By Prashant Gandhi1
1Principal Financial Services, ThoughtWorks
Regulators want to ensure that they are both protecting the sanctity of the financial system as well as protecting customers. Given the recent history of supervisory failures, financial crisis and systemic fraud at large scale, regulators are increasing their scrutiny and are creating more rule-based regulations to prevent further market crashes.
Heavy enforcement actions are also forcing banks to rethink their approach to compliance. Banks have often responded to regulatory obligations with tactical responses, especially after the 2008 crisis. Tactical implementations, coupled with manual processes, create complex layers that become hard to untangle. This creates a huge demand on the banks’ resources, both in terms of manpower and capital available to invest. A common trend has been to add large number of compliance professionals to the ranks every year to meet the increasing demand. In 2015, for example, JP Morgan added 13,000 resources to support regulatory and compliance efforts and spent more than $600 million in regulatory and compliance technology.
This is not a race that banks will win. Chris Skinner, a noted author and futurist, often quotes a statistic that a global bank must deal with 185 regulatory changes per day.1 It is inevitable that banks will need to automate compliance checks and regulatory changes.
This is not a race that regulators will win, either. Andrew Haldane, the chief economist at the Bank of England, observed that the type of complex regulation developed over recent decades might not just be costly and cumbersome but sub-optimal for preventing any financial crisis.2 The answer for the regulators may lie in defining their policies using complexity theory through network analysis and behavioural modelling.
One key idea from complexity theory is that regulators should propose simple rules that market participants can follow. In uncertain environments, human behaviour tends to follow simple heuristics better than complicated rules. One way to define these simple heuristics is through the lens of customer needs. Consumers are typically at the wrong end of information asymmetry when dealing with banks. So, a heuristic might be about increased transparency, which can be about visibility, costs or better market comparison. Another heuristic is around control, which can be delivered by providing the power of execution and choices in the hands of the end customer.
The stated goal of the electronification project at JP Morgan is to deliver greater choice, transparency, liquidity and efficiency for the customers and dealers.3 It uses machine learning techniques to learn from past market behaviour, spotting liquidity opportunities, and avoiding market impact through simulation and best execution price.
Customers have access to neural network powered execution capabilities for greater control and deeper trading strategies that reduce information asymmetry and provide best execution guarantees. These changes are partially driven by the regulators (MiFID II), and partly by the competitive landscape. Importantly, simple rules and machine learning capabilities provide best execution compliance in accordance with the intended spirit of the regulations.
Trader behaviour is another major area of concern, whether it is about the use of inside information or collusion with market participants or front running customer trades. The hard thing with bad actors is to identify whether an event has occurred and finding an appropriate resolution for it. Harder still is to find leading indicators that highlight the probability of a future event. It may be easier to change behaviour through deterrence.
Bad actors can precipitate a crisis. For example, Time magazine calls “The LIBOR scandal” the crime of the century. While a student paper had long ahead identified possibilities of rate manipulation,4 it took years for regulators and banks to identify this collusion and take corrective measures. In the LIBOR scandal, communication logs between different bank participants had long hinted there was a potential collusion afoot.
Identifying these communication anomalies in a traditional 3-tier compliance model is a gargantuan task. A natural language processing solution that monitors all communications including email, telephone call transcripts, calendar entries and chat rooms can provide an early indicator of collusion efforts. Non-verbal and verbal cues from traders on the trading floor can also be detected through computer vision and flagged through sentiment analysis. Such a proactive monitoring solution can act as an effective deterrent for the traders against fraudulent activities.
Market participants inevitably need to develop a data and AI strategy to acquire the right set of capabilities to achieve their business outcomes while remaining compliant. A key part of such a strategy is to create a flywheel of data. Simplifying customer journeys can create opportunities for additional data acquisition that can serve multiple goals. For example, capturing a geotagged photo of the person at their home with a proof of address during the application process can reduce fraud, deliver efficiencies for the back-office and speed up the overall onboarding process.
Market participants also need to avoid the fake AI phenomenon, as solutions regularly get branded as AI solutions even if they implement simple business rules or have a ‘Wizard of Oz’ implementation with humans providing the intelligence. Solutions need to deliver unique data sets and insights, demonstrate deep subject matter expertise and ability to integrate with other services.
Greater efficiencies can be achieved by combining human decision-making with predictions from AI solutions. Humans cannot provide the scale and speed to combat the increased volume of work, whereas AI solutions need multiple feedbacks to deliver more accurate results. Aided by a machine learning solution, it is easier for an auditor to review for any incriminating communications, or for a lawyer to preserve confidentiality through redacting select parts of the document. Regulators can also aim for an online dashboard that integrates data, methods and indicators to assess the financial stability. Real-time simulations using this data with stressed inputs can provide better insights into the financial stability than infrequent reports.
Well-designed AI solutions can provide both customer benefits and compliance as an outcome by enabling: