Managing Risk: The Power of Federated Learning
Background: Expanding the Conversation Around Risk
In the first part of our exploration into the impact of Federated Machine Learning (FML) on financial risk management, we focused primarily on the concept of de-risking. This practice, borne out of a desire to mitigate the considerable risks associated with financial crime, has led to a harmful trend: financial institutions are exiting entire markets and severing relationships with categories of clients. While this does indeed limit exposure to financial crime risk, it also results in unintended collateral damage such as economic exclusion and stifled growth, particularly in developing economies.
Our focus was on the correspondent banking system, a vital artery in the body of global finance in which larger, correspondent banks facilitate international services for smaller, respondent banks. We noted that this system is becoming constricted due to de-risking. Larger banks, in their quest to mitigate potential risks, are stepping back from their correspondent roles, leaving smaller respondent banks in the lurch. As we highlighted, this scenario not only impacts the financial institutions involved but also significantly affects individuals, businesses, and the wider global financial system that relies on these services.
However, de-risking represents just one strategy in the broader landscape of risk management. It’s an approach that swings the pendulum to one extreme: eliminating risk exposure by completely avoiding certain markets or client categories. Part 2 of our exploration expands this conversation, to put de-risking in context among the other approaches to risk management, and to examine how FML can play a transformative role in not only mitigating the challenges of de-risking, but also enhancing risk management more generally.
Risk Management
In banking, four primary strategies form the bedrock of risk management:
1. Risk Transference
2. Risk Acceptance
3. Risk Avoidance
4. Risk Reduction
1. Risk Transference
Risk transference is essentially passing the risk to a third party willing to bear it – a key feature of insurance agreements. However, when it comes to financial crime, banks cannot transfer the risk. It is an inherent risk tied directly to the bank’s operations, and the buck ultimately stops at the institution’s door. Even with the introduction of innovative technologies like FML, this fact remains unchanged. A common misconception is that the FML central model might bear the risk, but in reality, it aids in mitigating the risk through improved control mechanisms, leading us back to risk reduction.
2. Risk Acceptance
Risk acceptance refers to a conscious decision to bear the risk due to the cost of mitigation outweighing the potential loss, or because of the opportunities that come with the risk. A bank’s level of risk acceptance should generally be a relatively fixed parameter, established with an holistic view of its business operations and strategies including its broader risk appetite. This level should not be changed on a frequent basis. Therefore, when assessing the risk of customers or markets, it’s safe to assume that a bank’s level of acceptance is approximately fixed – this leaves only two remaining strategies to draw on when assessing the financial crime risk of new markets.
3. Risk Avoidance
Risk avoidance is the total elimination of risk by refraining from actions that could lead to its manifestation. In banking, this implies a complete disengagement from markets, products or customers that may potentially expose a bank to financial crime. This strategy can be viable when risks are potent enough to cause significant harm, and effective control measures to mitigate them are absent.
However, an overemphasis on risk avoidance can lead to missed opportunities. In the context of de-risking, risk avoidance becomes the default approach when a bank can’t accept, transfer, or reduce risk to align with its risk appetite.
4. Risk Reduction
An alternative to the strategies discussed so far is risk reduction. Banks can aim to reduce their risk exposure within a particular group by selecting exactly which parts of the group they wish to engage with, based on individual-level risk assessments.
While they cannot change the risk associated with the group as a whole, banks can choose a subset of customers from the market whose risk is within the bank’s acceptance level, thereby opening new opportunities for revenue and promoting financial inclusion. Achieving this in new, unfamiliar markets requires advanced technologies to discern the true risk of each customer accurately. This is where technologies like FML come into play, offering a nuanced, effective approach to risk management.
The Power of Federated Learning: Expanding Capabilities within Risk
FML is an innovative approach to machine learning that has shown immense potential in expanding the capabilities within risk management, especially in the context of correspondent banking. At its core, FML involves distributing the process of training a machine learning model across multiple devices or servers holding local data samples and communicating the local updates to a global model residing on a central server.
In the realm of correspondent banking, FML allows respondent banks to adopt the central model, a system that is familiar, comprehensible, and trusted by the correspondent banks. This adoption bolsters correspondent banks’ engagement in transactions, as the respondent bank’s customers are evaluated using a proven and trusted model.
Through FML, banks can utilise a central model which has been exposed to a wide variety of data sources, making it more robust to scenarios outside of their usual engagements, such as entirely new markets. Therefore, FML is uniquely positioned to maximise a bank’s ability to measure the risk in a market accurately, and thereby reduce it, as opposed to avoiding the market altogether due to uncertainty in the absence of risk-measuring technology.
Risk vs Uncertainty
While risk pertains to situations where the outcomes are unknown but the distribution of outcomes is known, uncertainty refers to situations where the outcomes and the distribution of outcomes are unknown.
In this context, the respondent banks also face risk concerns. They rely almost entirely on the KYC/CDD conducted by their correspondents when mitigating the risks of correspondent banking. In the traditional operating model, respondents also do CDD on correspondents and vice versa, when establishing these relationships. Beyond that, a respondent bank will have no view of the risk on the correspondent’s side of the transaction, relying entirely on the correspondent’s CDD.
With FML, correspondent banks “share technology” with respondent banks. This sharing grants the respondent banks a view into the correspondent’s underlying controls and risk appetite, thereby inferring the level of risk that they will accept with their own customers directly. This level of visibility and control is particularly crucial in light of major AML failings by large banks[1].
Conclusion
FML is more than a valuable tool for mitigating the problems of de-risking through risk avoidance. It’s a powerful asset within any risk management framework, expanding a bank’s capacity to precisely measure risk and differentiate it from uncertainty – enabling firms to take advantage of opportunities within accepted levels of risk, and to avoid those that are not.
By promoting technology sharing between correspondent and respondent banks, FML enhances the visibility of respondent banks, who might otherwise be exposed to substandard anti-money laundering controls in large correspondents. Thus, FML is transforming the landscape of risk management in banking, ensuring safer and more inclusive financial systems.
