Expert Judgment vs Data-Driven Climate Risk Models

𝗝𝘂𝘀𝘁 𝗥𝗲𝗹𝗲𝗮𝘀𝗲𝗱: 𝗨𝗡𝗘𝗣 𝗙𝗜 & 𝗚𝗹𝗼𝗯𝗮𝗹 𝗖𝗿𝗲𝗱𝗶𝘁 𝗗𝗮𝘁𝗮 𝗦𝘂𝗿𝘃𝗲𝘆 𝘀𝗵𝗼𝘄𝘀 𝗰𝗹𝗶𝗺𝗮𝘁𝗲 𝗿𝗶𝘀𝗸 𝗶𝘀 𝗰𝗿𝗲𝗱𝗶𝘁 𝗿𝗶𝘀𝗸. 𝗕𝗮𝗻𝗸𝘀 𝗮𝗿𝗲 𝘀𝘁𝗶𝗹𝗹 𝗻𝗼𝘁 𝗱𝗼𝗶𝗻𝗴 𝗲𝗻𝗼𝘂𝗴𝗵. Regulators have raised the bar for climate disclosure, but banks are still a long way from embedding climate risks into their business. 🔸  Collateral value adjustment remains low. Just 12% of banks adjust collateral for physical risk, and only 4% for transition risk. 🔸  ESG integration is fragmented. Over half of banks have internal ESG scoring, but there's no consensus. Few banks tie ESG directly into credit decisions, methods vary and full integration into ratings is rare. 🔸  Scenario analysis is widespread, but validation is not. 85% of banks use NGFS climate scenarios, but fewer than 5% regularly backtest climate impacts in credit models. 🔸  Incorporating climate into key credit metrics is lagging. Metrics like Probability of Default (PD), Loss Given Default (LGD) and Internal ratings-based (IRB) models remain inconsistently or only partially integrated with climate risk considerations. 🔸  Adjustments to ECL (Expected Credit Loss), RWA (Risk-Weighted Assets) and Economic Capital remain low and still in early, exploratory stages. Most banks report financial impact for climate risks between 0-2.5%. For transition risks, this increases to 5-10% but this is not reflected in key metrics. There remains a significant gap in quantification and adoption for capital impact. 🔸 Many banks still rely on expert judgement over data-driven models. While climate risk is assessed across major portfolios, most banks depend on judgement, due to data and methodological constraints. 🔸 Data quality & granularity are key obstacles. Obstacles to robust, forward-looking climate data (especially Scope 3) push banks toward proxies and general averages. 𝗠𝘆 𝗧𝗮𝗸𝗲 The UNEP report shows the banking sector still struggles to consistently quantify and integrate climate risk in credit portfolios, capital models, and client processes. Most banks remain reliant on expert judgment and qualitative overlays, mainly due to the lack of granular, forward-looking data and practical scenario analytics. Scenario analysis exists but is rarely deeply embedded in major decisions, and backtesting is the exception. This is where data-driven platforms are critical. Delivering granular scenario analysis, data harmonisation, and dynamic simulation enables banks to move beyond overlays to defensible, auditable climate risk insights. The leaders will be the banks who industrialise scenario analytics and make regulatory pressure a driver of real competitive advantage. #ClimateRisk #CreditRisk #Banking #ESG #RiskManagement #SustainableFinance Source: https://lnkd.in/eC4S8mRN ___________ 𝘛𝘩𝘦𝘴𝘦 𝘷𝘪𝘦𝘸𝘴 𝘢𝘳𝘦 𝘮𝘺 𝘰𝘸𝘯. 𝘍𝘰𝘭𝘭𝘰𝘸 𝘮𝘦 𝘰𝘯 𝘓𝘪𝘯𝘬𝘦𝘥𝘐𝘯: Scott Kelly

The UNEP FI – Bridging Climate and Credit Risk report delves into how 32 global banks are incorporating climate-related risks into their credit risk frameworks. These banks evaluate physical and transition risks across sectors like real estate, energy, and transport, focusing on exposure classes such as large corporates and SMEs. While expert judgment remains crucial, there is a notable shift towards data-driven approaches. The outcomes of climate risk assessments impact regulatory reporting, credit decisions, and client interactions, influencing activities like loan repricing and risk ratings. Despite progress in integrating climate risks into Probability of Default (PD) and Loss Given Default (LGD) models, their integration into internal models like IRB or rank-ordering is still limited. While scenario analysis, including NGFS scenarios, is prevalent, challenges persist with Scope 3 emissions data. More than half of the banks surveyed employ ESG scoring frameworks, but the methods of integration vary due to issues like data quality, methodological constraints, and resource limitations. The report advocates for refining climate-credit risk models, strengthening data governance, and promoting closer engagement with regulators. It emphasizes the necessity for banks to embrace proactive measures like stress testing, margin of conservatism, and broader sustainability integration to effectively navigate long-term climate-related credit risks.

When do complex risk models actually beat structured expert judgment? I’ve managed risk across different companies for 15+ years. The question isn’t whether sophisticated modeling works (it does). The question is: at what stage does complexity deliver proportional returns? 2025 changed accessibility ⚡ Monte Carlo simulations that required $5K licenses or $15K consultants now run free in ChatGPT or Claude. But democratization makes it easier to build sophisticated models on weak foundations. 🎯 The data maturity threshold Quantitative models are only as good as underlying assumptions and data quality. Running 10,000 iterations on expert estimates without validation creates precise outputs from imprecise inputs. Early-stage companies lack comparable data. For organizations with robust datasets and validated distributions? Sophisticated modeling delivers insight. For those building foundations? Simpler frameworks often outperform. 📊 Methodology follows maturity Most common mistake: deploying risk modeling before establishing processes that make models reliable. Complex models magnify both insight AND error. Great methodology + sophisticated tools = powerful. Weak methodology + sophisticated tools = expensive false confidence. ✅ When quantitative modeling delivers ROI: • Portfolio risk across uncorrelated positions • Material concentration risk threatening survival • Historical data enables validation • Complex interdependencies exceeding cognitive capacity • Regulatory compliance 💡 Better cost-benefit for developing organizations: • Three-scenario analysis with documented assumptions • Sensitivity analysis on 3-5 variables driving 80%+ variance • Actuals-vs-estimates tracking (builds modeling foundation) • Pre-mortem exercises surfacing tail risks 📉 Context matters Median seed-to-Series A: 774 days. That’s 26 months of capital efficiency. 15-person startup investing $15K in modeling before disciplined planning? 2-3 weeks of runway for premature capability. 500-person organization managing portfolio risk? Same investment prevents one seven-figure error. 50x ROI. Same tool. Different maturity. Different value. ❓ Where’s your inflection point? At what scale, data maturity, or risk concentration does sophisticated modeling and software deliver returns over expert frameworks? CFOs, risk leaders: what signaled it was time to upgrade methodology? P.S. Exploring this in Risk University 🎓 DM for access. #RiskManagement #StartupFinance #CRO #QuantitativeAnalysis #RiskAssessment #ModelRisk

#Alhamdulillah #NewPaperAlert Pleased to share our latest #OpenAccess #publication in Frontiers in Environmental Science tackling a persistent challenge of translating #complex socio-ecological data into concrete, #operational #landuse zones. We provide a Dual-Logic #Spatial #Zoning #Model (DLSZM) to overcome the challenge! Key Policy Insights from Our Analysis: -The Power of a Dual Lens: We found that while expert-driven rules provide normative clarity, the machine #learning pathway identified a much larger area (approx. 2,046 km²) needing #restoration. This suggests that data-driven models can capture cumulative stress signals that rigid thresholds might miss, offering a more #intervention-oriented #perspective for #degraded #landscapes. -Convergence & Divergence as Decision Tools: Both methods agreed on vast areas of "Managed Use" (~7,630 km²), indicating stable zones for #sustainable #resource #management. However, the significant divergences in coastal transition zones highlight priority areas where #planning outcomes are most #sensitive to the underlying logic—and thus require closer #governance attention and on-the-ground validation. -Bridging the Science-Policy Gap: Spatial analysis usually stops at producing "#maps." Our DLSZM framework is designed as a transparent #decision-support tool that helps planners answer the crucial question: Given the #ecological value, #human #pressure, and #vulnerability here, what should the #management regime be? This work provides a reproducible #pathway for #coastal regions worldwide to make more informed, adaptive, and conflict-sensitive land-use decisions. Full Text Link: https://lnkd.in/gcZNiRwQ #SpatialPlanning #LandUse #MachineLearning #CoastalManagement #Sustainability #EnvironmentalScience #GIS #Policy

I spoke to the NYT in a story out today about Zillow's Climate Risk scores. While my whole response to their inquiry didn't make it into the quote I thought I'd share the whole thing here: https://lnkd.in/e-6ivGtk With forward-looking models, we can't check their outputs against a record, they are necessarily forecasting events that have not yet happened. This means that the only way to ensure the accuracy of these models is through methodological transparency. In the public sphere, models and methods are peer-reviewed and there is intellectual debate over the "best" results. We don't have this insight when private companies use black-box models. In general, the science of modeling granular, asset-level risk, is very much in progress, especially when we are talking about longer timescales, like decades from now. So it seems especially important that we have expert review of the methods behind these models. People are obviously interested in finding the best or most accurate model, but the important question to ask is less about what model is the best or most accurate model in general, and more about which model is the most appropriate for the particular question or use case you might have. The First Street models might provide a good enough answer for certain questions or certain hazards... it might be a good first approximation of risk from a portfolio perspective for example, for an investor in many homes across the country or region. But what level of accuracy is good enough changes substantially if the question is about one specific property you are about to spend your life savings on. I don't think it's completely a bad thing that Zillow has been displaying climate risk information, it's clearly relevant information that homebuyers should be thinking about. It just needs to be taken with a substantial grain of salt. Hopefully a high climate risk score would prompt a homebuyer to do more digging. The best risk maps may be those produced by your local municipality or state as part of a climate adaptation planning process. Risk projections that are designed for a specific area and that take into account detailed information like drainage infrastructure are likely going to provide better information than an off the shelf national score. Ultimately, I hope we get to a place where the burden is not solely on an individual consumer to try to investigate their climate risk exposure. There needs to be a broader institutional solution to this problem, including getting banks and regulators to do a better job of risk assessment on their end. 

💡 A Practical Guide to Climate Scenarios! Really pleased to have written the forward to this valuable report on the types and applications of climate scenarios by MSCI Inc. and my former United Nations Environment Programme Finance Initiative (UNEP FI) FI colleagues Looking for a handy summary of the types of scenarios from qualitative to quantitative? Here it is: 1. 𝗙𝘂𝗹𝗹𝘆 𝗡𝗮𝗿𝗿𝗮𝘁𝗶𝘃𝗲 𝗦𝗰𝗲𝗻𝗮𝗿𝗶𝗼𝘀 These scenarios are qualitative descriptions of potential climate futures. ✅ Strengths: - Easily customizable - Useful for high-level strategic discussions - Can capture complex risks that are difficult to quantify ⚠️ Limitations: - Subjective and vulnerable to bias - Lack of numerical outputs makes them hard to integrate into risk models 2. 𝗤𝘂𝗮𝗻𝘁𝗶𝗳𝗶𝗲𝗱 𝗡𝗮𝗿𝗿𝗮𝘁𝗶𝘃𝗲 𝗦𝗰𝗲𝗻𝗮𝗿𝗶𝗼𝘀 This type builds on fully narrative scenarios by adding expert-driven quantitative estimates (macroeconomic forecasts, asset class returns, regional physical risks). ✅ Strengths: - Balances qualitative storytelling with numerical data - Allows for scenario comparisons without requiring sophisticated models - Easier to communicate results with clear quantitative insights ⚠️ Limitations: - Can give a false sense of precision if assumptions are weak - Still dependent on subjective expert input, leading to potential biases 3. 𝗠𝗼𝗱𝗲𝗹-𝗗𝗿𝗶𝘃𝗲𝗻 𝗦𝗰𝗲𝗻𝗮𝗿𝗶𝗼𝘀 These scenarios rely on integrated quantitative models to project how climate change and transition risks might evolve under different policy and economic conditions, using macroeconomic models, IAMs, and energy system models. ✅ Strengths: Highly structured and data-driven, reducing subjectivity. Can produce detailed, sector-specific outputs useful for investment decisions. Widely used by regulators and financial institutions for stress testing. ⚠️ Limitations: - Expensive and time-consuming to develop and maintain - “Black box” nature of complex models makes interpretation difficult - Results are only as good as underlying assumptions and data inputs 4. 𝗣𝗿𝗼𝗯𝗮𝗯𝗶𝗹𝗶𝘀𝘁𝗶𝗰 𝗦𝗰𝗲𝗻𝗮𝗿𝗶𝗼𝘀 Probabilistic models go beyond single-scenario forecasting by assigning probabilities, variance, and uncertainty estimates to different climate outcomes. ✅ Strengths: - Models uncertainty, improving risk management - Enables sophisticated stress testing for asset prices, portfolios, and corporate exposure - Valuable for insurance, catastrophe modeling, and financial risk assessments ⚠️ Limitations: - Highly complex and computationally demanding - Requires strong assumptions about uncertainty - Limited research on how climate change affects probability distributions #ClimateFinance #ClimateScenarios #SustainableInvesting #RiskManagement #ScenarioAnalysis #Risk #Finance

The “subjectivity beast” in risk analysis: Are statistical models better than expert opinions?   This post is a matter of the heart. I have heard and read so many (misleading) statements about the superiority of “objective” (i.e., statistical) over “subjective” (i.e., expert opinion-based) risk analysis.   It is a complex topic that deserves much more than a simple post. I can’t cover the complexity and nuances surrounding it. See it as a starting point for a hopefully great discussion.   It is true that, for some risks, data-driven risk analysis and even simple quantitative algorithms regularly outperform experts, as clearly shown by the evidence. There are many reasons for this, like biases at play, an environment where experiences lead to learning, too little experience with certain risks, and many more.   It is not true that statistical models mean “objective risk analysis.” Many decisions remain highly subjective, such as the choice of the statistical model, the choice of the sample, and assumptions about the causality embedded in the model. It is tempting to confuse objectivity with “quantitative” risk analysis and subjectivity with “qualitative risk analysis.” I'm afraid that's not right. Here is why:   Pure quantitative statistical models can also entirely rely on subjective probability and impact distributions assessed by experts. For example, I can conduct a Monte Carlo simulation based on a triangular distribution in which experts guess the worst, best, and most likely scenarios.   Also, statistical analysis results require human interpretation, which might be biased. A statistical model fails to ensure the analysis problems are correctly framed (e.g., risk scenarios that only cover short-term impacts). Statistical analysis starts and ends with subjective decisions. Specifically, in the case of rare risks, expert opinion may outperform statistical analysis just because no data exists. Remember that probability theory cannot be applied to assessing single-event risks that have yet to occur.   Experts may hint at the wrong model assumptions, have some data, and have an educated opinion (the combination may be better than just relying on data). Experts may use scenario analysis to reveal wrongly framed risks. Experts may decompose complex risks by using event tree analysis. Experts may adjust the results of data-driven analysis.   So what does that mean? Two things: First, there is no such thing as objective risk analysis, even if your risk management is fully “quantitative.” It may even lead to the paradox that quantitative risk analysis is more biased as it is believed to be objective. Second, for some risks, the dominant strategy is to rely on expert opinion. For a good reason: Experts may outperform statistical analysis in assessing rare (but detrimental) risks. Institut für Finanzdienstleistungen Zug IFZ Lucerne University of Applied Sciences and Arts #ifzriskmanagement

I have no data for Monte Carlo simulation! Like many people in the world of risk management, an expert once told me something very common: “We don’t have reliable historical data to run quantitative models. So we rely on evaluators’ judgment and on a few past events.” This argument is often repeated and ends up becoming a vicious cycle: “We don’t run simulations because we don’t have data. And we don’t have data because we don’t run simulations.” Here’s my step-by-step response: 1. It’s a valid concern… but incomplete. Yes, it’s true that sometimes we don’t have complete or well-structured databases. But that doesn’t mean data doesn’t exist. The data is in the minds of the experts. They’re not numbers in a table, but experiences, judgments, and intuitions used every time someone rates a risk on a heat map. When someone says a risk is “very likely” or its impact is “severe,” they are using information based on what they’ve seen, lived through, or learned. And that, in itself, is a form of data. 2. There are tools made for exactly this situation. There are probability distributions that don’t require massive databases to work. Some, like the PERT or Triangular distribution, only need the expert to answer: What’s the worst-case scenario? What’s the most likely value? What’s the best-case scenario? With those three values, a distribution can be built that reflects the potential variability of the risk. And the best part: there are well-developed methods to build these distributions rigorously, minimizing bias or judgment errors. ⚠️ Note: Cognitive biases don’t just appear when using distributions—they’re also present when using heat maps. The key is recognizing them and applying tools to reduce their effect. 3. We can improve over time—thanks to Bayes. Bayes’ theorem gives us a solid foundation to refine our estimates as we gather new data. In simple terms: we start with what we know (or think we know), and we improve as we learn more. With every new data point, our estimates become more accurate. 4. A distribution is always better than a fixed cell. When we place a risk in a cell on a heat map, we’re giving it a single value—as if there were no uncertainty or margin for error. But we all know the real world doesn’t work like that. On the other hand, if we use a probability distribution, we give the risk the freedom to vary within a realistic range. We’re not locking it in a rigid box. This way, we represent uncertainty more effectively—and make more informed decisions. Want to take the next step? Contact me. I’ll show you how to move beyond those limiting heat maps and toward a deeper, more quantifiable risk assessment that actually supports your decision-making.

I am excited to share our latest publication in PLOS Climate (thanks to great contributions from Victor Korir, Brenda chepngetich, Victor Villa, Livia Sagliocco, Anna Belli, Leonardo Medina Santa Cruz, Grazia Pacillo and Peter Läderach): 🔗 Understanding local expert perceptions of climate security hotspots using participatory mapping 👉 https://lnkd.in/dMB7sSPt Climate security risks don’t emerge in isolation—they arise from the interaction between climate hazards, socio-economic vulnerabilities, and conflict dynamics. Yet, these interactions are highly context-specific and often poorly captured by purely data-driven approaches. 🧭 In this study, we explore how local experts perceive and map climate security hotspots using participatory mapping. By combining spatial analysis with stakeholder knowledge, we bridge the gap between quantitative models and lived realities. This is a key step towards pursuing context-sensitive adaptation solutions. => Have a read to unpack more findings from the study. #ClimateSecurity #ParticipatoryMapping #ClimateChange #Resilience Ibukun Taiwo