AboutPeopleEventsThe Actuarial WattContactJoin

Cracking portfolios under pressure

Written by .
Published 1st November 2018.
Stress testing is a method used to predict how portfolios will fare in a financial crisis; but how did it all begin and how will it evolve?

Bank stress testing as a discipline has come a long way since the financial crisis of 2008, with governments around the world placing increasing demands on banks to demonstrate their ability to sustain economic shocks. The models have become more complex, the stress testing scenarios are more rigorously selected and, overall, banks in the UK have actively improved their financial health to meet the requirements from the regulator. However, increasing data demands and emerging risk areas such as cyber risk continues to drive innovation in this space. Ever more complex models drive a need to better manage model risk and provide adequate review and challenge of model inputs and outputs.

The stress testing of both individual banks and banking systems is an exercise in assessing the resilience of the institution(s) in response to extreme macro-economic events. For many regulatory tests, the bottom line becomes consideration of a bank’s Common Equity Tier 1 Capital Ratio (CET1) - a much-used indicator of bank health. Implicit in this ratio are the outputs of the bank risk models reliant upon macro-economic data. It is natural to stress this ratio under macro-economic shock scenarios (where variable paths are generated in response to a hypothetical or historical economic crisis, such as an oil-price snapback or a global recession) and impose ratio thresholds upon banks to determine if they ‘pass’ or ‘fail’ the test. Regulators can then intervene in the event of a stress test ‘failure’ to attempt to increase the bank’s resilience through mandatory management actions.

The outputs of these analyses are becoming ever more relevant to bank management in developing strategy and to regulators in setting macro and micro prudential policy due to the increasing quantities of data that are available for processing. Actuarial techniques concern themselves with the understanding of risk, in all forms, and this skillset is well aligned to meeting the extreme requirements of these exercises.

The past

The defining event which marked a shift in the use and methodologies of stress testing was the financial crisis. Before the crisis, from the early 1990s onward, stress tests were used predominantly as management information exercises (down to trading desk manager levels) and regulatory tests tended to be much less rigorous. It was not until the introduction of Basel II (an international capital regime) in 2004 that the regulatory landscape shifted in favour of banks submitting formal credit risk (CR) calculations for regulatory supervision and to perform stress tests upon the output of these models. CR is the majority contributor to a bank’s Risk Weighted Assets (RWA) which drives the CET1 ratio and, given the huge scope of this modelling process, banks were still in the process of building out these models in the run-up to the financial crisis.

The types of stress testing models being used pre-crisis were often simpler than more recent models. While simpler models are easy to understand, and often to compute and maintain, many were limited to a collection of single factor stress analyses. This type of analysis concerns itself with stressing only one macro-variable and quantifying a direct impact upon bank health and, as a result, outputs were easy to communicate to stakeholders. The downside of many of these early stress tests is that they did not consider the interconnectivity of the macro-variable inputs to models of bank risk and income, leading to poor capturing of economic phenomenon such as contagion effects and feedback loops.

The financial crisis of 2008, widely accepted to be in part a symptom of banks’ aggressive mortgage selling and bundling policies, was an event which stressed the banks far more than any economic scenario inspired by recent history could have. A quotation from a Bank of England (BoE) working paper, published in 2005 examining the stress testing of UK banks, is helpful in expressing the inability of many early scenario generation techniques to comprehend the growing likelihood of a global crash. This paper attempted to address the issue of feedback effects less present in other contemporary stress tests.

[…] even if the most extreme economic stress conditions witnessed over the past two decades were repeated, the UK banking sector should remain robust.” — Bank of England Working Paper No. 282, November 2005.

It is unfair to dismiss this conclusion as poor retrospectively, given that exercise was performed based only on historical data from the 1980s-2000s. However, it does illustrate the irrefutable challenges of the scenario generation and macro-variable expansion phase of a stress test. It remains to this day one of the most challenging and challenged elements of a bank’s stress test, requiring Board of Director involvement and approval.

The effect of the crisis upon stress testing as a tool was profound as regulators adopted new methodologies to better assess and control their systems’ health. The crisis led to the UK’s banking regulator, the BoE, initialising its annual concurrent stress testing exercises which began in 2014.

The future

Stress testing in the UK (and further afield), currently operating under the Basel III framework, is undergoing a period of transition through a fundamental reform of the International Financial Reporting Standard, pertaining to the measurement of credit impairment (the charge banks must declare on their profit and loss account to provision for loans that are expected to “go bad”). The paradigm has shifted from an incurred loss model, where impairment was calculated based on past experience, to a forward-looking expected loss model under IFRS9. This has proven challenging for stress testers, as it has required considerable econometric techniques to be introduced to many models. In combination with the fact that stressed scenarios usually operate outside of “normal” bank model parameters, this adds extra complication to the design and management of stress testing models. The BoE specifically designed this year’s annual cyclical scenario (ACS) to help banks quantify the impact of IFRS9 on stress test results. The outcome of this exercise will be released by the BoE by year end. The BoE has also published a supervisory statement “Model Risk Management Principles for Stress Testing” to address the issues associated with models becoming ever more complex and non-harmonised.

Looking beyond the present challenges, the stress testing space is likely to undergo some important changes to better deliver value to both banks and regulators. Principal amongst these changes is likely to be a greater emphasis being placed upon Big Data engines such as Hadoop to allow more powerful inferences to be made across models. Equally, the industry is keen to develop automation of many reporting and basic mechanical functions within the stress testing process to free up resources to focus upon understanding and challenging model outputs.

A ‘hot-topic’ in the industry at present is the discussion surrounding how to better model cyber risk and how this should be regulated. In the BoE’s most recent Systemic Risk Survey, cyber risk was cited as a key source of risk by 61% of banks. While cyber events are most often not systemic to the industry, the BoE has expressed concern that inhibition of bank services to the real economy, after an outage, could create ripple effects throughout the economy. To address this, the BoE intends to collaborate with the National Cyber Security Centre to devise a separate test to the annual concurrent stress test.

There are many other lower-level methodological changes being implemented by the BoE in upcoming stress tests, not least amongst these is the treatment of group risk and double leverage requirements in light of new bank ring-fencing regulation. More information can be found on the Bank of England’s website.