A stress test asks a simple question: if a bad but plausible world arrives, how much capital does the bank lose, and does it still meet its requirements? A climate stress test asks the same question for a world shaped by climate policy, technology and weather. At Meridian Bank the exercise is owned by Enterprise Risk Management (ERM), a team of the Risk function that sits above the individual risk types (credit, market, liquidity, operational) and adds them up into one view of the bank’s resilience.
Where it sits. ERM reports to the Chief Risk Officer. Inside it, the stress testing team runs the bank-wide exercises: the annual internal stress tests used in the Internal Capital Adequacy Assessment Process (ICAAP), the supervisory exercises (the EU-wide stress test coordinated by the European Banking Authority, thematic exercises of the European Central Bank), reverse stress tests and ad hoc scenarios requested by the board. The climate and ESG risk team, also second line, owns the climate risk framework and the materiality assessment; for the climate stress test it supplies the sector heatmaps, the transition and physical risk data and the narrative of the scenarios. The economic research team (often in Finance or in the chief economist’s office) writes the macroeconomic paths. The credit risk methodology team owns the satellite models that turn macro variables into probabilities of default (PD) and loss given default (LGD) (see E01). Capital planning in Finance projects the balance sheet, profits and capital ratios. Model validation reviews every model used. Internal audit, the third line, reviews the whole process every two or three years.
Who the clients are. The internal clients are the board’s risk committee, which approves the scenarios and the results; the asset and liability committee and the capital committee, which use them for capital planning; the business lines, whose sector limits and pricing are affected (see B01, B03, E03). The external client is the supervisor: for a significant institution such as Meridian, the Joint Supervisory Team of the ECB, which reads the ICAAP every year and runs or coordinates the supervisory exercises. Auditors, rating agencies and investors read the published results and the bank’s climate disclosures.
How it “earns” money. It does not, but it decides how much capital the bank must hold and where. Orders of magnitude for Meridian (case assumptions consistent with E02): risk-weighted assets (RWA) of EUR 118bn, CET1 capital of EUR 16.52bn, a CET1 ratio of 14.00%; one basis point of CET1 ratio is EUR 11.8m of capital. The supervisory stress test feeds the Pillar 2 guidance (P2G), a capital recommendation on top of requirements; one extra percentage point of P2G on Meridian’s RWA is EUR 1.18bn of capital that cannot be distributed or lent, which at a 12% hurdle must earn about EUR 140m a year after tax. A climate stress test does not yet set P2G in the EU, but its results enter the qualitative assessment of the supervisory review and, through the ICAAP, the bank’s own capital buffer. A climate stress testing team in a bank of Meridian’s size counts five to fifteen people, plus the time of modellers and data engineers borrowed from other teams.
A realistic week (October, eight weeks before the board). Monday: the stress testing lead chairs the scenario working group with economic research and the ESG risk team; the agenda is the translation of the latest NGFS variables into Meridian’s three internal scenarios. Tuesday: the credit methodology team delivers the sector PD shifts; the climate risk analyst checks them against the bottom-up re-ratings of the twenty largest carbon-intensive names done by the credit analysts (see E01 for Castellan). Wednesday: the physical risk analyst runs the flood module on the commercial real estate (CRE) book with the geocoded collateral file and finds that 8% of addresses could not be matched; she writes the data-quality note. Thursday: capital planning runs the balance-sheet projection; the first CET1 depletion figure of the severe scenario breaches the risk appetite threshold. Friday: a call with the head of corporate coverage, who disputes the oil and gas sensitivities; the stress testing lead keeps them and records the challenge. The draft goes to model validation the following week, then to the risk committee, then to the board.
WHO DOES WHAT — Who signs what: the stress testing lead owns the process and the results pack; the ESG risk head signs the scenario narratives; the head of credit methodology signs the satellite models; model validation issues an opinion; the CRO presents the results; the board risk committee approves the scenarios in spring and the results in autumn; the board approves the ICAAP. Each signature leaves a document that the supervisor can ask for.
IN ONE SENTENCE — A stress test projects the bank’s income, losses, balance sheet and capital over a fixed horizon under a hypothetical adverse scenario, and compares the resulting capital ratio with requirements and with the bank’s own appetite.
INTUITION — A stress test is the bank’s fire drill. Nobody expects the building to burn this year, but you want to know whether the exits are wide enough, whether the alarms ring and who carries the extinguisher. The number at the end (how much capital is lost) matters less than what the drill reveals about where the exits are blocked.
The classic stress test, as run by the EBA every two years with the ECB and the European Systemic Risk Board (ESRB), has five building blocks.
FORMULA
is the exposure at default of portfolio in year ; and are the stressed one-year PD and LGD under scenario ; is the horizon in years; the extra loss compared with the baseline; the tax rate (taken at 25% when the loss is assumed to reduce taxable profit); the increase in RWA from rating migration; and the starting values.
The second formula shows the two ways a stress test hits capital: losses reduce the numerator; rating migration inflates the denominator. For a bank using internal ratings-based (IRB) models, the second effect is often as large as the first, and for a well-rated corporate book it can be larger. A simple simulator makes the first effect concrete.
INTERACTIVE LAB — Three-year loss of a credit book under a stressed PD
See how a PD multiplier, the horizon and LGD drive the stressed loss of a static book, and how much of it is extra compared with the baseline.
| Parameter you can move | Range | Default |
|---|---|---|
| Exposure at default | 1 to 100 EUR bn | 10 EUR bn |
| Baseline one-year PD | 0.05 to 5 % | 0.5 % |
| LGD | 10 to 70 % | 40 % |
| PD multiplier in the stress | 1 to 6 x | 2.5 x |
| Horizon | 1 to 5 years | 3 years |
| Result at default settings | Value |
|---|---|
| Baseline loss over the horizon | 60.0 EUR m |
| Stressed loss over the horizon | 150.0 EUR m |
| Extra loss from the stress | 90.0 EUR m |
| Extra loss in bp of exposure | 90.0 bp |
Try it. Set the horizon to 1 year: the extra loss is a third of the three-year figure. A short horizon is the main reason why climate stress tests over three years produce small numbers.
On a EUR 10bn book with a PD of 0.50% and an LGD of 40%, a static three-year stress that multiplies the PD by 2.5 produces EUR 150m of losses against EUR 60m in the baseline: EUR 90m of extra loss, 90 bp of the exposure.
Three tools are often confused.
A further distinction matters for climate. Stress testing in the strict sense measures capital adequacy over a short horizon (three to five years). Scenario analysis for resilience looks at the business model over ten to thirty years: which sectors shrink, which revenues disappear, whether the strategy holds. The EBA’s guidelines on environmental scenario analysis, discussed in section 3, make this distinction explicit.
A top-down stress test is run by the supervisor (or by the bank’s central team) with its own models applied to data reported by the bank, usually at sector or portfolio level. It is consistent across banks and quick, and it ignores the specificity of each client. A bottom-up stress test is run by the bank with its own models, at counterparty or loan level, following a common methodology and scenario; the supervisor then challenges the results with its own top-down benchmarks (“quality assurance”). The EBA exercise is a constrained bottom-up test. Most internal climate stress tests are hybrids: bottom-up re-rating of the largest and most exposed clients, top-down sector shifts for the rest. Section 5 shows both on the same book.
The stress test is not an end in itself. Under the Capital Requirements Directive, a bank must have an ICAAP, a documented process showing that its capital covers all material risks over a forward-looking horizon, in a baseline and adverse scenarios. The ECB reads the ICAAP and runs the Supervisory Review and Evaluation Process (SREP), which sets the Pillar 2 requirement (P2R) for risks not covered by Pillar 1 and the P2G derived mainly from the EBA stress test. The bank, for its part, uses stress test results in its risk appetite statement: the board sets thresholds (“CET1 depletion in the severe scenario no greater than X bp”), early-warning levels and limits by sector or country. A breach triggers an escalation and a remediation plan. Section 5.8 calibrates such a threshold and the sector limit that follows from it.
PITFALL — Treating the stress-test loss as a forecast. It is a conditional projection: “if this scenario happens, this is the loss”. Whether the scenario is likely is a separate question, which most climate scenarios do not answer: the NGFS does not attach probabilities to its scenarios.
WHY IT MATTERS ON THE DESK — Every number the stress testing team produces travels: into the ICAAP, the risk appetite, the sector limits that coverage bankers must respect, the capital allocated to business lines and therefore their RAROC (see E03). A sensitivity set without care on Tuesday becomes a lending limit in January.
Climate stress tests reuse the machinery of section 2 but break four of its habits.
Physical and transition risks enter the classic risk types through transmission channels: lower corporate profits (carbon costs, stranded assets, damage, business interruption) raise PDs; damaged or devalued collateral raises LGDs; repriced securities create market losses (see E09); deposit outflows or drawn credit lines after a disaster affect liquidity (see C15); and the whole economy may slow, which affects every portfolio (see A01).
The Network for Greening the Financial System (NGFS), a network of central banks and supervisors created in 2017, publishes the climate scenarios that almost every European bank and supervisor starts from. It had 152 members and 24 observers in January 2026; the US Federal Reserve left the network in January 2025.
Long-term scenarios. The NGFS long-term scenarios are produced by three integrated assessment models (GCAM, MESSAGEix-GLOBIOM and REMIND-MAgPIE), which simulate energy systems, land use and emissions, and the NiGEM macroeconomic model, which turns them into GDP, inflation, interest rates and other variables by country, with a horizon to 2100. The current vintage at the time of writing is Phase V, published in November 2024, complemented in November 2025 by explanatory notes on key assumptions, narratives and energy investment, and by an updated guide to climate scenario analysis for central banks and supervisors. Phase V contains seven scenarios in four families:
| Family | Scenario | Idea |
|---|---|---|
| Orderly | Net Zero 2050 | Early, smooth policy; 1.5 °C |
| Orderly | Below 2 °C | Gradual tightening; below 2 °C |
| Orderly | Low Demand | Behaviour change lowers energy demand |
| Disorderly | Delayed Transition | Little action to 2030, then abrupt policy |
| Hot house world | NDCs | Pledges delivered, even if not legislated |
| Hot house world | Current Policies | Only policies in force; about 3 °C |
| Too little, too late | Fragmented World | Late and uneven action; high physical and transition risk |
The main novelty of Phase V was its physical-risk damage function, taken from Kotz, Levermann and Wenz (2024), which estimates much larger GDP losses from warming than earlier vintages; the NGFS itself noted in 2025 that the paper had received academic critiques and that physical estimates would be revisited in Phase VI. The NGFS strategy for 2026-2027 lists the update to Phase VI of the long-term scenarios, with enhanced physical risk assessment, and a second phase of the short-term scenarios. Phase VI was announced for 2026; students should check the NGFS scenarios portal for its release before using Phase V variables in an exercise.
Orders of magnitude. In Net Zero 2050, the global shadow carbon price reaches roughly USD 100 to 250 per tonne by 2030 (in 2010 dollars) according to the NGFS’s own summary, and much more by 2050; transition-driven GDP losses stay within a couple of per cent of the baseline. In Current Policies the shadow price barely moves and all losses come from physical risk, which grows over the century.
Short-term scenarios. In May 2025 the NGFS published its first short-term scenarios, built for stress testing over five years (to 2030), with regional and sector detail and shocks that hit the business cycle. There are four:
| Scenario | Risk | Narrative in brief |
|---|---|---|
| Highway to Paris | Transition | Credible, coordinated policy; carbon revenue recycled into green investment |
| Sudden Wake-up Call | Transition | Abrupt policy shift; confidence and financial shock |
| Disasters and Policy Stagnation | Physical | Compound extreme events in 2026-2027; no new policy |
| Diverging Realities | Both | Advanced economies transition, others suffer disasters |
The NGFS’s summary gives a global GDP loss of about 1.3% relative to baseline by 2030 in Sudden Wake-up Call against about 0.4% in Highway to Paris, and an EU loss of about 1.7% in Diverging Realities. These magnitudes are modest compared with a classic EBA adverse scenario, which typically assumes a cumulative GDP fall several times larger: the NGFS short-term scenarios are designed to be combined with, not to replace, a macro-financial stress.
INTUITION — The NGFS scenarios are like a set of weather forecasts for a sailing trip, each saying “if the wind does this, the sea will do that”. They do not tell you which forecast will come true. The bank’s job is to check that its boat survives each of them, and to know which one it is most exposed to.
PITFALL — Using NGFS variables “off the shelf”. NGFS scenarios are global and regional, at country level at best, and often starting from data two or three years old. A bank must translate them: pick the relevant variables (EU ETS price rather than a global shadow price; national GDP; sector gross value added), recalibrate the starting point, and decide how to add a financial shock. The translation is where most of the judgement, and most of the supervisor’s questions, lie.
ECB 2022 climate risk stress test. Results published on 8 July 2022. 104 significant banks took part in a three-module exercise: a questionnaire on their own climate stress testing capabilities, metrics on their dependence on carbon-intensive sectors, and, for 41 banks, a bottom-up projection under transition scenarios (a three-year disorderly short-term scenario and a 30-year long-term set) and a one-year physical-risk scenario combining a major flood with a drought and heatwave. Credit and market losses for the 41 banks totalled about EUR 70bn, a figure the ECB itself said significantly understated the risk. About 60% of banks had no climate stress testing framework, and nearly two thirds of banks’ income from non-financial corporates came from greenhouse-gas-intensive industries. There was no direct capital effect: the results fed the SREP qualitatively.
Fit-for-55 one-off exercise (the three ESAs and the ECB). Results published on 19 November 2024. The exercise tested whether the EU’s Fit-for-55 package (the 2030 target of a 55% emissions cut) could be implemented without endangering financial stability, across 110 banks, more than 2,300 insurers, more than 600 occupational pension funds and about 22,000 investment funds, over 2022-2030. It used a baseline and two adverse scenarios: a “run on brown” in which investors dump carbon-intensive assets, and the same combined with a classic macro-financial stress. The conclusion: transition risk alone is unlikely to threaten financial stability, but combined with a macroeconomic shock it raises losses sharply; for banks, losses in the combined scenario were about 1.6 times those of the transition-only case (roughly 11% against 7% of the exposures in scope). It was a top-down, system-wide exercise with no bank-level results.
EBA 2027 EU-wide stress test. The EBA consulted early, in June 2026, on a simplified methodology for the 2027 exercise that, for the first time, includes a dedicated climate module covering transition risk and flood risk on corporate and real estate exposures. The final methodology and draft templates were published on 9 November 2026. The adverse macro and climate scenarios, developed by the ESRB and the ECB, are to be released with the launch in January 2027, and results are scheduled for publication in summer 2027. At this stage the climate module does not affect the core results, and therefore not P2G.
Bank of England CBES. The 2021 Climate Biennial Exploratory Scenario tested the largest UK banks and insurers over 30 years under three scenarios (Early Action, Late Action, No Additional Action), with counterparty-level data for the largest corporate borrowers. Results, published on 24 May 2022, showed climate losses equivalent to an average annual drag on profits of about 10 to 15% if firms did not respond, and roughly doubled loan loss rates in Late Action. Its most cited finding was methodological: for the same shared counterparties, the most conservative bank’s loss estimate was on average about ten times the least conservative. In December 2025 the PRA replaced its 2019 supervisory statement with SS5/25 on managing climate-related risks, which leaves firms freedom to choose scenarios proportionate to their exposure and allows narrative scenarios over long horizons.
NOTE — Lessons common to all four exercises: losses over a short horizon are moderate compared with a classic recession stress; data gaps (emissions, geolocation, transition plans) dominate the uncertainty; the dispersion of bank estimates for the same client is very large; and the interaction of climate shocks with macro-financial stress matters more than either alone.
WHY IT MATTERS ON THE DESK — For a junior analyst, the regulation translates into deliverables with deadlines: a scenario analysis covering the business model to at least ten years by the 2027 ICAAP, a climate module in the EBA templates by spring 2027, Pillar 3 tables every half-year, and a plan that the supervisor will compare with all three.
The “deal” of this desk is the annual internal climate stress test, which runs from January to December and feeds the ICAAP submitted to the ECB in the spring of the following year. In a supervisory year (2027 for the EBA exercise) the internal exercise and the EBA climate module share data, models and people, and the calendar is compressed.
INTERACTIVE LAB — The annual climate stress test at Meridian Bank
Follow the internal climate stress test from scenario design to the board decision and the ICAAP.
| # | Who | Does what | Deliverable |
|---|---|---|---|
| 1 | Board risk committee | Approves the stress testing programme and the scope of the climate exercise (January) | Programme memo with scope and calendar |
| 2 | ESG risk team | Updates the materiality assessment and the sector heatmap; selects the portfolios to stress | Materiality assessment, portfolio perimeter |
| 3 | Economic research with ESG risk | Selects NGFS scenarios and translates them into internal scenarios with EU and sector variables | Scenario narratives and variable paths |
| 4 | Board risk committee | Approves the scenarios (March) | Approved scenario book |
| 5 | Data office and physical risk analyst | Extracts exposures, sector codes, emissions and geocoded collateral; runs data-quality checks | Stress test data set and data-quality report |
| 6 | Credit risk methodology | Calibrates satellite models: sector PD shifts, LGD and collateral haircuts | Model documentation |
| 7 | Credit analysts | Re-rate the largest exposed counterparties bottom-up under each scenario | Counterparty scenario ratings |
| 8 | Model validation | Reviews satellite models and the bottom-up overrides | Validation opinion with findings |
| 9 | Stress testing team with capital planning | Runs losses, RWA migration and capital under static and dynamic balance sheets | Results pack, CET1 depletion by scenario |
| 10 | Risk appetite owner and CRO | Compares results with thresholds; proposes limits and management actions | Escalation note and remediation plan |
| 11 | Board | Approves results, limits and the capital add-on (November-December) | Board minutes, updated risk appetite statement |
| 12 | Capital planning | Integrates the results into the ICAAP and the Pillar 3 narrative | ICAAP chapter, disclosures |
Programme and scope (January). The stress testing lead drafts a programme memo: which exercises run this year, with which scenarios, on which perimeter, with which deadlines. For climate, scope is set by the materiality assessment required by the EBA guidelines: Meridian’s assessment concludes that transition risk is material for corporate credit and that flood risk is material for CRE and agricultural lending, so these are stressed quantitatively; market risk on the trading book is covered by sensitivities (see E09); operational and reputational risks are covered qualitatively. The memo is short, but a supervisor will compare it with the final results to check that nothing was quietly dropped.
Materiality and perimeter. The ESG risk team refreshes the sector heatmap, which scores each NACE sector on transition and physical drivers, and maps exposures to it. The perimeter of the exercise is the corporate loan book (EUR 60bn of EAD), the largest single consumer of credit capital. Retail mortgages are covered by a separate energy-efficiency and flood exercise (see B08).
Scenarios. Economic research and ESG risk select the NGFS scenarios, choose the variables that matter for Meridian (EU ETS price, euro area and national GDP, sector gross value added, energy prices, a flood hazard index), re-base them on current data and write narratives. Three choices are documented: which scenario is “severe”, how the climate shock is combined with a macro shock, and the horizon. The board risk committee approves the scenarios in March; changing them later requires a new approval.
Data. This is the longest step. The data office extracts EAD, ratings and sector codes from the risk data warehouse; the climate risk analyst adds emissions (reported or estimated with PCAF data-quality scores, see A03 and E07), transition-plan classifications from the coverage teams, and the geocoded addresses of collateral. Each gap is logged: a client without reported emissions gets a sector proxy; an address that cannot be geocoded gets the worst hazard score of its postcode. The data-quality report travels with the results.
Models. Credit methodology calibrates the satellite model (section 5.2-5.3) and the collateral haircuts (section 5.4). The model documentation states the method, the parameters, the expert judgement and its owner, and the known limitations.
Bottom-up re-ratings. For the twenty to fifty largest exposures in high-risk sectors, credit analysts project the client’s financials under each scenario and propose a scenario rating, as for Castellan in E01. These overrides replace the top-down shift for those names.
Validation. Model validation challenges the satellite model, the overrides and the data proxies. Its findings are graded (high, medium, low); a high finding must be remediated or explicitly accepted by the CRO before results go to the board.
Run and results. The stress testing team runs losses, RWA migration and capital under a static balance sheet (the reference) and a dynamic one (the plan), producing a results pack: CET1 depletion by scenario, contributions by sector and the ten largest contributors by name.
Escalation and decision. If a risk appetite threshold is breached, the CRO proposes actions: a sector limit, a pricing change, a capital add-on, hedges or more data. The board decides and records the decision. The results then go into the ICAAP, the Pillar 3 narrative and, in a supervisory year, the EBA templates.
PITFALL — Running the exercise as a compliance project, with results arriving after the budget and the limits are set for the year. A stress test that cannot change a limit or a price is not used, and supervisors look for exactly that evidence of use (“use test”).
Question. The board risk committee has asked whether Meridian’s corporate loan book could withstand an orderly transition, a disorderly transition and a hot-house world with an acute flood, over the three-year horizon of the ICAAP (2027-2029), and whether the result calls for limits or capital. The stress testing lead, Karim Benali, owns the answer; the ESG risk analyst, Sofia Lindqvist, owns the scenarios and the physical module.
Data (case assumptions). The book has EUR 60bn of EAD. It is split into nine lines: seven sectors stressed top-down, Castellan Cement stressed bottom-up with the scenario ratings of E01, and the rest of the cement and lime book (the forty names of E01, section 5.7). PDs are the average one-year model PDs of each line; LGD is the F-IRB 40% for senior unsecured loans, and 25% for CRE, which is secured (an internal best estimate used for stress losses and, for simplicity, for RWA as well). Maturity is 2.5 years throughout. Meridian’s CET1 capital is EUR 16.52bn and its total RWA EUR 118bn (E02).
| Line | EAD (EUR bn) | PD | LGD |
|---|---|---|---|
| Oil and gas | 4.000 | 0.40% | 40% |
| Power and utilities | 6.000 | 0.25% | 40% |
| Cement and lime excl. Castellan | 0.825 | 0.618% | 40% |
| Castellan Cement (bottom-up) | 0.149 | 0.20% | 40% |
| Steel, chemicals, heavy industry | 3.026 | 0.45% | 40% |
| Automotive and transport | 7.000 | 0.35% | 40% |
| Commercial real estate | 9.000 | 0.50% | 25% |
| Agriculture and food | 3.000 | 0.60% | 40% |
| Other sectors | 27.000 | 0.30% | 40% |
| Total | 60.000 |
The first five lines, EUR 14.0bn, are the “high transition risk” perimeter that A02 used in its simplified example with a single BB grade; here the actual rating mix is used, which is better on average. The PD of the rest of the cement book, 0.618%, is the EAD-weighted average of the two sub-books of E01: EUR 525m at 0.40% and EUR 300m at 1.00%.
Calculation of the baseline. With a static balance sheet, the baseline loss over three years is . The one-year expected loss of the book is EUR 80.66m, so the three-year baseline loss is EUR 241.97m. With the CRR3 F-IRB formula, the book carries EUR 29.47bn of RWA, a density of 49.1%: a quarter of Meridian’s total RWA.
Reading. The baseline is not a forecast of impairments but the reference against which climate losses are measured: everything that follows is “extra” loss compared with a world with no climate shock. Using the through-the-cycle model PDs as the baseline is a choice; supervisory exercises use a projected baseline macro path. The choice is recorded in the methodology note.
Question. Which scenarios, which variables, and how do they reach the sectors?
Data. Sofia Lindqvist’s working group builds three internal scenarios, each anchored on NGFS narratives but expressed in variables that Meridian’s models can use. All values are deviations from the baseline at the end of the horizon (case assumptions, calibrated by expert judgement on the NGFS long-term and short-term scenarios).
| Scenario | NGFS anchor | EU ETS price vs baseline | GDP vs baseline | Flood index |
|---|---|---|---|---|
| S1 Orderly | Net Zero 2050, Highway to Paris | +EUR 50/t | −0.2% | 0 |
| S2 Disorderly | Delayed Transition, Sudden Wake-up Call | +EUR 120/t | −1.5% | 0 |
| S3 Hot house and flood | Current Policies, Disasters and Policy Stagnation | 0 | −1.0% | 1 |
The flood index is 1 for a one-in-100-year river flood hitting two of Meridian’s main lending regions in the first year of the horizon, 0 otherwise. S2 is the severe scenario: a delayed and then abrupt carbon-price rise with a loss of confidence, whose GDP effect is close to the upper end of the NGFS short-term range for the EU.
The satellite model translates these three variables into a shift of each sector’s credit quality, measured as a shift of the distance to default (in standard deviations of asset value). It is linear in the three variables:
FORMULA
is the shift for sector in scenario ; the carbon price increase in EUR per tonne; the GDP deviation in per cent (negative in a downturn); the flood index; , and the sector’s sensitivities to carbon (per EUR 100), to GDP (per point) and to the flood.
The sensitivities come from three sources: the carbon sensitivity from the carbon cost as a share of EBITDA of a representative company in the sector, net of pass-through (the method of E01 applied to sector averages); the GDP sensitivity from Meridian’s existing macro satellite models; the flood sensitivity from the share of the sector’s assets in flood zones and its vulnerability.
INTERACTIVE LAB — From scenario variables to sector shifts in distance to default
Translate a carbon price shock, a GDP shock and a flood into a shift of credit quality for each sector of the book.
| Parameter you can move | Range | Default |
|---|---|---|
| EU ETS price vs baseline | 0 to 250 EUR/t | 120 EUR/t |
| GDP vs baseline | -5 to 1 % | -1.5 % |
| Flood index (0 none, 1 one-in-100-year event) | 0 to 2 | 0 |
| Result at default settings | Value |
|---|---|
| Shift, oil and gas | 0.570 sd |
| Shift, power and utilities | 0.255 sd |
| Shift, cement and lime | 0.690 sd |
| Shift, heavy industry | 0.540 sd |
| Shift, automotive and transport | 0.369 sd |
| Shift, commercial real estate | 0.285 sd |
| Shift, agriculture and food | 0.180 sd |
| Shift, other sectors | 0.192 sd |
Try it. Set the carbon shock to zero and the flood index to 1 (scenario S3): the ranking of sectors inverts. Real estate and agriculture move to the top, cement and oil and gas to the bottom.
Calculation. In S2, cement shifts by standard deviations, oil and gas by 0.570, heavy industry by 0.540, automotive and transport by 0.369, CRE by 0.285, power by 0.255, other sectors by 0.192 and agriculture by 0.180. In S1 the largest shift is cement’s 0.245. In S3 the order inverts: agriculture 0.430, CRE 0.400, automotive 0.200, and 0.150 for oil and gas, power and cement, which only feel the GDP shock and a small flood effect.
Reading. Two features of the table drive everything that follows. First, the carbon sensitivity is concentrated: four sectors with a quarter of the book carry almost all of it. Second, the GDP sensitivity is spread over the whole book, including the EUR 27bn of “other sectors”, which have almost no carbon exposure. A disorderly transition is dangerous for a bank not only through its brown clients but through the recession it may cause.
PITFALL — Reading the power sector’s low carbon sensitivity as a sign that it is safe. Meridian’s utility clients are a mix of renewable producers, which gain from a higher carbon price, and gas and coal generators, which lose; the average hides both. The sector line should be split as soon as data allow, as the EBA templates for 2027 ask for exposures by NACE code at a finer level.
Question. How does a shift in distance to default become a PD, and is the result consistent with the Vasicek approach of E01?
Data. The rest of the cement book: PD 0.618%, shift 0.690 in S2.
FORMULA
is the baseline one-year PD; the default threshold in standard deviations (for 0.618%, ); the shift of section 5.2; the standard normal distribution function; the stressed one-year PD.
INTUITION — Picture each borrower as a walker on a path along a cliff edge; default is a fall. The PD measures how close to the edge the average walker is. A climate shock pushes the whole group sideways by : a group that walked comfortably far from the edge barely notices, a group already close sees many more falls.
INTERACTIVE LAB — Stressed PD for a shift in distance to default
Turn a sector shift into a stressed PD and compare it with the Vasicek conditional PD used in E01.
| Parameter you can move | Range | Default |
|---|---|---|
| Baseline PD | 0.05 to 5 % | 0.618 % |
| Shift in distance to default | 0 to 1.5 sd | 0.69 sd |
| Vasicek sensitivity rho (for comparison) | 5 to 50 % | 20 % |
| Vasicek climate factor X (for comparison) | 0 to 3 sd | 2 sd |
| Result at default settings | Value |
|---|---|
| Stressed PD | 3.502 % |
| Multiple of baseline PD | 5.67 x |
| Vasicek conditional PD at X | 3.617 % |
Try it. Compare the three curves at a shift of 0.7: the PD of the best borrowers is multiplied by about seven, that of the weakest by about five. In relative terms the shock hurts good credits more; in euros, it hurts weak ones more.
Calculation. For the rest of the cement book in S2, , 5.67 times the baseline. The Vasicek conditional PD of E01 at and gives 3.617% for the same average PD. Applied to the E01 sub-book at 0.40%, the shift gives 2.488% against 2.470% for Vasicek.
Result. The cement line’s stressed PD is 3.502% in S2. The two methods agree to within a few per cent for the PDs of this book, because a shift of 0.69 is what a factor realisation of 2 with a sensitivity of 20% produces after rescaling.
Reading. The shift form is preferred for stress testing because it leaves the PD unchanged when the shock is zero, whereas the Vasicek conditional PD at is lower than the average PD (E01 shows 0.065% against 0.20% for BBB), which would make a mild scenario look like an improvement. Roncalli’s notebook (section 13.1.3) presents the family of methods (shocks to the default column of the migration matrix, Vasicek conditioning on a climate factor, a climate add-on to the Markov generator); the shift used here is the simplest member, and it moves only the PD, not the full migration matrix.
Question. How large is the LGD add-on for CRE in S3, where a one-in-100-year river flood hits two lending regions?
Data (case assumptions). CRE EAD EUR 9.0bn with an average loan-to-value (LTV) of 60%, so collateral worth EUR 15.0bn. The geocoding of collateral addresses against flood maps places 20% of collateral value in the flooded area. The depth-damage (vulnerability) curve of the property types gives an average damage of 40% of value at the modelled water depth. Insurance covers 30% of damage on average for this mix of properties, which is consistent with the low insured shares recorded for floods in Europe. A forced sale after default is assumed to recover the remaining collateral value less a 30% haircut.
FORMULA
is collateral value; the share of collateral in the hazard zone (exposure); the damage ratio from the vulnerability curve (hazard intensity times vulnerability); the insured share of damage; the haircut on a forced sale; the exposure of the book. The formula assumes that loans remain collateral-limited, so each euro of uninsured damage, net of the sale haircut, is a euro less recovered.
INTUITION — Physical risk is the product of three things: a hazard (how deep the water gets), an exposure (what stands in the way) and a vulnerability (how much a building suffers per metre of water). Insurance moves part of the bill elsewhere. Roncalli’s chapter 12 builds the same chain, from climate models to hazard maps to damage functions.
INTERACTIVE LAB — Flood damage, insurance and the LGD add-on on the CRE book
Compute uninsured damage to collateral after a flood and the resulting increase in the LGD of the CRE book, and see the role of insurance.
| Parameter you can move | Range | Default |
|---|---|---|
| CRE exposure | 1 to 30 EUR bn | 9 EUR bn |
| Average loan-to-value | 30 to 90 % | 60 % |
| Collateral in the flooded area | 0 to 60 % | 20 % |
| Damage ratio at the modelled depth | 5 to 80 % | 40 % |
| Insured share of damage | 0 to 100 % | 30 % |
| Forced-sale haircut | 0 to 50 % | 30 % |
| Annual probability of the event | 0.1 to 10 % | 1 % |
| Result at default settings | Value |
|---|---|
| Collateral value | 15.00 EUR bn |
| Total damage to collateral | 1.200 EUR bn |
| Uninsured damage | 0.840 EUR bn |
| LGD add-on for the CRE book | 6.53 pp |
| Expected annual uninsured damage | 8.4 EUR m |
Try it. Lower the insured share from 30% to 10%, as happens when insurers withdraw from a flood zone: the LGD add-on rises from 6.53 to 8.40 points. The insurance gap is a credit risk for the bank.
Calculation. Damage is ; uninsured damage is EUR 0.84bn; net of the sale haircut, recoveries fall by , which on EUR 9.0bn of EAD is an LGD add-on of 6.53 percentage points: CRE LGD rises from 25% to 31.53% in S3. Weighted by the 1% annual probability of the event, the expected annual uninsured damage is EUR 8.4m.
Reading. Three remarks. First, the flood also raises PDs (the 0.25 flood sensitivity of CRE in section 5.2), through vacancy and repair costs; the LGD add-on comes on top. Second, the expected annual figure (EUR 8.4m) and the stress figure (EUR 0.84bn of uninsured damage) answer different questions: pricing and provisioning use the first, capital adequacy the second. Third, the insurance gap is large in Europe: the European Environment Agency estimates weather- and climate-related losses in the EU at about EUR 822bn over 1980-2024, of which less than 20% was privately insured (published October 2025). If insurers raise premiums or withdraw, the bank’s LGD rises even if the climate does not change.
PITFALL — Geocoding at postcode level. Flood risk varies within a few hundred metres: a postcode average assigns the riverside warehouse and the hilltop office the same score. Meridian’s data-quality note records the 8% of addresses that could not be matched and assigns them the worst score of their postcode, a conservative but crude choice.
Question. What are the extra losses, the extra RWA and the CET1 depletion in each scenario?
Data. The book of 5.1, the shifts of 5.2, the PD mapping of 5.3, the CRE LGD add-on of 5.4 (applied in S3 only), and Castellan’s scenario PDs from E01: 0.353% in S1 (E01’s orderly path), 1.145% in S2 (disorderly) and 0.379% in S3 (current policies). Static balance sheet over three years: each year, each line keeps its EAD and suffers its stressed PD and LGD. RWA are recomputed with the stressed PDs and LGDs at the end of the horizon. Losses reduce CET1 after 25% tax, the convention used in A02.
INTERACTIVE LAB — The climate stress test of the corporate book, scenario by scenario
Run the three scenarios on the nine lines of the book and compute extra losses, RWA migration and CET1 depletion, with the contribution of each sector.
| Parameter you can move | Range | Default |
|---|---|---|
| Scenario (1 orderly, 2 disorderly, 3 hot house and flood) | 1 / 2 / 3 | 2 |
| Severity of the scenario shocks | 0 to 200 % | 100 % |
| Horizon | 1 to 5 years | 3 years |
| CRE LGD add-on in S3 (from 5.4) | 0 to 20 pp | 6.53 pp |
| CET1 capital | 5 to 30 EUR bn | 16.52 EUR bn |
| Total RWA | 50 to 300 EUR bn | 118 EUR bn |
| Tax rate on losses | 0 to 35 % | 25 % |
| Result at default settings | Value |
|---|---|
| Baseline one-year expected loss | 80.66 EUR m |
| Baseline loss over the horizon | 241.97 EUR m |
| Baseline RWA of the book | 29.47 EUR bn |
| RWA density of the book | 49.1 % |
| Extra loss from the scenario | 357.3 EUR m |
| Extra loss in bp of the book | 59.5 bp |
| RWA migration | 12.63 EUR bn |
| Stressed CET1 ratio | 12.44 % |
| CET1 depletion | 155.9 bp |
| of which losses | 22.7 bp |
| of which RWA migration | 133.1 bp |
| Extra loss, oil and gas | 70.4 EUR m |
| Extra loss, CRE | 40.4 EUR m |
| Extra loss, other sectors | 74.4 EUR m |
| Extra loss, Castellan (bottom-up) | 1.69 EUR m |
| Extra loss, Castellan if stressed top-down | 2.20 EUR m |
| Bottom-up below top-down, Castellan | 23.4 % |
| Stressed loss over baseline loss | 2.48 x |
| After-tax extra loss (ICAAP add-on basis) | 267.96 EUR m |
Try it. Set the horizon to 1 year in S2: the extra loss falls to a third, but the depletion barely moves, because RWA migration does not depend on the horizon. Then set the severity to 0: everything returns to the baseline.
Calculation (S2, the severe scenario). Each line’s stressed PD is computed as in 5.3; the extra loss of a line is . For oil and gas, , so the extra loss is . Summed over the nine lines, the extra loss is EUR 357.29m, 59.5 bp of the book. RWA rise by EUR 12.63bn as PDs migrate (the oil and gas risk weight alone goes from 55.7% to 100.1%). The CET1 ratio falls from 14.00% to 12.44%: a depletion of 155.9 bp, of which 22.7 bp come from losses and 133.1 bp from RWA migration.
Result (all scenarios).
| Scenario | Extra loss (EUR m) | RWA migration (EUR bn) | Stressed CET1 | Depletion (bp) |
|---|---|---|---|---|
| S1 Orderly | 62.63 | 3.07 | 13.61% | 39.4 |
| S2 Disorderly | 357.29 | 12.63 | 12.44% | 155.9 |
| S3 Hot house and flood | 250.17 | 10.15 | 12.74% | 125.5 |
In S3 the largest contributor is CRE, EUR 92.11m of extra loss, of which about EUR 26m come from the flood’s LGD add-on (without it, CRE’s extra loss would be EUR 66.04m); “other sectors” follow with EUR 54.80m.
INTERACTIVE LAB — What depletes CET1: losses or RWA migration?
Split the CET1 depletion of each scenario into the part due to losses and the part due to RWA migration, and test the effect of the output floor.
| Parameter you can move | Range | Default |
|---|---|---|
| Share of IRB RWA migration that reaches the CET1 ratio | 0 to 100 % | 100 % |
| CET1 capital | 5 to 30 EUR bn | 16.52 EUR bn |
| Total RWA | 50 to 300 EUR bn | 118 EUR bn |
| Result at default settings | Value |
|---|---|
| Depletion, S1 | 39.4 bp |
| Depletion, S2 | 155.9 bp |
| Depletion, S3 | 125.5 bp |
| Share of RWA migration in S2 depletion | 85.4 % |
Try it. Set the share of migration to 0%, the situation of a bank whose capital is set by the output floor at the margin: the S2 depletion falls from 155.9 bp to 22.7 bp.
Reading. Four lessons, each of which the results pack spells out for the board.
The validation challenge. Model validation’s opinion on the run, delivered ten days before the risk committee, raises three findings that show where the result is fragile.
Non-linearity. The severity input of the e08-stress block scales all shocks of a scenario. At half the S2 shocks the extra loss is EUR 137.34m and the depletion 79.8 bp; at twice the shocks the extra loss is EUR 1,221.93m and the depletion 310.8 bp, of which 77.7 bp from losses. Doubling the shock multiplies the extra loss by more than three, because the normal distribution function is convex in the tail where these PDs sit. A scenario calibrated “a bit too mildly” understates the loss by much more than “a bit”. Validation asks for this sensitivity table to appear on the first page of the results pack, next to the central figures, and the risk committee adopts it.
Sensitivities set by judgement. The carbon sensitivities of section 5.2 come from sector-average carbon costs and pass-through assumptions; the GDP sensitivities come from satellite models estimated on 2008-2023 data, a period without a carbon shock. Validation grades this a medium finding: the method is reasonable but the parameters cannot be back-tested. The remediation is to cross-check each sector’s sensitivity against the bottom-up re-ratings (the cement shift is consistent with Castellan and with the E01 overlay; the oil and gas and heavy industry shifts have no bottom-up cross-check yet) and against the sector results of the supervisory exercises when the EBA 2027 data become available.
What is not in the model. The run has no second-round effects (defaults of suppliers and customers of the stressed sectors, a fall in collateral values beyond the flood zone, a funding-cost shock), no interaction between S2 and S3 (a disorderly transition in a year of floods), and no change in the behaviour of other banks. Each omission biases the result downwards. Validation does not ask for all of them to be modelled, which would add false precision, but for the results pack to list them and for one combined scenario to be run qualitatively in the narrative.
CASE — The CET1 depletion of S2, 155.9 bp, breaches Meridian’s risk appetite threshold of 150 bp for the severe climate scenario, and S3’s 125.5 bp is just above the early-warning level of 125 bp. Karim Benali writes the escalation note on the Thursday of the week described in section 1.
Question. Does the bottom-up re-rating of Castellan change the result, and what does it teach about the top-down lines?
Calculation. Bottom-up, Castellan’s S2 PD is 1.145% (E01, path 3), and its extra loss over three years is . Top-down, applying the cement shift of 0.690 to Castellan’s 0.20% gives a stressed PD of 1.433% (section 5.3) and an extra loss of EUR 2.21m.
Reading. The bottom-up figure is 23.4% lower, because Castellan’s intensity is below its sector’s and its transition plan is credible enough to deliver most of its 2030 target in the disorderly path. At book level the difference is invisible (half a million euros in EUR 357m), but the point is not the number. Bottom-up re-rating of the largest names does three things that a sector shift cannot: it rewards credible transition plans, it identifies the few large names that drive the tail of each sector, and it connects the stress test to the credit file, where the analyst records triggers and conditions (E01, section 5.8).
The opposite case also exists. For a heavy-industry client whose plan depends on a public subsidy not yet granted, the bottom-up scenario rating may be worse than the sector shift implies. The stress testing team keeps both figures and uses the more conservative one unless the analyst’s projection has been validated.
WHO DOES WHAT — The credit analysts’ bottom-up projections are first-line work used in a second-line exercise. Model validation checks that the analysts did not use the client’s own scenario (often a single, optimistic one) and that overrides of the sector shift are documented with the client data that justify them.
Question. Meridian’s strategic plan already reduces its exposure to oil and gas, the rest of the cement book and heavy industry. What happens to the S2 result if the plan is taken into account?
Data. Dynamic balance sheet: the three high-carbon lines (EUR 7.851bn of EAD, Castellan and power excluded) run down by 15% a year, and the freed exposure is redeployed into “other sectors”. Each year’s extra loss uses that year’s exposures; RWA are recomputed at the end of 2029.
INTERACTIVE LAB — Static versus dynamic balance sheet in the disorderly scenario
Compare year-by-year extra losses and the CET1 depletion when high-carbon lines run down and the exposure is redeployed into other sectors.
| Parameter you can move | Range | Default |
|---|---|---|
| Annual run-down of high-carbon lines | 0 to 50 % | 15 % |
| Risk appetite threshold, severe scenario | 50 to 300 bp | 150 bp |
| CET1 capital | 5 to 30 EUR bn | 16.52 EUR bn |
| Total RWA | 50 to 300 EUR bn | 118 EUR bn |
| Tax rate on losses | 0 to 35 % | 25 % |
| Result at default settings | Value |
|---|---|
| Extra loss, static balance sheet | 357.3 EUR m |
| Extra loss, dynamic balance sheet | 321.9 EUR m |
| Loss avoided by the run-down | 35.4 EUR m |
| CET1 depletion, static | 155.9 bp |
| CET1 depletion, dynamic | 142.5 bp |
| High-carbon exposure at end-2029 | 4.82 EUR bn |
| Run-down needed to meet the threshold | 6.05 % a year |
| High-carbon exposure consistent with the threshold | 6.51 EUR bn |
| Extra loss per year, static | 119.1 EUR m |
| Extra loss in 2027, dynamic | 112.6 EUR m |
| Extra loss in 2029, dynamic | 102.3 EUR m |
Try it. Lower the threshold to 125 bp: the run-down needed jumps to about 48% a year. Beyond a point, cutting high-carbon lending cannot fix a result driven by the recession channel on the rest of the book.
Calculation. With a 15% annual run-down, high-carbon exposure falls from EUR 7.851bn to EUR 4.82bn by end-2029. The extra loss falls from EUR 119.1m a year to EUR 112.6m in 2027 and EUR 102.3m in 2029: EUR 321.9m over three years instead of EUR 357.3m, a saving of EUR 35.4m. The CET1 depletion falls from 155.9 bp to 142.5 bp, inside the 150 bp threshold.
Reading. The dynamic result is useful for strategy and dangerous for capital adequacy. Useful, because it shows the board what its own plan buys. Dangerous, because it assumes that the bank can execute the run-down in the middle of a disorderly transition, when every other bank tries to do the same and the borrowers most in need of refinancing are those it wants to leave. This is why the EBA uses a static balance sheet, why the ECB’s 2026 reverse stress test judged some management actions (portfolio sales under stress) overly optimistic, and why Meridian’s results pack reports the static figure as the reference and the dynamic one as a management view.
PITFALL — Counting the run-down twice: once as a reduction of stress losses and again as a reduction of financed emissions. Both are real effects of the same decision, but a run-down of loans to a client that keeps emitting changes Meridian’s portfolio, not the atmosphere. E07 and E12 discuss why supervisors and investors increasingly ask for both numbers with that caveat.
Question. What does the board decide?
Data. Meridian’s risk appetite statement contains two climate metrics, approved in 2025: CET1 depletion in the severe climate scenario no greater than 150 bp (threshold), with an early-warning level at 125 bp; and a cap on high-carbon exposure, currently set at EUR 9bn and never binding.
Calculation. The dynamic block of 5.7 solves for the run-down that brings S2 exactly to the threshold: 6.05% a year, which means a high-carbon exposure of EUR 6.51bn at end-2029 instead of EUR 7.85bn today. If the board wanted S2 at the early-warning level of 125 bp, the run-down would need to be 48.25% a year, leaving EUR 1.09bn: an exit from three sectors in three years.
Reading and decision. The CRO’s proposal, which the board approves in December, has five parts.
WHY IT MATTERS ON THE DESK — This is the “use test” the supervisor looks for: the stress test changed a limit, a capital number, an origination rule and a data plan, and the minutes say why the board did not go further.
| Stage | Key figure | Value |
|---|---|---|
| 5.1 | Book EAD; one-year EL; 3-year baseline loss | EUR 60bn; EUR 80.66m; EUR 241.97m |
| 5.1 | Book RWA; density | EUR 29.47bn; 49.1% |
| 5.2 | S2 shifts: cement, O&G, heavy industry | 0.690; 0.570; 0.540 sd |
| 5.2 | S3 shifts: agriculture, CRE | 0.430; 0.400 sd |
| 5.3 | Cement stressed PD in S2; multiple | 3.502%; 5.67x |
| 5.3 | Vasicek benchmark (X = 2, rho 20%) | 3.617% |
| 5.4 | Flood: uninsured damage; CRE LGD add-on | EUR 0.84bn; 6.53 pp |
| 5.4 | Expected annual uninsured damage | EUR 8.4m |
| 5.5 | Extra loss S1 / S2 / S3 | EUR 62.63m; 357.29m; 250.17m |
| 5.5 | RWA migration S1 / S2 / S3 | EUR 3.07bn; 12.63bn; 10.15bn |
| 5.5 | CET1 depletion S1 / S2 / S3 | 39.4; 155.9; 125.5 bp |
| 5.5 | S2 depletion: losses / RWA migration | 22.7 bp; 133.1 bp (85.4%) |
| 5.6 | Castellan S2 extra loss: bottom-up / top-down | EUR 1.69m; 2.21m |
| 5.7 | S2 dynamic (15% run-down): loss; depletion | EUR 321.9m; 142.5 bp |
| 5.8 | Run-down for 150 bp; high-carbon EAD 2029 | 6.05% a year; EUR 6.51bn |
| 5.8 | ICAAP climate add-on (after tax) | EUR 268m |
Climate stress testing does not create a Pillar 1 capital charge. Under CRR3, Meridian’s corporate exposures are weighted with the F-IRB formula: own PD, regulatory LGD (40% senior unsecured; lower values for eligible collateral), regulatory conversion factors and an effective maturity of 2.5 years. Climate enters Pillar 1 only through the PD the rating system assigns (E01) and, for secured exposures, through the collateral values. The stress test shows what would happen to Pillar 1 if those inputs moved: the EUR 12.63bn of RWA migration in S2 is the capital the bank would need if every rating moved as the scenario implies.
Pillar 2 is where climate capital lives today. The ECB can set a P2R add-on for climate risk management shortcomings found in the SREP, and the bank’s ICAAP must show that internal capital covers material climate risks; Meridian’s EUR 268m add-on of section 5.8 is such an internal measure. The P2G is derived from the EBA stress test, whose 2027 climate module will not, at this stage, affect it. The output floor (72.5% of standardised RWA from 2030, phased in before) limits the capital effect of rating migration for floor-bound banks, as the e08-compare block shows: a bank bound by the floor sees its marginal capital follow the standardised weights (75% to 100% for corporate exposures rated BBB to BB by an agency, 100% for unrated ones), which move little with internal downgrades. The systemic risk buffer could in principle be used by national authorities against climate-related systemic risk; it has been discussed by the ESRB but not applied for climate in the euro area.
The stress test is not a deal, but its result is priced into deals through capital allocation. The question for the oil and gas book: what does the allocation of the ICAAP climate add-on do to its RAROC, and what margin would restore the hurdle?
Data (case assumptions). Oil and gas book: drawn EUR 3.2bn and undrawn commitments EUR 2.0bn, so EAD EUR 4.0bn with the 40% conversion factor; average PD 0.40%; drawn margin over FTP 120 bp; commitment fee 35 bp; ancillary revenue (hedging, bond issuance, cash management) EUR 40m a year carrying EUR 2.5 of RWA per euro. Costs: 40% of lending revenue, 60% of ancillary revenue, and the 40 bp carry cost of HQLA held against 10% of undrawn commitments (the LCR outflow rate for committed credit facilities to non-financial corporates). Climate add-on: the after-tax extra loss of the book in S2 over three years, allocated in full.
INTERACTIVE LAB — RAROC of the oil and gas book with the climate capital add-on
Compute the RAROC of a sector book line by line, add the allocated ICAAP climate add-on, and find the margin that would restore the hurdle.
| Parameter you can move | Range | Default |
|---|---|---|
| Drawn amount | 0.5 to 10 EUR bn | 3.2 EUR bn |
| Undrawn commitments | 0 to 5 EUR bn | 2.0 EUR bn |
| Drawn margin over FTP | 50 to 300 bp | 120 bp |
| Commitment fee | 0 to 100 bp | 35 bp |
| Ancillary revenue | 0 to 100 EUR m | 40 EUR m |
| Baseline PD | 0.05 to 3 % | 0.4 % |
| Stressed PD in S2 | 0.05 to 10 % | 1.867 % |
| Stress horizon | 1 to 5 years | 3 years |
| Share of the after-tax stress loss allocated | 0 to 100 % | 100 % |
| Also allocate RWA migration (1 yes, 0 no) | 0 / 1 | 0 |
| Result at default settings | Value |
|---|---|
| Revenues | 85.40 EUR m |
| Operating costs incl. HQLA carry | 42.96 EUR m |
| Expected loss | 6.40 EUR m |
| Risk weight | 55.75 % |
| Risk weight at the stressed PD | 100.12 % |
| RWA incl. ancillary | 2,330.0 EUR m |
| Allocated capital (13% of RWA) | 302.90 EUR m |
| RAROC before the add-on | 8.92 % |
| Extra loss in S2 over the horizon | 70.42 EUR m |
| Climate capital add-on | 52.81 EUR m |
| RAROC with the add-on | 7.60 % |
| Margin for 12% without the add-on | 184.7 bp |
| Margin for 12% with the add-on | 228.7 bp |
Try it. Set the ancillary revenue to 0: the lending-only RAROC collapses. The relationship is profitable only through its cross-sell, which is exactly what a sector limit puts at risk.
Line by line. Lending revenue is ; with EUR 40m of ancillary revenue, revenues are EUR 85.40m. Costs are . Expected loss is . At PD 0.40% the risk weight is 55.75%, so RWA are EUR 2,330.0m including EUR 100m for the ancillary business, and allocated capital is EUR 302.90m. RAROC , below the 12% hurdle before any climate consideration; the drawn margin would need to be 184.7 bp to reach it.
With the climate add-on. The book’s S2 extra loss over three years is EUR 70.42m (the e08-stress figure); after tax, EUR 52.81m of capital is added. RAROC falls to 7.60% and the margin needed for the hurdle rises to 228.7 bp. If the RWA migration were also allocated, the add-on would be EUR 283.56m and RAROC 4.61%.
Reading. No investment-grade oil and gas client will pay 229 bp. The add-on does not lead to repricing; it changes which clients Meridian chooses, how long it lends (shorter tenors shrink the share of the transition the loan lives through), and how much ancillary business it requires from them. It also makes visible a fact the coverage team already knew: this book earns its hurdle only with ancillary revenue. Allocating only the loss part, not the migration, is a judgement that Meridian’s capital committee made and recorded; a bank allocating the full stress would effectively price itself out of the sector.
The stress test of this module is a solvency exercise, but climate scenarios have liquidity channels that the ILAAP (the liquidity counterpart of the ICAAP) must consider. After a flood, corporate and SME clients draw their credit lines to fund repairs and bridge insurance payments, and households in the affected area withdraw deposits; the EBA’s environmental scenario guidelines ask banks to assess such effects where material. In a disorderly transition, wholesale funding spreads of banks perceived as exposed to brown sectors could widen. Meridian’s ILAAP runs a sensitivity: a 20% extra drawdown of undrawn lines to agriculture and CRE clients in the flooded regions, which consumes part of the liquidity buffer but leaves the LCR well above 100%. On the FTP side, Treasury does not charge a climate premium on funding today (see C15); the climate cost of a loan enters through capital allocation, as in 6.2, not through funding.
Stress-test losses are not accounting provisions. IFRS 9 expected credit losses are probability-weighted over a range of forward-looking scenarios, measured over 12 months or the lifetime of the loan depending on its stage, and must be unbiased. A stress scenario is deliberately severe and is not a probability-weighted forecast. The two meet in two places. First, the scenarios of the stress test inform the climate overlays of the ECL model: E01 shows how Meridian weights a disorderly scenario at 20% in its sector overlay for cement. Second, the auditors and the supervisor compare the two: if the stress test shows that a sector’s PD would quintuple in a plausible scenario while the ECL model contains no climate effect at all, they ask why. IAS 1 and IFRS 7 require disclosure of significant judgements and of the sensitivity of ECL to assumptions; banks increasingly disclose ECL under alternative scenarios that include climate (see E05).
The stress test uses and feeds the bank’s sustainability reporting.
PITFALL — Publishing a stress-test number without its assumptions. A “climate loss of EUR 357m” in a sustainability report invites comparison with peers who used different scenarios, horizons and balance-sheet assumptions. The CBES showed tenfold dispersions for the same counterparties; disclosures should state horizon, scenario, static or dynamic, and what is not covered.
The risks of a climate stress test are model, data and governance risks: a wrong number that looks precise, or a right number that nobody uses. A specific form of greenwashing exists too: presenting a stress test as evidence of resilience, or of alignment, that it cannot support.
| Risk | Control | Owner |
|---|---|---|
| Scenario too mild to bite | Benchmark against NGFS short-term and EBA scenarios; reverse stress test | ESG risk, stress testing |
| Satellite sensitivities set by judgement | Documented calibration, challenge by validation, sensitivity table | Credit methodology, validation |
| Sector averages hide outliers | Bottom-up re-rating of the largest names; split mixed sectors | Credit analysts |
| Geocoding gaps in physical module | Match-rate KPI; conservative proxies; remediation plan | Data office, physical risk analyst |
| Insurance cover assumed, not verified | Check policies at origination and renewal in hazard zones | Credit operations |
| Optimistic management actions | Static balance sheet as reference; dynamic as a separate view | Stress testing lead |
| RWA migration ignored or double counted | Reconcile with capital planning; show floor-bound view | Capital planning |
| Results not used | Link to risk appetite thresholds and limits; board minutes | CRO, board risk committee |
| Inconsistent disclosures | Reconcile stress perimeter with Pillar 3 and CSRD | Finance, sustainability reporting |
Red-flag checklist for a reviewer (supervisor, auditor, validator or analyst).
Scenarios (as of November 2026).
Supervisory exercises.
Regulation. EBA Guidelines on the management of ESG risks apply from 11 January 2026 (a year later for small and non-complex institutions); EBA Guidelines on environmental scenario analysis apply from 1 January 2027; CRD VI requires ESG stress testing and a horizon of at least ten years in risk management.
Orders of magnitude.
What changed recently. Climate moved from stand-alone supervisory exercises (2022, 2024) into the regular EU-wide stress test (2027); guidelines turned expectations into rules with application dates; short-term scenarios gave stress testers a five-year tool; and physical-risk damage functions became much more severe, then contested.
Open debates.
INTERACTIVE LAB — Climate stress testing: check your understanding
Test the key concepts, regulatory facts and calculations of the module.
Q1. Which vintage of NGFS long-term scenarios was the latest confirmed release available to Meridian’s team in autumn 2026?
Answer: C. Phase V was published in November 2024 with explanatory notes in November 2025; Phase VI was announced for 2026.
Q2. What distinguishes the NGFS short-term scenarios published in May 2025?
Answer: A. The four short-term scenarios run to 2030 and are designed for stress testing; they complement the long-term set.
Q3. In a static balance-sheet stress test, what happens to a loan that matures during the horizon?
Answer: B. The static assumption freezes the balance sheet to make banks comparable and exclude convenient management actions.
Q4. In Meridian’s severe scenario, which component drives most of the CET1 depletion?
Answer: C. 133.1 bp of the 155.9 bp depletion, 85.4%, comes from RWA migration; losses account for 22.7 bp.
Q5. Why does the module prefer a shift of distance to default to the Vasicek conditional PD for stress testing?
Answer: B. The Vasicek conditional PD at X = 0 is below the average PD, so mild scenarios could look like improvements.
Q6. A flood damages 40% of the value of collateral in the flooded zone. What reduces the bank’s LGD effect most directly?
Answer: B. Insurance transfers part of the damage; uninsured damage, net of the sale haircut, is lost recovery.
Q7. What did the ECB’s 2022 climate stress test conclude about capital?
Answer: B. The 2022 exercise was a learning exercise; about EUR 70bn of losses for 41 banks, judged an underestimate, with qualitative SREP use.
Q8. What is new in the EBA 2027 EU-wide stress test?
Answer: A. A climate module on transition and flood risk for corporate and real estate exposures, without effect on the core results at this stage.
Q9. What was the most cited methodological finding of the Bank of England’s 2021 CBES?
Answer: B. For shared counterparties, the most conservative estimate was on average about ten times the least conservative.
Q10. Meridian’s bottom-up stressed loss on Castellan is lower than the top-down figure. Why?
Answer: B. Bottom-up projections reward a credible transition plan; the sector shift treats Castellan as an average producer.
Q11. Why does the module treat the dynamic balance-sheet result as a management view, not the reference?
Answer: B. Exiting sectors during a disorderly transition is uncertain; supervisors judge such management actions sceptically.
Q12. From which date do the EBA Guidelines on environmental scenario analysis apply?
Answer: C. The final report was published on 5 November 2025, with application from 1 January 2027.
Q13. What is the main reason a stress-test loss is not an IFRS 9 provision?
Answer: B. IFRS 9 requires an unbiased, probability-weighted estimate; a stress scenario is deliberately severe.
Q14. Lowering Meridian’s threshold from 150 bp to 125 bp would require a run-down of about 48% a year of high-carbon lines. What does this show?
Answer: B. Most of the depletion comes from the GDP channel and RWA migration across the whole book, which sector exits do not remove.
A EUR 2bn steel book has a PD of 0.80% and an LGD of 40% (F-IRB, M = 2.5). A scenario raises the EU ETS price by EUR 80 per tonne and lowers GDP by 1%. The sector’s carbon sensitivity is 0.30 per EUR 100 and its GDP sensitivity 0.12 per point. Compute the shift, the stressed PD, the extra loss over three years with a static balance sheet, and the change in RWA.
Solution. Shift: . Threshold: ; stressed PD , about 2.5 times the baseline. Extra loss: . RWA: the risk weight at 0.80% gives EUR 1,509.6m; at 2.024%, EUR 2,048.6m; migration of EUR 539.0m, which at 13% is EUR 70m of allocated capital, more than twice the after-tax loss of EUR 22.0m.
A bank holds EUR 4bn of CRE loans at an average LTV of 50%. A flood map places 15% of collateral value in a one-in-100-year flood zone; the damage ratio at the modelled depth is 30%; insurance covers 50% of damage; forced sales recover remaining value less 25%. Compute the uninsured damage, the LGD add-on and the expected annual uninsured damage. What happens if insurance cover falls to 20%?
Solution. Collateral: . Damage: . Uninsured: . LGD add-on: percentage points. Expected annual uninsured damage: . With 20% insurance, uninsured damage is EUR 0.288bn and the add-on 5.4 points: the insurance gap raises the LGD effect by 60%.
A bank has CET1 of EUR 8bn and RWA of EUR 60bn. A climate scenario produces EUR 200m of extra credit losses over the horizon and EUR 4bn of RWA migration. Tax is 25%. Compute the stressed CET1 ratio and the depletion, split into losses and migration. The bank’s threshold is 100 bp: what is the simplest diagnosis?
Solution. Starting ratio: . After-tax loss: EUR 150m. Loss-only ratio: , so losses deplete 25.0 bp. Full stressed ratio: ; total depletion 106.8 bp, of which 81.8 bp from migration. The threshold is breached by 6.8 bp, and three quarters of the depletion comes from the denominator: the remedy lies in the concentration of rating-sensitive exposures or in how the floor applies, not in provisions.
1. What is the difference between climate scenario analysis and a climate stress test? A stress test measures capital adequacy over a short horizon, typically three to five years, under a severe but plausible scenario, and compares the result with requirements and appetite. Scenario analysis explores the resilience of the business model over ten to thirty years under several plausible futures, without necessarily computing a capital ratio. The EBA’s environmental scenario guidelines formalise the distinction. A bank needs both: the first for the ICAAP, the second for strategy and the transition plan.
2. Walk me through how you would translate an NGFS scenario into PDs for a corporate book. I would pick the variables that matter for the book (EU carbon price, national GDP, sector value added, energy prices), re-base them on current data, and write the narrative. A satellite model links those variables to each sector’s credit quality, for example as a shift in distance to default calibrated on carbon cost as a share of EBITDA and on historical GDP sensitivities. The shift maps into a stressed PD by sector. For the largest exposed names I would replace the sector shift by a bottom-up re-rating with projected financials.
3. Why are climate stress test losses often small, and should we worry? Short horizons, NGFS GDP effects of one or two per cent, linear models and the absence of second-round effects all push losses down. The ECB said its 2022 figure of EUR 70bn significantly understated the risk. Small numbers should prompt questions about what the model excludes rather than reassurance: compounding physical events, insurance withdrawal, financial amplification, and the combination with a classic recession, which the Fit-for-55 exercise showed matters most.
4. How do RWA migration and the output floor interact in a climate stress test? For an IRB bank, stressed PDs raise risk weights, often more than losses reduce capital, especially for a well-rated book. In Meridian’s case 85% of the severe-scenario depletion comes from migration. If the bank is bound by the output floor, its marginal capital follows standardised risk weights, which barely move with internal downgrades, so the migration effect is muted. A good results pack shows both views.
5. A business head says the stress test penalises clients who are transitioning. How do you answer? A sector shift does treat all clients as average, which is why the largest names are re-rated bottom-up: a credible transition plan lowers the scenario rating, as Castellan’s case shows. The answer is to bring data, not to remove the client from the stress test. I would also point out that the add-on changes client selection and tenor more than price, and that transition finance with conditions tied to milestones can improve the result.
6. What would you look for when validating a physical risk module? The geocoding match rate and how unmatched addresses are treated; the source and resolution of the hazard maps and whether they reflect future climate or only history; the vulnerability curves by property type; the treatment of insurance and whether it is verified or assumed; and whether the module captures chronic risks and business interruption, not only damage. I would also check that results are reconciled with the Pillar 3 physical risk tables.
7. Should climate stress test results feed Pillar 2 guidance? Arguments for: the risks are real, and capital that reflects them gives banks an incentive to manage them. Arguments against: the models and data are immature, results vary enormously between banks for the same client, and capital charges could raise the cost of transition finance. The EBA has kept the 2027 climate module out of P2G for now. My view is that qualitative use and internal ICAAP add-ons are appropriate until the models converge.
8. How would you use a reverse stress test for climate risk? I would fix an outcome, for example a 300 bp CET1 depletion or a breach of the combined buffer, and ask which combination of carbon price, recession, flood and insurance withdrawal would produce it. The answer shows how far the plausible scenarios are from trouble and which assumption the result is most sensitive to. It avoids the trap of attaching probabilities to climate scenarios and gives the board a concrete narrative to judge.
Scenarios
Supervisory exercises
Regulation and guidelines
Data and orders of magnitude
Academic and textbook references
Other modules: A01, A02 (regulatory stack and materiality), A03 (carbon data), B01 (transition-plan scorecard), B08 (real estate), C15 (treasury), E01 (Castellan re-rating, Vasicek overlay), E02 (capital, output floor), E03 (RAROC), E05 (IFRS 9), E07 (financed emissions), E09 (market risk), E12 (transition plan).