Foundations of System-Wide Financial Stress Testing with Heterogeneous Institutions

In a highly connected financial system, seemingly localised shocks can be amplified and propagated to take on systemic importance. During the financial crisis from 2007-2009, problems that started in the real economy with increasing sub-prime mortgage defaults quickly spread. Asset price falls on mortgage-backed securities prompted margin calls that put pressure on hedge funds leading to correlated sales that further depressed prices. Banks’ common exposures to these assets put pressure on their solvency, leading to the wholesale funding run on Lehman Brothers. Its subsequent default triggered solvency contagion to hedge funds, banks and money market funds as well as a freeze in interbank markets. Similarly, the ‘dash for cash’ in March 2020, which saw different sectors scramble for liquidity in response to shifts in asset prices and accompanying margin calls, highlighted the interrelationships between different institutions and markets. 

These dynamics underscore the challenge for regulators to understand the plethora of different types of financial institutions and markets, their interconnections, and their interactions under stress. This challenge is compounded by the fact that the financial system constantly changes, not least in response to new regulation, that relevant data of sufficient quality is often lacking, and that the interacting effects of the suite of regulatory reforms can lead to new risks. The challenge for regulators is to constantly evaluate these risks to the resilience of individual institutions and the financial system as a whole.

Few regulatory instruments embody this challenge as clearly as financial stress tests. Microprudential stress tests, like the US Federal Reserve’s Supervisory Capital Assessment Program introduced in 2009, focus on modelling the first order impact of a defined macroeconomic scenario on banks’ balance sheets. Although they can provide useful information about institutions’ balance sheet exposures under stress, their narrow focus on the resilience of individual institutions makes them ill-suited to study systemic risk. ‘The system is not the sum of its parts’, so to understand how shocks can propagate and amplify, regulators need stress tests that capture the effect of endogenous shock amplification.

This realisation has motivated a push to develop macroprudential stress tests. At first, such efforts focused on the interaction between similar institutions, primarily banks, but researchers and regulators increasingly recognise that this is not enough: endogenous shock amplification also involves interactions with non-banks and a multitude of different markets. In tandem with developments in macroprudential stress tests for banks, regulators therefore increasingly look for ways to develop macroprudential stress tests that are ‘system-wide’.

Such macroprudential stress testing of the wider financial system is still in its infancy. Pioneering work in this field has focused on specific markets and the interactions among representative sectors, representing an important but ultimately incomplete advance in the understanding of system-wide resilience. The main obstacles to further progress are associated with the difficulty in designing a modelling framework that (1) comprehensively captures amplification of solvency and liquidity shocks and (2) takes account of the heterogeneity of institutions and their responses to these shocks given the constraints they face. Even if this modelling challenge is solved, scaling and flexibly adjusting such system-wide models in data-driven ways to account for the ever-changing characteristics of different (subsets of) financial systems presents novel computational challenges that microprudential stress tests avoid.

In this new Bank of England working paper, we address those challenges and propose a structural framework for the development of system-wide financial stress tests with multiple interacting contagion and amplification channels as well as heterogeneous financial institutions. This framework conceptualises financial systems through the lens of five building blocks: financial institutions, contracts, markets, constraints, and behaviour. These blocks can be flexibly implemented to form a dynamic multiplex network using the accompanying software engine and library (the ‘Economic Simulation Library’). Depending on the needs of regulators and researchers and the data they have access to, this framework supports both stylised stress testing models as well as large-scale, data-driven models that map out the financial system with a high degree of verisimilitude.

Using this framework, we implement a system-wide stress test model for the European financial system. This model captures solvency and liquidity channels, incorporates four interacting amplification channels, and takes account of the heterogeneity of financial institutions. To evaluate the complementary value of this system-wide approach, we compare our findings to the results of the regular micro-prudential European Banking Authority (EBA) stress test from 2018.

This comparison yields three main findings, which are robust to extensive sensitivity and robustness checks. First, even if microprudential stress tests suggest resilience to a particular financial shock, the financial system may still be unstable. This result hinges on the shock-amplifying tendency of the financial system (also covered here), which we show can in turn be heightened by the interaction of different contagion channels and different types of institutions. This finding suggests that there is a complementary role for system-wide financial stress tests to evaluate financial stability.

Our second finding is that banks’ willingness to draw on their capital buffers to absorb losses—which we term the ‘usability’ of capital buffers—significantly affects the shock-amplifying tendency of a financial system. If banks take actions to avoid using their buffers in response to an adverse shock, this can generate pro-cyclical dynamics that substantially increase system-wide losses. In light of this result, regulators should be mindful of how the design and enforcement of regulatory buffers may affect their ‘usability’ in times of financial stress, a challenge that is especially salient now that regulators have lowered buffers to facilitate the provision of capital to the real economy in response to the COVID-19 shock.

Finally, we find that microprudential stress tests that omit endogenous amplification mechanisms may underestimate the (usable) regulatory buffer that is required to ensure the resilience of individual institutions and the financial system as a whole. Currently, regulators mostly use the results of banking sector stress tests to calibrate the discretionary (time-varying) capital requirements under Pillar II of the Basel capital adequacy framework; while the Bank of England also uses its banking sector stress test to calibrate the macroprudential countercyclical capital buffer. Our findings suggest that system-wide stress tests should be more widely used to meaningfully complement microprudential stress tests when calibrating capital buffers.

Doyne Farmer is Professor of Mathematics at the University of Oxford and Director of Complexity Economics at INET Oxford. 

Alissa Kleinnijenhuis is a Postdoctoral Associate at the MIT Sloan School of Management and a Research Fellow at INET Oxford. 

Paul Nahai-Williamson is an Analyst at the Bank of England. 

Thom Wetzer is Associate Professor of Law and Finance at the University of Oxford and a Senior Research Fellow at INET Oxford. 

The views expressed in this post and the associated paper are solely those of the authors and cannot be taken to represent those of the Bank of England, members of the Bank of England’s Monetary Policy Committee, Financial Policy Committee or Prudential Regulation Committee, or to state Bank of England policy.