# Scenario Design and Application of Climate Stress Testing ## Introduction: Why Climate Stress Testing Is No Longer Optional When I first started working in financial data strategy, climate risk was a footnote in most boardroom presentations. Back in 2018, I remember sitting in a meeting where a senior risk officer dismissed climate scenarios as “an ESG checkbox thing.” Fast forward to today, and that same officer is now leading a cross-functional task force on climate stress testing. The shift hasn’t been subtle—it’s been seismic. Climate stress testing is no longer a theoretical exercise reserved for academic journals. It has become a core tool for banks, insurers, asset managers, and even central banks to assess how their portfolios might withstand the physical and transition risks of a warming planet. The European Central Bank (ECB) ran its first economy-wide climate stress test in 2022, covering over 4 million firms and 2,900 banks. The results were sobering: under a severe scenario, losses could reach €70 billion for euro-area banks alone. The Bank of England followed suit with its Climate Biennial Exploratory Scenario (CBES) in 2021–2022, and the Federal Reserve has been slowly but steadily building its own framework. But here’s the catch: a stress test is only as good as the scenarios it runs. Bad scenario design produces misleading outputs, false confidence, or worse—paralysis. This article dives deep into the art and science of scenario design and application in climate stress testing. We’ll explore the mechanics, the pitfalls, the real-world cases, and the future directions that I believe will define this field over the next decade. Whether you’re a risk manager, a data scientist, or a policy wonk, I hope this piece gives you both practical insights and a bit of food for thought. ---

Physical vs. Transition Risk Scenarios

The first and most fundamental distinction in climate stress testing is between physical risk and transition risk. Physical risk refers to the direct impacts of climate change—floods, droughts, heatwaves, sea-level rise, wildfires—on assets, operations, and supply chains. Transition risk, on the other hand, stems from the process of adjusting toward a low-carbon economy: policy changes, technological disruption, market sentiment shifts, and reputational damage. Both are material, but they operate on different timescales and require different modelling approaches. Let me illustrate with a real case. In 2021, a global reinsurance company I worked with wanted to stress-test its property portfolio in Southeast Asia. We ran two distinct physical scenarios: a “chronic heat stress” scenario (gradual temperature rise +2.5°C by 2050) and an “acute flood” scenario (a 1-in-200-year flood event occurring twice within a decade). The chronic scenario affected energy demand patterns and worker productivity, while the acute scenario caused immediate structural damage and business interruption losses. The results were striking—the acute scenario produced nearly 40% higher losses in the first five years, but the chronic scenario became more damaging over a 30-year horizon. This taught me that scenario design must explicitly match the risk profile of the portfolio and the decision horizon of the stakeholders. For transition risk, scenario design often revolves around policy timing and technological adoption curves. The Network for Greening the Financial System (NGFS) has developed six reference scenarios, ranging from “Net Zero 2050” (orderly transition) to “Delayed Transition” and “Hot House World” (disorderly or no transition). Each scenario embeds assumptions about carbon prices, energy mix, and sectoral shifts. But here’s the thing—these macro-level scenarios need to be downscaled to firm-level or asset-level granularity to become useful. That’s where the data pain begins. I’ve seen countless teams struggle with mismatched data granularity—macro scenario outputs at national level vs. portfolio data at postal-code level. The gap is real, and it’s not trivial to bridge. A practical tip I often give to practitioners: don’t start with the most extreme scenario. Start with a “middle-of-the-road” scenario (e.g., NGFS’s Delayed Transition) and then stress-test the key assumptions under sensitivity analysis. This builds credibility with senior management, who are often skeptical of overly apocalyptic numbers. Also, remember that physical and transition risks are not independent—they interact. A delayed transition today might lead to more severe physical risk tomorrow. Your scenario design should capture these feedback loops, even if imperfectly. In my experience, an imperfect integrated model beats two perfect siloed models. ---

Data Infrastructure and Granularity Challenges

If scenario design is the brain, data is the nervous system. Without high-quality, granular, and forward-looking data, any climate stress test is a house of cards. But here’s the uncomfortable truth: most financial institutions still rely on backward-looking, aggregated data that is woefully inadequate for climate risk assessment. A 2023 survey by the International Financial Reporting Standards (IFRS) Foundation found that over 70% of banks admitted to using proxy data for scope 3 emissions, and nearly half said they lacked confidence in their climate data quality. Let’s break down the data challenge into three layers: exposure data, hazard data, and vulnerability data. Exposure data tells you what assets you have and where they are. Hazard data tells you what climate events might occur at those locations. Vulnerability data links the two—how sensitive your assets are to those hazards. Most institutions have decent exposure data (though often siloed in different legacy systems), but hazard and vulnerability data are frequently sourced from third-party vendors with varying methodologies and accuracy. For example, flood hazard maps from one vendor might use a different return period definition than another, leading to wildly different loss estimates. I recall a project with a mid-sized European bank where we tried to stress-test their mortgage portfolio in coastal regions under a sea-level rise scenario. The bank had excellent property-level data, but the hazard layer—from a government source—only provided coastal flood zones at a 1-km resolution. That was too coarse. A 500-meter difference in elevation could mean the difference between dry and underwater. We ended up building a custom downscaling algorithm using LIDAR elevation data, which cost us three extra weeks but fundamentally changed the risk numbers. The lesson? Invest in data infrastructure before you invest in complex models. Garbage in, gospel out, as they say. Another critical aspect is forward-looking data—climate risk is inherently dynamic. Historical weather data alone cannot predict future risk under different emission pathways. This is where climate models (GCMs and RCMs) come in, but they are large, complex, and computationally expensive. Many institutions don’t have the in-house capability to run them, so they rely on pre-processed climate projections from providers like the Copernicus Climate Change Service or the World Bank’s Climate Change Knowledge Portal. The challenge is translating these global datasets (often at 25-50 km grids) into actionable portfolio-level insights. Statistical downscaling methods, machine learning interpolation, and expert judgment all play a role—and all have their own uncertainties. From a data strategy perspective, I’ve seen the most successful institutions build a “climate data mesh” approach, where datasets are federated across business units rather than centralized in a single data lake. This respects the domain expertise of individual teams while enabling cross-functional integration for stress testing. It’s not easy—it requires strong data governance, clear metadata standards, and a pragmatic approach to data sharing. But the payoff is enormous: faster scenario runs, better auditability, and more trust in the results. ---

Methodologies: Top-Down vs. Bottom-Up Approaches

When it comes to actually running the numbers, there are two broad methodological schools: top-down and bottom-up. A top-down approach starts with macroeconomic or sector-level impacts (e.g., GDP loss by sector under a given warming scenario) and then applies sectoral loss factors to the portfolio based on its sectoral composition. It’s fast, scalable, and relatively cheap. But it’s also crude—it ignores geographical and asset-specific nuances. A top-down model might say that “commercial real estate in coastal areas faces an 8% default rate increase,” but it won’t tell you that *your* specific building in Miami is on a flood plain while *your* building in Boston is not. The bottom-up approach, by contrast, starts at the individual asset or counterparty level, applying location-specific hazard layers, asset-specific vulnerability curves, and firm-level financial statements to estimate risk. This produces far more precise results, but the data, computational requirements, and modelling effort are significantly higher. For a large bank with millions of loan exposures, a full bottom-up exercise can take months to run and require dedicated data science and climate science teams. Is it worth it? In my opinion, yes—but only for material portions of the portfolio, not the entire book. Let me give you a concrete example from the insurance sector. Munich Re, one of the world’s largest reinsurers, has long used a bottom-up approach for its catastrophe models. Their models combine high-resolution wind speed and flood depth simulations with detailed building vulnerability curves, and then translate those into insured loss estimates. This works very well for physical risk in property and casualty lines. However, for transition risk (e.g., carbon-intensive lending), they increasingly rely on top-down input-output models that trace supply chain dependencies. The key is to use the right tool for the right question—and to be transparent about which approach you’re using and why. In the banking world, the ECB’s 2022 stress test actually used a hybrid approach. They started with a top-down macroeconomic module to estimate sector-level GDP impacts, then bottom-up elements to assess individual bank exposures to high-carbon firms. The result was a more actionable picture than either approach alone could provide. The takeaway for practitioners: don’t view top-down and bottom-up as either/or. Design a two-layer architecture where top-down results set the boundary conditions and bottom-up results refine the portfolio-specific impacts. This sounds logical, but in practice, I’ve seen many institutions get stuck in endless debates about which “school” to follow. Just pick a pragmatic split—say, 70% of effort on your most material risk areas using bottom-up, and 30% on the rest using top-down, and iterate from there. One more technical point: the time horizon. Climate stress tests often look at 5, 10, or 30 years. Short horizons (5–10 years) are more reliable for modelling but underestimate chronic risks. Long horizons (30+ years) capture tipping points and feedback loops but are inherently uncertain. I tend to recommend a “staged horizon” approach: run a 5-year liquidity and solvency test using near-term physical risk, and a 30-year climate scenario for strategic capital planning. This way, you speak to both the regulator’s short-term prudential concerns and the board’s long-term strategic lens. ---

Regulatory Frameworks and Their Evolution

No discussion of climate stress testing is complete without acknowledging the regulatory tide that is pushing it into the mainstream. The Basel Committee on Banking Supervision (BCBS) has issued its Principles for the Effective Management and Supervision of Climate-related Financial Risks, explicitly calling for banks to use scenario analysis as part of their risk management. The European Banking Authority (EBA) has included climate risk in its stress testing guidelines, and the European Commission’s Sustainable Finance Disclosure Regulation (SFDR) mandates climate risk disclosures for financial market participants. Even in the United States, where federal banking regulators have been more cautious, the Federal Reserve has run pilots with the six largest U.S. banks to explore climate scenario analysis. But here’s the thing about regulation—it’s a double-edged sword. On one hand, it forces action and creates a level playing field. On the other hand, overly prescriptive requirements can lead to box-ticking exercises that consume huge resources without generating genuine risk insights. I’ve been in countless workshops where the primary goal was to “pass the regulator’s methodological review” rather than to truly understand the portfolio’s vulnerabilities. That’s a problem. The best firms treat regulation as a floor, not a ceiling, and use climate stress testing as a strategic tool for capital adequacy, pricing, and even product innovation. Let me share a personal experience from working with a large Asian bank during the HKMA (Hong Kong Monetary Authority) climate stress testing exercise in 2021. The HKMA required banks to run three scenarios: an orderly transition, a disorderly transition, and a hot house world. Initially, our bank’s teams were overwhelmed—they had no existing climate data infrastructure and had to rely heavily on external consultants. The first run produced results that were frankly unusable—negative loss rates in some sectors due to modelling errors. We spent the next three months debugging, cleaning, and recalibrating. By the final submission, the results were coherent and actually revealed an unexpected concentration risk in the shipping finance portfolio under the disorderly transition scenario. That insight led to a strategic reevaluation of that business line. So, yes—regulation can be a pain, but it also forces you to see sharp edges in your portfolio you didn’t know were there. Looking ahead, I expect regulatory frameworks to converge around a common taxonomy of scenarios (thanks to NGFS) but to diverge in terms of implementation details. For instance, the ECB might emphasize P&L impacts, while the Bank of England might focus on capital adequacy and procyclicality. Firms with global footprints should build a flexible scenario design engine that can accommodate multiple regulators’ specifications without multiple rewrites. This is easier said than done—but a well-abstracted data model and modular scenario generator can make it feasible. In a sense, regulation is a forcing function, but it’s not the destination. The destination is a fundamentally more resilient financial system. ---

Case Studies: Lessons from Real-World Applications

To bring theory down to earth, let’s look at three real-world applications that illustrate both successes and cautionary tales. The first is the Bank of England’s CBES (Climate Biennial Exploratory Scenario), which ran in 2021–2022. The BoE required major UK banks and insurers to test their portfolios under three scenarios: early action, late action, and no additional action. One key finding was that the “late action” (disorderly transition) scenario could result in combined losses of around £100 billion for UK banks—roughly 40% of their aggregate common equity tier 1 capital. But more interesting than the headline numbers was the qualitative learning. Many banks reported that the exercise surfaced data and modelling gaps that they hadn’t previously considered material. It forced them to build new data pipelines and cross-functional communication channels. In that sense, the process itself was the product. The second case is from my own backyard—a regional commercial bank in Southeast Asia that I advised on physical risk stress testing for its agricultural lending portfolio. We designed a bottom-up model that linked specific crop types (rice, palm oil, rubber) to site-specific temperature and precipitation projections under a +2.5°C scenario. The model then translated yield shocks to borrower repayment capacity using farm-level financial data. The results showed that without adaptation measures, the default rate could increase by six percentage points over a 20-year horizon. But here’s the interesting part: by overlaying a targeted adaptation scenario (drought-resistant seeds, improved irrigation), the default impact was cut by half. This opened a new business opportunity—the bank launched a “green agricultural loan” product with preferential rates for farmers adopting climate-adaptive practices. The stress test didn’t just quantify risk; it identified a market opportunity. The third case is a cautionary tale from the asset management industry. A large pension fund in Europe decided to run a climate stress test on its equity portfolio using a top-down sectoral approach. The model indicated that technology stocks were relatively “climate-safe” because they had low direct emissions. But this ignored the supply chain dependencies and energy sources of those tech companies. When a second-layer analysis was done—incorporating upstream electricity consumption and supply chain vulnerabilities—the picture changed dramatically. Several large tech names were flagged as highly exposed to carbon price spikes and energy volatility in a disorderly transition. The pension fund narrowed its scenario design to include indirect emissions, and the resulting portfolio adjustments saved them an estimated €300 million in potential losses over a decade. The lesson? Always validate your methodology before trusting the output. A comfortable answer is often an incomplete answer. These cases illustrate three key principles: (1) climate stress testing is as much about organizational learning as it is about numbers; (2) bottom-up approaches unlock actionable granularity that top-down cannot match; (3) unexpected insights are the gold you’re digging for—don’t be afraid of surprises. The downside? Each of these exercises took anywhere from 9 to 18 months from design to delivery. Patience, persistence, and iterative improvement are non-negotiable. ---

Integration with Financial and Strategic Planning

Here’s a trap I see far too often: climate stress testing is treated as a standalone “risk project” that lives in a corner of the risk department and never connects with the broader financial and strategic planning processes. That’s a huge missed opportunity. A well-designed climate stress test is not just a risk measurement tool—it’s a strategic planning tool that can inform capital allocation, loan pricing, product development, and even M&A strategy. But to unlock that value, you need to integrate the outputs into your firm’s existing planning cycles. Let’s get concrete. Your annual budget exercise typically defines targets for revenue growth, credit loss rates, and capital ratios. If climate stress testing is done properly, it should feed directly into those assumptions. For instance, if your climate scenario suggests that 15% of your commercial real estate loans are in zones with high flood risk, your loan loss provisions and capital buffers should reflect that. If you’re a lender to the steel industry, your credit approval standards should embed carbon costs under a transition scenario. This isn’t rocket science, but it requires breaking down organizational silos. In a recent engagement with a mid-sized Asian bank, we built an integrated framework that linked the climate stress test outputs (estimated credit losses, market losses, and operational impacts) to the bank’s ICAAP (Internal Capital Adequacy Assessment Process) and its annual business plan. The CFO was initially skeptical—she saw climate stress testing as a compliance exercise. But after we demonstrated that the 99th percentile climate loss was roughly 40% of the bank’s regulatory capital, she quickly changed her tune. The bank subsequently adjusted its lending limits for exposure to coal-fired power plants and increased pricing for high-emission transport loans. The stress test wasn’t a reporting exercise anymore; it was a competitive tool. To make this integration work, I recommend three structural moves. First, align the time horizons of your stress tests with your strategic planning horizon. If your plan is a 3-year cycle, run at least one climate scenario over a 3-year horizon, not just 30 years. Second, embed scenario outputs into your risk appetite statement. Define concrete limits, like “no more than X% of the portfolio in extreme flood zones.” Third, create a cross-functional committee—comprising risk, finance, strategy, and sustainability—that reviews stress test results at least quarterly. This ensures that climate risk doesn’t get stuck in a silo and that recommendations actually reach decision-makers. ---

Uncertainty, Limitations, and the Art of Communication

Let’s be honest: climate stress testing is fundamentally uncertain. We don’t know the exact emission pathway, the timing of policy interventions, or how technology will evolve. The models we use have large confidence intervals, and small changes in input assumptions can produce very different outputs. This uncertainty is uncomfortable for financial institutions, which are conditioned to seek precision and volatility-based risk measures. But the goal of climate stress testing is not to predict the future—it’s to explore a range of plausible futures and identify vulnerabilities across that range. One of my favourite teaching moments was when a senior risk director asked me, “What’s the probability of this disorderly transition scenario actually happening?” I answered honestly: “I don’t know, and neither does anyone else. But if it does happen, your portfolio loses 6% of its value. Are you comfortable with that?” That reframing from “what will happen” to “what happens if” is key. Climate stress testing is about resilience, not prediction. That said, we still need to communicate this uncertainty effectively. Regulators and boards often want a single number, but giving a single number without confidence bands is dishonest. I always recommend presenting output as a probability distribution or at least as low/base/high sensitivity cases. Communication is also about storytelling. I’ve seen excellent technical models fail to influence decision-makers because the results were presented in dense technical jargon that executives couldn’t connect with. On the flip side, I’ve seen simple visualizations—a map with coloured flood zones overlaying a loan portfolio, or a bar chart of sectoral losses across scenarios—cut through the noise in minutes. My rule is simple: the first slide of any climate stress test presentation should be a one-page executive summary with the headline numbers and a heatmap. The technical details go in annexes. The goal is not to impress with modelling sophistication but to drive action. Another limitation is the temporal mismatch between climate cycles and financial cycles. Climate risk builds up slowly, while financial risk can materialize quickly. This means that a typical annual stress test might miss the slow burn of climate risk. I recommend supplementing traditional stress tests with “reverse stress tests” that ask the question: “What combination of climate events could make our firm insolvent?” This backward-looking exercise is often more powerful for revealing hidden fragilities. It forces you to think about tail risks in a structured way, rather than just tweaking forward-looking scenarios. ---

Future Directions and Emerging Innovations

As we look to the future, several emerging trends are set to reshape climate stress testing. The first is the integration of **machine learning and AI**. Traditional climate models are computationally intensive and struggle with non-linearities, but surrogate models trained on climate outputs can run thousands of scenarios in seconds. These AI-based emulators are not a replacement for physics-based models, but they enable massive sensitivity analysis and real-time stress testing—something that’s simply not possible with traditional approaches. At BRAIN TECHNOLOGY LIMITED, we’ve been experimenting with transformer-based time-series models to downscale climate projections to asset-level granularity, and the results are remarkably accurate while being orders of magnitude faster. Second, **real-time and predictive analytics** are making their way into climate risk. Rather than running quarterly or annual stress tests, some institutions are moving towards “continuous stress testing” where dashboards show live risk metrics that update as new climate or portfolio data arrives. This is a game-changer for early warning systems. Imagine a bank that receives an alert when a new flood hazard map is published for a region where it holds significant mortgage exposure—that’s not far-fetched; it’s already happening in pilot projects. Third, **scenario entanglement**—the idea that physical and transition risks dynamically interact—will become standard. Simple additive models will give way to coupled models that simulate, for example, how a carbon tax might accelerate one region’s transition while causing job losses and credit defaults in another, all while affecting global commodity prices. These integrated models are complex, but advances in computing power and multi-agent simulation are making them tractable. Finally, there’s a push towards **standardization and open data**. The NGFS scenarios are already publicly available, but we need similarly open datasets on assets, hazards, and vulnerabilities. The more we share data, the more we can benchmark and validate models. In the long run, I believe climate stress testing will evolve from an exotic exercise into a standard, almost mundane, part of regular risk management—just like credit risk and market risk are today. The pioneers who build this capability now will have a decisive competitive advantage in the coming decades. ---

Conclusion: Turning Risk into Resilience

Climate stress testing is not a crystal ball; it’s a flashlight. It doesn’t tell you exactly what will happen, but it illuminates the dark corners of your portfolio where hidden risks lurk. The journey from scenario design to application is messy, iterative, and full of uncertainty. But it’s also one of the most intellectually stimulating and strategically important exercises that a financial institution can undertake. I’ve seen firsthand how a well-designed stress test can shift entire business strategies, uncover new opportunities, and strengthen the institution’s credibility with regulators, investors, and customers. The core message I want to leave you with is this: don’t wait for perfect data or perfect models. Start somewhere, iterate, and be transparent about your limitations. Use a hybrid top-down/bottom-up approach, invest in data infrastructure, and integrate stress tests into your planning cycles. And above all, communicate the results in a way that drives action, not just awareness. In the era of climate change, sitting still is the riskiest scenario of all. --- ## BRAIN TECHNOLOGY LIMITED’s Insights At BRAIN TECHNOLOGY LIMITED, we view climate stress testing as one of the most compelling data and analytics challenges of our time. Having developed financial data strategy frameworks and AI-based risk engines for clients across Asia and Europe, we’ve observed that the gap between climate science and financial modeling remains the hardest to bridge. Our belief is that success lies in building **adaptive data architectures** that can ingest multi-resolution climate data, transform it into portfolio-level risk metrics, and enable continuous learning as new data emerges. We advocate for a “no bright lines” approach—meaning that climate risk shouldn’t be quarantined in a separate risk category, but rather embedded into every credit, market, and operational risk decision. We’ve also found that explainability is paramount; board members and executives need to understand not just the “what” but the “why” of stress test outputs. That’s why our AI models emphasize feature attribution and scenario visualisations over black-box predictions. We are also pioneering the use of **large language models** to parse unstructured climate-related disclosures and news, enriching traditional structured stress test data with forward-looking signals. This is still nascent, but our pilots show it can improve the accuracy of transition risk assessments. We believe the next frontier is collaborative. No single organization, research group, or regulator has all the answers. Open standards, shared data, and joint scenario design will accelerate progress and ensure that the financial industry becomes a true enabler of a just and resilient transition.