holistic surveillance
Deloitte Risk & Financial Advisory Managing Director Roy Ben-Hur is also seeing a movement towards more holistic market surveillance. For example, most firms have evolved beyond static, off-the-shelf and manual solutions to adopt more automated and specialized surveillance systems that leverage analytics to monitor diverse data types and identify patterns in both structured and unstructured data.
Additionally, firms are increasingly integrating their trade monitoring with communications surveillance programs. Communication monitoring is growing in size, budget, and importance. The combination of trade and communication surveillance approaches is helping firms to establish more holistic surveillance approaches that combine data and indications from both. Bodor also expects to see more firms replace rule and lexicon-based surveillance methods with AI-based models:
Many firms are keen to capture the compliance and risk management benefits of integrating AI into surveillance strategies, including the ability to detect more sophisticated fraud and manipulation schemes, such as those committed cross-product or cross-market; improve detection and false positive rates, manage alert volume, and improve overall program efficiency; and optimize analyst review and disposition.
Not a panacea
Even though newer AI models can improve surveillance efforts, they also introduce new problems. Mara observes that firms implementing new AI and machine learning capabilities will continue to work with vendors to meet new regulatory requirements around the explainability of these models. For example, they can introduce bias based on historical data that lead to ineffective scenario and lexicon surveillance and the associated reviews.
AI, as any other technology, is not a panacea solution for market surveillance needs. On the one hand, AI models are quite capable of detecting fraud and manipulation schemes actively being committed within markets. However, it is far more difficult to use those same deep learning models for fraud prevention despite AI's ability to identify patterns and make more precise predictions about future market behavior. An important puzzle for regulators and the private sector alike will be around how to use AI to effectively and ethically minimize the risk of fraud in markets, and not just as a tool for identifying misconduct after the fact.
Ben-Hur is also concerned that the growing reliance on AI for market surveillance could raise new risks, including ones that may not be readily visible or play out in unexpected ways. Organizations will need to understand how AI models, including complex deep learning models, are created and adapted over time to manage these risks. Ben-Hur predicts:
As AI transitions to support more decision-making in the surveillance space, we are expecting regulatory expectations to increase, requiring firms to provide greater transparency and clarity on items such as how model data is being sourced and used, how potential biases are identified and managed, the logic that describes algorithms, and comparisons between established policies, and detection of potential tampering.
The financial industry is already highly regulated and benefits from the ability to track trades and monitor communication channels. It is also highly motivated to avoid the significant fines being levied against different forms of abuse and bad conduct.
Currently, regulators, AI vendors, and the enterprises that use these new tools are still trying to sort out what to track and manage. The AI industry also has to contend with various expectations of privacy and confidentiality as individuals and enterprises use these tools and services.
The broader AI industry may take a while to catch up with the recent progress in financial market surveillance. However, large fines and lawsuits relating to bias, abusive marketing practices, and copyright issues might drive the adoption of better surveillance tools and processes in the AI industry as they have in finance.