Malgorzata Tynecka (Nov 24)

The BTRM Newsletter: BTRM Faculty Opinion 

Then and Now of Stress Testing

The financial world looked very different when I joined RBS Global Banking and Markets in September 2008, at the height of the global financial crisis. I witnessed first-hand the stress testing frameworks and risk assessments that, despite their best intentions, proved insufficient in the face of unprecedented systemic shocks. Now, 15 years later, stress testing has evolved profoundly, traversing themes such as “lower for longer” and “higher for longer” interest rates, globalisation, deglobalisation, and shifting regulatory priorities. Here’s my reflection on the journey from 2009 to 2024.

From “Lower for Longer” to “Higher for Longer”

Shortly after 2008, central banks worldwide slashed interest rates to historic lows, making “lower for longer” the norm. Banks were compelled to reimagine their business models in an environment where traditional interest income was squeezed. Stress testing focused on prolonged low rates and liquidity stress, with scenarios aimed at evaluating the impact of deflationary threats and credit risk. In those early years, designing a “worst-case” scenario was straightforward, as the 2008 crisis benchmark remained a relevant reference.

Over a decade on, we’re operating under a new paradigm. Interest rates have trended towards a “higher for longer” trajectory as central banks tackle persistent inflation. Today’s stress testing must reflect this shift, modelling not only the financial impact of high funding costs but also the second-order effects, such as changes in customer behaviour and market volatility. Rising interest rates underscore the importance of stress testing scenarios that assess the potential rise in credit risk due to strained borrower affordability, especially in interest-rate-sensitive sectors like real estate and consumer loans.

Furthermore, technology and social media have accelerated changes in customer behaviour. Remarkably, it was only in May 2011 that the Royal Bank of Scotland launched the world’s first fully functional banking app. In a low-rate environment, there was less incentive for “rate shopping,” but now, customers can transfer funds with the flick of a finger. Are your projections truly reflecting this new reality?

Shifting Sands of Financial Institution Interconnectedness

Nearly everyone who was in banking pre-2008 remembers what they were doing when Lehman Brothers collapsed. It was a historical event and the ripple effects were felt globally, underscoring the need for banks to include global interconnectedness in their stress testing scenarios. I recall intense discussions on how international linkages and dependencies could amplify shocks, exploring scenarios where foreign market turmoil could stress the UK’s financial system.

Fast forward to today, and the world has changed. With rising geopolitical tensions and the fallout from events like Brexit and COVID-19, deglobalisation has reshaped the landscape. Today, stress testing must account for the impact of regional disruptions and fractured supply chains. For instance, a bank’s stress testing framework might now more frequently include a severe scenario simulating the effects of sudden regulatory shifts on cross-border flows or disruptions in critical imports due to sanctions.

From Crisis Prevention to Risk Management

In those early post-crisis years, regulators were laser-focused on liquidity and capital adequacy, demanding robust buffers to withstand severe downturns. Stress testing primarily aimed to shore up capital and liquidity – a critical priority to rebuild public trust.

Over time, regulatory priorities have shifted to encompass a more holistic risk management approach, including operational resilience and ESG factors. Stress testing is no longer just about surviving a crisis. In practice, this means banks now incorporate scenarios assessing the potential financial impact of climate change on lending portfolios or the repercussions of a prolonged cyber-attack on operational resilience. Today, stress testing should move beyond regulatory compliance, preparing banks for a broad array of potential shocks and enabling proactive, strategic decision-making.

How Artificial Intelligence is Transforming Stress Testing

Could this even count as an article written in 2024 without a section on AI? Probably not. One of the most exciting recent developments has been the incorporation of artificial intelligence (AI) into stress testing. AI allows banks to move beyond traditional, manual models to more sophisticated, real-time stress testing. Take the scenario generation process, for example. In the past, we relied on relatively static, backward-looking data and scenario libraries to model potential shocks. AI, however, can process vast amounts of real-time data and simulate a broader range of scenarios based on emerging trends.

Machine learning algorithms can analyse and predict customer behaviours under various interest rate paths, enabling banks to anticipate changes in risk profile dynamically. AI-driven models can also uncover correlations between non-traditional datasets, such as customer transaction data or real-time social media sentiment, and market risks. These insights can help model scenarios where economic shocks are amplified or mitigated by behavioural trends that traditional models might miss.

It’s remarkable to see this technology taking its first steps in action. Instead of relying on annual stress tests based on last year’s data, banks could potentially conduct more frequent, “on-the-fly” stress tests to capture current market conditions and adjust risk appetite. This level of sophistication and agility was unimaginable in 2008, yet it’s essential in today’s complex financial environment.

However, I wouldn’t be a risk professional if I didn’t consider the risks associated with this wave of technological change. Having spent years developing risk and projection models, I am fully aware of the challenges a “black box” element can introduce to understanding the underlying risk profile. Transparency, interpretability, and regulatory alignment are critical in the context of the UK’s Prudential Regulation Authority (PRA) expectations. The complexity of AI models, often involving intricate algorithms and high-dimensional data, can obscure the model’s inner workings, making them difficult to explain and justify. Additionally, AI models may present issues with data biases and overfitting, potentially distorting stress test outcomes and ultimately impacting risk management decisions. This is why I am cautious about jumping on the “AI solution for all” bandwagon.

While AI can automate many routine tasks, human judgement remains indispensable in risk management. The most successful AI implementations are those that augment human decision-making rather than replace it. The professionals must develop a working knowledge of AI and machine learning principles to collaborate effectively with data scientists and technology team.

The onus is on today’s banking professionals to adapt and modernise stress testing frameworks. As we face new challenges, let’s leverage the lessons of the past and the technologies of the present to build a future-ready financial system capable of withstanding whatever lies ahead.