Giovanni Covi, James Brookes and Charumathi Raja
How banks are exposed to the financial system and real-economy determines concentration risk and interconnectedness in the banking sector, and in turn, the severity of tail-events. We construct the Global Network data set, a comprehensive exposure-based data set of the UK banking sector, updated quarterly, covering roughly 90% of total assets. We use it to study the UK banking system’s microstructure and estimate the likelihood and severity of tail-events. We find that during the Covid-19 (Covid) pandemic, the likelihood and severity of tail-events in the UK banking sector increased. The probability of an extreme stress event with losses above £91 billion (roughly 19% of CET1 capital) increased from 1% before the pandemic to 4.1% in 2020 Q2, subsequently falling to 1.7% in 2021 Q4.
The role of concentration risk and interconnectedness in the economic and financial system
Concentration risk in the economic system, such as vulnerability to shocks to large non-financial corporations, may lead to remarkable fluctuations in economic activity (Gabaix (2011)). The level of interconnectedness in economic activity, such as a high level of interdependency in the intersectoral input-output linkages of firms, that is, how a firm’s output is used in the production function of another firm as input, may explain aggregate fluctuations in output (Acemoglu et al (2012)). These network features – concentration risk and interconnectedness – also play an important role within the financial system in determining fluctuations in the level of systemic risk. Stress-testing models aimed at capturing tail-risk interdependence and the level of systemic risk need therefore to take these network features into account, so as to model the financial system’s stability through the lens of its market microstructure.
Stochastic microstructural stress-testing models
Developing policies that reduce the build-up of systemic risk and preserve the stability of the financial system is an increasingly relevant task for regulators worldwide. The risk environment is continuously evolving, and risks may arise from within the system depending on how banks’ exposures are distributed across asset classes, firms, sectors and countries. This requires developing sound analytical tools to interpret and forecast risks. There are different methodologies aiming at assessing the propagation of risks from the real economy to the banking sector’s balance sheet. In this post, we use a microstructural stress-testing methodology to assess solvency risk. This methodology is very handy because it allows regulators to decompose the sources of risk according to each individual component of the network, and perform ad-hoc counterfactual policy exercises.
Our modelling approach measures solvency risk of the UK banking sector as a function of:
- The network structure of UK banks’ exposures, thereby capturing the role played by interconnectedness and concentration risk.
- Counterparty risk such as counterparties’ probability of default (PD) and loss given default (LGD) parameters, capturing the severity of potential shocks (one year ahead) to the real economy. This set of parameters is estimated by UK banks according to the sector and country of the counterparty using obligor level data and they are provided as supervisory data COREP template C.09.02.
- A correlation matrix of counterparties’ default probabilities, which aims to approximate the inter-sectoral input-output linkages of firms and so models tail-risk interdependence.
Finally, we perform this methodology for 20,000 simulations in order to derive a full distribution of banks’ losses and so model scenario uncertainty over time. This stochastic approach to scenario design allows us to capture the entire spectrum of the severity of potential stress events and assess their outcome in probabilistic terms. Further details about the methodology and results are provided in this working paper.
Measuring capital at risk
We derive two forward-looking measures of solvency risk (one year ahead) – a capital at risk measure (CAR) and a conditional capital at risk measure (CCAR). The former aims to track the build-up of expected losses or average risk in the UK banking sector, whereas the latter is calibrated to the 99th percentile of the loss distribution to capture extreme stress events or to the 97.5th percentile, to capture severe stress events. We therefore track the build-up of the average and tail risks in the UK banking sector and compare their likelihood over time, focusing on sizing the build-up of tail-risk during the Covid pandemic.
Big granular data sets
We construct the Global Network data set, which comprises of loan, security and derivative exposures from a number of different data sets collected for supervisory purposes – Table A. This covers roughly £9.4 trillion or 90% of the UK banking system’s assets. The data set consistently maps UK banks’ exposures to counterparties across various sectors of the economy and countries. The data set is divided into two main categories of exposures. Granular exposures refer to exposures mapped at an entity-to-entity level which account for 43% of total exposure amounts (£4.1 trillion). The remaining aggregate exposures at mapped at a sector-country level.
Table A: The global network data set (£ billion)
Note: GG refers to general government, FC to non-bank financial corporations, CI to credit institutions, HH to the household sector, NFC to non-financial corporations and CB to central banks.
We find that the probability of experiencing an extreme stress event above £91 billion losses which is equal to 19% of UK banking system’s CET1 capital (3.4 times the average loss) reached its peak of 4.1% in 2020 Q2, from 1% during the pre-pandemic period (left-hand panel, Chart 1). Moreover, the severity of extreme stress events has also increased, with CCaR estimates amounting to £147 billion, almost 62% higher compared to the pre-pandemic period. Similarly, the likelihood and severity of severe (97.5th percentile) stress events has increased too. Most of this increase is due to higher counterparty risk in the corporate sector (higher PDs) and due to the build-up of risk outside the UK. Last, we estimated expected losses (CaR) – the mean of the loss distribution – which averaged at £27 billion pre-pandemic and £37 billion at the peak of the crisis, representing an increase of 36%. In 2021 Q4, the CaR estimate was still above the pre-pandemic level and close to £31 billion.
Chart 1: Probability and severity of tail events in the UK banking sector
Conclusions and policy implications
The probability and severity of extreme stress events in the banking sector depends, first of all, on the level of fragility in the real economy, that is, on current economic and financial conditions which are captured by the set of risk factors – PD and LGD parameters. Nevertheless, banks’ exposure to the real economy and the financial system further exacerbate the severity of rare tail events as well as increase their probability. Thanks to stochastic microstructural stress-testing methodologies, we are able to assess how the microstructure of the banking system and its defining features – concentration risk and interconnectedness – jointly with the structure of the real economic network play a key role in the realisation of such rare extreme stress events like the 2008 Great Financial Crisis. The very same probabilistic scenario may result in a very different outcome depending on how financial and real economic relationships are distributed, and on the set of firms that are negatively affected by the deterioration in economic and financial conditions, that is, the distribution of shocks. Overall, a higher level of counterparty risk in the real economy, or a higher level of input-output integration among firms in the real economy, and a more interconnected and concentrated banking system’s network of exposures increase the severity and probability of rare tail events affecting the banking sector. This microstructural approach has been applied to measure the impact of the Covid pandemic on the UK banking sector’s probability of experiencing such extreme stress events. We found that this probability has increased by 310% at its peak in 2020 Q2 and still in 2021 Q4 remains higher compared to pre-pandemic levels due to higher counterparty risk.
Giovanni Covi works in the Bank’s Stress Test Strategy Division, James Brookes works in the Bank’s Advanced Analytics Division and Charumathi Raja works in the Bank’s Banking Capital Policy Division.
If you want to get in touch, please email us at email@example.com or leave a comment below.
Comments will only appear once approved by a moderator, and are only published where a full name is supplied. Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.