Stress testing in a changing risk environment

Keynote speech by Patrick Montagner, Member of the Supervisory Board of the ECB, at the 2026 workshop of the European Systemic Risk Board (ESRB) Standing Task Force on Scenario Design and Development

Kraków, 23 October 2026

Good morning,

I am very pleased to join this workshop, as its programme illustrates how the scope of stress testing is expanding beyond banks and private markets to encompass geopolitical and cyber risks as well as new approaches to scenario design.

The next crisis is unlikely to resemble any scenario we design in advance. We may identify some of its components, but the precise sequence of events, the speed at which they interact and the responses of institutions, markets and public authorities are much more difficult to predict.

This does not mean that stress tests are without value. Quite the opposite: although tress tests cannot predict the next crisis, they can reveal weaknesses, challenge assumptions and improve our capacity to respond. Their usefulness hinges on asking a clear question, choosing an appropriate method and interpreting the results with due caution.

In my remarks today, I will start by describing the different purposes served by stress testing. I will then discuss how we can use stress-testing exercises to assess emerging risks before outlining some priorities for further developing our stress-testing framework.

Stress tests serve different purposes

Stress testing has played an important role since the establishment of the European Banking Authority (EBA) in 2011 and the Single Supervisory Mechanism in 2014.

Since then, European banks have significantly strengthened their capital and liquidity positions. They have also improved their risk management practices, although progress remains uneven. Meanwhile, stress testing has continued to evolve, both in its objectives and in the methodologies used.

Today, EU-wide stress tests are an important input into the Supervisory Review and Evaluation Process (SREP) and for determining Pillar 2 guidance, which is a key element of the ECB supervisory toolkit and should remain part of it. These tests combine several elements. First, banks predict the impact of a common scenario using their own models, subject to certain methodological constraints. Supervisors then conduct an extensive quality assurance process, challenge banks’ submissions and compare results across institutions. Top-down models and benchmarks provide additional points of reference. This framework serves a clear supervisory purpose: to have sufficiently comparable results across banks to support consistent assessments. Common definitions and constraints are therefore essential.

But comparability comes at a price. The constrained balance-sheet approach ignores many of the decisions that banks take during an actual downturn. Likewise, a common scenario will inevitably be more relevant for some business models than for others. And a three-year projection can only provide a simplified representation of the complexity of a crisis and the speed with which it unfolds. These are clear limitations, but they do establish the boundaries within which the results should be interpreted. In short, EU-wide stress tests provide information on the impact of a specified scenario under a common framework. They reveal differences in banks’ exposures, models, data and capacity to explain their results, and provide insights to support capital assessment and ongoing supervision.

Banks’ internal stress tests have a different objective. They are meant to reflect each bank’s specific exposures and funding structures, since a common supervisory scenario cannot capture the particular vulnerabilities of every bank. For a bank that relies heavily on market funding, for example, assessing the impact of a sudden loss of market access may yield more valuable information than further refining a common macroeconomic scenario.

Internal stress tests are most useful when they inform strategic decisions. They should feed into capital and liquidity planning, risk appetite, business strategy and contingency arrangements. This requires board members and senior management to be actively involved, by challenging the scenarios, understanding the main drivers of losses and assessing whether proposed management actions would be feasible under the envisaged conditions. The value of stress testing is significantly reduced when it becomes a specialised modelling exercise disconnected from managerial decisions. Technical sophistication is only useful if the results are understood and acted upon.

Macroprudential stress tests address a different set of questions. Their purpose is to assess the resilience of the financial system and understand the channels through which stress can spread. A bank-by-bank analysis may show that each institution has sufficient capital to absorb a given first-round shock, but it does not capture the consequences of simultaneous reactions.

For example, individual banks may seek to sell the same assets, reduce credit to the same counterparties or raise funding in the same markets. Meanwhile, redemptions and margin calls can increase demand for liquidity, triggering additional asset sales. This can put pressure on prices and funding conditions, generating further losses. In a systemic crisis, correlations increase and defensive actions that are prudent from the perspective of an individual bank may end up amplifying stress across the system.

Macroprudential exercises try to capture these common exposures and interactions. They assess whether the financial system can continue to provide its essential services to households and firms during a severe disruption.

Before moving on, let me recap the difference between these stress tests: supervisory solvency tests support capital assessment and cross-bank comparability, banks’ internal exercises inform their own decisions and macroprudential exercises examine collective reactions and feedback loops. We should assess each exercise against the purpose for which it was designed. No single stress test can simultaneously deliver a comparable capital assessment, capture each institution’s specific vulnerabilities and reflect the behaviour of the financial system as a whole.

Combining risks: the limits of scenarios

A conventional macroeconomic stress test usually starts with a specified deterioration in economic and financial conditions. These shocks are translated into credit losses, market losses, income effects and changes in risk-weighted assets. There is substantial uncertainty at every stage of this process. Nevertheless, historical evidence of past recessions, interest rate movements, market corrections and credit cycles can help inform the scenario and provide a basis for challenging the results.

Interactions between risks are much harder to assess. Even when historical data on these relationships are available, future shocks may propagate through the financial system differently. Moreover, risks that have remained dormant for some time can suddenly re-emerge, as recent geopolitical events have shown. Climate and nature-related risks are another example. Although they were identified more than three decades ago, the financial sector has only recently started to take them seriously enough. I am well aware, however, that accounting for these risks is no easy task, as they operate over different horizons and may combine slow-moving developments and acute shocks. Current climate scenarios do not yet capture every relevant source of risk. Non-linear effects, tipping points, compound events and feedback loops remain particularly difficult to quantify. Their results should therefore be understood as partial estimates that need to be complemented by sensitivity analysis and expert judgement^[1].

The origin of a shock tells us little about its ultimate financial impact. Geopolitical, cyber and climate-related events can cut across established risk categories, generating financial and operational effects that interact as stress unfolds.

As these interactions develop, historical correlations may become less reliable. Behaviour has a decisive influence on the outcome. Institutions and investors may revise their expectations, withdraw financing, call for additional collateral or seek to reduce exposures. Government and central bank measures may alter the course of events. We are therefore faced with radical uncertainty. We can identify relevant vulnerabilities and possible transmission channels, but we cannot map every plausible sequence of events.

Adding detail to a scenario can improve the analysis of a particular transmission channel, but it cannot make the scenario comprehensive. Beyond a certain point, additional complexity may create an false sense of precision. This is particularly important when results are communicated. A projected capital impact may appear exact even though it rests on uncertain assumptions about behaviour, market functioning and policy responses. We should therefore be cautious about both extrapolating past patterns into the future and interpreting model outputs far beyond the scope for which the models were designed.

Against this background, exploratory exercises can be a particularly useful complement to the supervisory, internal and macroprudential tests I talked about earlier. These exercises can assess preparedness for a new risk, test the relevance and potential materiality of identified vulnerabilities and inform the design of future stress-testing activities. Their findings extend beyond a final estimate of losses.

The ECB’s 2022 climate risk stress test illustrated this broader approach. The exercise assessed banks’ exposure to physical and transition risks. It also revealed weaknesses in data, methodologies and governance. Many banks relied on proxies and broad sectoral assumptions and struggled to assess risks at the level of individual counterparties and physical assets. They were still developing their ability to incorporate climate-related risks into strategy and risk management. These capability gaps were important supervisory findings. They highlighted the areas that banks needed to improve before climate risk could be measured with greater confidence. Subsequent supervisory work has continued to focus on more granular counterparty analysis, physical-risk assessment and the integration of climate and nature-related risks into banks’ internal processes.

The ECB’s 2024 cyber resilience stress test is another example. It focused on banks’ ability to respond to and recover from a severe cyber incident. It examined crisis-management procedures, internal and external communication, business continuity and the restoration of critical functions. Here, resilience meant more than being able to absorb financial losses. A bank may have a strong capital position and still face severe difficulties if it cannot restore access to reliable data, process payments or maintain essential services. The exercise also highlighted the importance of interdependencies. Banks rely on internal systems, third-party providers, market infrastructures and communication networks. An incident affecting a common provider or a widely used technology could therefore have consequences across multiple institutions simultaneously. Preparedness in this area depends on arrangements that can be tested directly: responsibilities, escalation procedures, recovery plans, data restoration and coordination with third parties. Stress test scenarios allow these capabilities to be assessed under pressure.

The 2026 reverse stress test on geopolitical risk used a different approach. We asked banks to develop plausible scenarios that would lead to a depletion of at least 300 basis points in their Common Equity Tier 1 ratios. This allowed banks to focus on the geopolitical risks that are most relevant to their specific business models and exposures, and resulted in meaningful scenarios involving conflicts, trade and energy disruptions, sanctions and cyber incidents. However, the exercise did also reveal weaknesses in certain areas: some risk assessments were not sufficiently granular, scenarios were not adequately translated into solvency and liquidity metrics, and certain mitigating management actions were not always credible. The credibility of management actions is particularly important, because responses that may appear feasible when considered by one bank in isolation – like selling assets, raising capital or reducing lending – become much less viable when several institutions plan on doing the same thing under stressed market conditions. Reverse stress testing helps expose these assumptions and identify vulnerabilities that a common scenario may not capture.

These three stress-testing exercises highlight an important lesson: the output of a stress test should answer the question being asked. A climate exercise may reveal shortcomings in risk measurement, a cyber exercise can expose weaknesses in recovery arrangements, and a geopolitical reverse stress test may identify critical combinations of shocks and unrealistic management assumptions. Even if such exercises do not produce a capital estimate comparable with that of the EU-wide stress test, they all yield relevant supervisory findings.

Exploratory exercises can also serve as testing grounds for the regular stress-testing framework. Their greater flexibility allows us to explore new risk factors, transmission channels and modelling approaches and to identify which components may eventually be mature enough for inclusion in the EU-wide stress test. The integration of a climate module into the regular ESRB stress test, which built on the lessons learned from earlier exploratory climate exercises, is a good example of how this learning process can work in practice.

Developing a broader and more agile toolkit

So how should the stress-testing framework evolve going forward? As I said before, we should start by considering the objective of the exercise. If the purpose is to inform capital guidance, comparability and robust quality assurance are essential. If the goal is to explore a new risk, a targeted and flexible design may be more effective. If we want to understand how shocks are amplified across the system, the exercise must consider institutions outside the banking sector and how they might react. All this underscores the need for a portfolio of complementary approaches. Applying different methods can provide a more robust view than adding more layers to a single comprehensive scenario. It also allows the outcomes of each exercise to remain interpretable.

It also requires a close dialogue between macroeconomic modellers and supervisors. Macroeconomic expertise supports the coherence and plausibility of a scenario, while supervisory knowledge helps identify the exposures, business models and transmission channels that merit particular attention. This combined perspective helps make the exercise more relevant and its results more interpretable. This Task Force provides a valuable forum for this cooperation.

We also need a more agile framework. Large-scale bottom-up stress tests take time. Banks must prepare extensive submissions, and supervisors must validate data, conduct quality assurance and ensure consistent treatment across institutions. This is justified when the exercise supports formal capital decisions.

Yet other questions call for a quicker response. Newly emerging vulnerabilities often require a faster initial assessment that uses desk-based analysis, targeted data requests or short sensitivity exercises to determine whether more extensive work is necessary. It is crucial that supervisors have access to relevant data, flexible analytical tools and staff capable of adapting existing methods. In this context, ECB Banking Supervision has made significant efforts in recent years to set up agile stress-testing frameworks that enable it to flexibly conduct multiple scenario analyses to support its forward-looking assessments of new and emerging risks. Banks should also be able to run additional scenarios without rebuilding their processes from scratch.

This work is linked to ECB’s and the EBA’s broader efforts to simplify stress testing. These simplification measures will help streamline requirements that consume considerable resources without adding commensurate supervisory value. They should also make thematic exercises more proportionate and improve the efficiency of quality assurance. The ultimate aim is to place greater emphasis on analysis and interpretation: an exercise is more useful when banks and supervisors have time to examine the main vulnerabilities, challenge alternative assumptions and understand differences in outcomes.

The next area in need of progress is system-wide stress testing. The financial system has changed dramatically since the first post-crisis stress tests. It is no longer just financial conglomerates that operate within a broader network of investment funds, insurers, pension funds, private credit providers, leveraged investors, central counterparties and other market infrastructures – banks now do so too. Risks can move across this network through financing, derivatives, collateral and common asset holdings before returning to banks.

This means that assessing each institution separately may provide an incomplete picture, especially when their reactions influence the severity of the stress itself. Individual institutions typically design their responses to shocks without knowing exactly how others will react. Yet when asset sales, portfolio rebalancing and reductions in financing occur simultaneously, they can reduce market liquidity, depress prices and generate further demands for cash or collateral.

System-wide stress tests should therefore examine how different institutions respond to the same shock and how their actions interact. They should capture common exposures, liquidity flows, collateral calls and feedback loops between solvency and liquidity. They should also test whether assumptions made by one sector are consistent with the expected behaviour and capacity of other sectors.

Several recent initiatives are taking this work forward. The ECB’s 2025 macroprudential stress test extension examined contagion between banks, investment funds and insurance corporations using a top-down system-wide model.^[2] Other examples include the system-wide exploratory scenarios conducted by the Bank of England^[3], with the support of the Prudential Regulation Authority and other UK authorities, and the first French system-wide stress test, conducted jointly by the Banque de France, the Autorité de contrôle prudentiel et de resolution and the Autorité des marchés financiers, which covered banks, insurers and asset managers in a single exercise.^[4]

These initiatives provide a strong basis for further work at the European level. The ESRB already plays a central role in this area. Its macroprudential mandate covers the financial system as a whole, and its members include central banks, banking and insurance supervisors, securities regulators and European authorities. Its institutional perspective allows it to bring together the sectoral expertise needed to understand how stress propagates across the European financial system.

Building on the recommendations of its High-Level Group, the ESRB has launched a project to develop a desktop-based system-wide stress test by 2028. This is a major step towards translating the existing analytical framework into a practical European exercise. From a banking supervision perspective, this work will provide valuable insights into risks that originate or become amplified outside the banking sector. It will help test whether assumptions made in bank stress tests about market liquidity, asset sales, access to funding and the behaviour of financial counterparties remain plausible when other sectors are subject to the same shock. It will also reveal channels through which stress may return to banks through markets, collateral and counterparty exposures.

The work of the ESRB and other authorities has provided a substantial analytical foundation for system-wide stress testing. At the same time, existing sectoral stress tests will continue to serve distinct supervisory and regulatory purposes for banks, insurers and central counterparties. Building on this groundwork, the ESRB can further strengthen links between sectoral analyses, promote greater consistency across exercises and help identify vulnerabilities that may remain hidden when each part of the system is assessed separately.

The next step is to implement these approaches more fully by connecting sectoral analyses and developing focused exercises in cases where collective behaviour and cross-sectoral feedback loops are central to the risk. Ensuring that the exercises remain interpretable will of course be essential, as the broader and more complex an exercise becomes, the harder it may be to identify the main drivers of its results and attribute outcomes to particular shocks or transmission channels.

Developing these exercises will be technically demanding, given the uneven availability of data across sectors and the difficulty of calibrating behavioural responses. We should therefore begin with focused exercises covering clearly identified channels and expand them as our data and analytical capacity improve. An interpretable exercise covering the most material interactions will usually provide more insight than an excessively complex model of the entire financial system.

The rapid expansion of investment related to artificial intelligence aptly illustrates the value of this system-wide perspective.

Major investments are now being made in semi-conductor production, data centres, energy infrastructure, cloud capacity and specialised technology. Large financing needs are emerging, supported by banks, capital markets, private credit and other non-bank investors. Competitive pressure encourages big and early investments but expected revenues may depend on rapid adoption and continued technological advantage. At the same time, some projects require major upfront expenditure before generating stable cash flows. Debt, leasing structures, special-purpose vehicles and private lending can make the distribution of risk more difficult to follow. Connections may also develop between technology companies, infrastructure providers and sources of finance. The same expected revenue growth may be used to support equity valuations, credit decisions and collateral values across several parts of the system. A reassessment of these expectations could therefore have a broad impact. Valuations could fall, credit spreads could increase and planned investments could be postponed. Borrowers might need to refinance before projects become profitable. Collateral values could also decline as lenders become more cautious.

Here, stress testing is best used to identify the conditions under which a sectoral adjustment could reach systemic dimensions, rather than to work out the probability or timing of a reassessment. One priority is to map where leverage is accumulating and which institutions are providing finance. It is crucial to understand where equity and credit exposures overlap, whether risks are concentrated among a limited number of institutions and how banks are connected to non-bank lenders and investors.

A system-wide exercise could then examine how falling valuations or weaker expected revenues would affect collateral, refinancing conditions and the behaviour of lenders and investors. This would help pinpoint the channels through which a sectoral adjustment could spread across markets and financial institutions.

For emerging risks of this kind, exposure mapping and targeted sensitivities are a useful starting point. More elaborate loss estimates can then follow as data and methodologies improve.

Conclusion

Stress testing has contributed greatly to the resilience of European banks. It helped repair balance sheets and restore confidence when European banking supervision was established and it continues to support our assessments of banks and their capital adequacy.

In the current risk environment, we need to use stress testing in a broader and more agile way. This will require continued cooperation across institutions and sectors, drawing on the ESRB’s capacity to bring together macroprudential analysis, scenario design and supervisory expertise. The EU-wide stress test will remain a central part of our framework. Alongside it, lighter and more targeted exercises can be launched rapidly to explore new risks and capture evolving trends across the financial system.

We must continue to test new approaches and select the instruments best suited to the risks we are trying to understand. Established exercises remain essential, but they should never become a routine that discourages challenge or creates a false sense of security. With their greater flexibility, exploratory exercises can uncover shortcomings in risk measurement and provide valuable guidance on integrating new risks and methodologies into the EU-wide stress-testing framework.

The financial system is changing fast, and our stress-testing framework must evolve with it.

Thank you.