The Internal Capital Adequacy Assessment Process (ICAAP) asks a simple question: does the bank hold enough capital to survive its risks, including risks not fully captured by Pillar 1 minimums, under adverse but plausible conditions? Credit risk usually dominates the answer. This article sets out how to build the credit-risk stress-testing spine of an ICAAP without either understating the stress or producing numbers regulators cannot follow.
The regulatory anchors are the EBA guidelines on stress testing (EBA/GL/2018/04), the EBA ICAAP guidelines (EBA/GL/2016/10) and — for EU banks — the biennial EBA stress-testing exercise methodology, which sets a de facto benchmark for credit-risk projection quality.
The three-layer architecture
An ICAAP credit-risk stress test operates in three layers.
- Macro layer: a coherent set of macro-financial paths (GDP, unemployment, house prices, corporate spreads, FX, equity, oil) over a three-year horizon.
- Satellite layer: models translating macro variables into portfolio-level credit risk parameters (PDs, LGDs, migration matrices) by segment.
- Balance-sheet layer: the resulting expected losses, RWA changes and capital ratio projections, integrated with the P&L and the funding plan.
Each layer has to be internally consistent and consistent with its neighbours. The most common ICAAP weakness is a macro layer that looks severe on paper — GDP down x%, unemployment up y% — combined with a satellite layer whose sensitivities produce mild credit-loss increases. Regulators recognise this pattern immediately.
Macro scenario design
A defensible scenario set includes at least:
- Baseline: the current best estimate, aligned with the plan.
- Adverse: a severe but plausible downturn — typically calibrated to a 1-in-25 to 1-in-50 year macro path over three years.
- Severe adverse: a deeper tail, calibrated to a 1-in-100 year path or aligned with an EBA-style scenario.
The severity anchor matters. A scenario that resembles the mildest downturns of the last thirty years is not adverse; a scenario that resembles the 2008–09 or 2020 shocks in one variable but not others is internally inconsistent. Cross-variable coherence — for example, unemployment paths consistent with the GDP contraction, or house-price paths consistent with the unemployment shock — is where scenario design earns its keep.
For methodology on transitioning between climate and macro scenarios, see NGFS Climate Scenario Guide and Cross-Asset Stress Testing.
Satellite models: linking macro to credit parameters
Satellite models are the technical core. For each material segment (retail mortgages, consumer credit, SME, mid-corporates, large corporates, financial institutions, sovereigns) the model links macro drivers to segment PD and LGD.
A typical PD satellite has the form:
logit(PD_seg,t) = α_seg + β_seg,1 · GDP_t + β_seg,2 · Unemployment_t + β_seg,3 · HPI_t + …
with lags chosen from the data (usually 1–4 quarters) and coefficients estimated on 15–25 years of history where available. The Vasicek-style scalar approach — mapping macro through a systemic factor Z to shift the PIT PDs — is a defensible alternative when direct macro data is sparse (see IFRS 9 PD/LGD/EAD for the connection).
LGD satellites typically link stressed LGD to house-price paths (for mortgages), commercial-property paths (for CRE), unemployment (for consumer) and corporate spreads (for corporates). Downturn LGD in the ICAAP is scenario-conditional; a fixed downturn add-on applied to all scenarios understates the baseline separation.
Migration matrices and grade-level dynamics
Rating migration is where much of the RWA impact under IRB emerges. A stressed migration matrix moves exposures to lower grades — increasing default probability, increasing RWA and, if severe enough, triggering IFRS 9 Stage 2 transfers and lifetime ECL. Two approaches dominate:
- Direct estimation of scenario-conditional migration matrices from historical stressed periods.
- Model-implied migration from the same PD satellite, using the credit-cycle factor Z to shift transition probabilities.
Under either approach, the stressed migration matrix should reconcile with the stressed portfolio default rate — the two are the same reality expressed at different granularities.
Concentration risk
Portfolio credit models with only granular exposure understate concentration risk. Under Pillar 2, banks are expected to hold capital against name, sector and geographic concentrations. Stress-test overlays include:
- Name concentration: replay the default of the top N single-name exposures under adverse conditions, computing the incremental capital impact.
- Sector concentration: apply sector-specific shocks (e.g. commercial real estate, energy, sovereign) with cross-sector correlation calibrated to historical downturns.
- Geographic concentration: for banks with cross-border activity, differential country stresses that respect currency and sovereign exposure.
The HHI on drawn exposure is a useful summary but not a substitute for the concentration stress overlay.
Balance-sheet integration
Credit stress-test outputs feed:
- Impairment / provision charges through the ECL calculation with the adverse scenario weighted at 100% for stress purposes.
- RWA changes through migration, defaulted exposures and IRB parameter updates (subject to the regulatory approach).
- Capital ratios through the retained-earnings impact and the RWA impact.
- Funding and liquidity impacts (see Liquidity Risk) — stress downgrades feed higher funding costs and can trigger contingent outflows.
The full projection over three years should show CET1, Tier 1 and total capital ratios by quarter, against internal risk-appetite limits and regulatory minimums including buffers.
Numerical intuition
For a mid-sized universal bank with two-thirds retail mortgages, one-quarter corporate and the balance in consumer:
- A severe adverse scenario (GDP contraction of 4–5% over three years, unemployment up 3–4 percentage points, house prices down 15–20%) typically produces a three-year cumulative credit-loss increase of 200–400 bps of RWA relative to baseline.
- Migration-driven RWA growth typically adds 15–25% to Pillar 1 credit RWA in the same scenario.
- Together, the CET1 ratio impact is often 300–500 bps over the three-year horizon.
Ranges depend heavily on portfolio mix; the point is that a well-calibrated credit stress produces material capital movement, not decoration.
Validation and independent challenge
Validators focus on:
- Scenario coherence: are macro variables and their lag structures internally consistent, and severity plausibly aligned with the target return period?
- Satellite calibration: are coefficients estimated on sufficient historical downturns, and are elasticities within the range implied by peer benchmarks?
- Migration reasonableness: do stressed migration matrices produce cumulative default rates aligned with the satellite PD projections?
- Overlay treatment: are concentration overlays and management overlays documented, quantified and sunset-dated?
See The Model Validation Lifecycle for the general validation framework.
Governance implications
Boards should see:
- The three-year projection of capital ratios by scenario, with the drivers of change decomposed (net interest income, credit losses, RWA migration, other).
- Sensitivity to key satellite parameters (unemployment elasticity, house-price elasticity).
- The point at which each management action in the recovery plan would be triggered.
Recovery-planning inputs from the stress test — dividend suspension trigger, capital-raising trigger, asset-disposal trigger — should feed directly into the recovery plan and the resolution planning dialogue.
Limitations
Every stress test is conditioned on a specific set of scenarios. Reverse stress testing — asking what scenarios make the bank unable to meet minimum requirements — is the honest complement, and increasingly a supervisory expectation. Institutions that treat the ICAAP stress as prediction miss the point; those that treat it as insurance against a class of futures use it well.
Conclusion
Credit-risk stress testing for ICAAP works when the macro layer is severe and coherent, the satellite layer maps macro to credit parameters with defensible elasticities, and the balance-sheet integration produces movements that are visible in the ratios. The banks that do this well tend to have fewer surprises when the real downturn arrives — because their internal view of the tail has been calibrated in advance.
References and further reading
- European Banking Authority, Guidelines on institutions' stress testing (EBA/GL/2018/04).
- European Banking Authority, Guidelines on ICAAP and ILAAP information (EBA/GL/2016/10).
- Basel Committee on Banking Supervision, Stress testing principles, BCBS 450.
- European Central Bank, ECB Guide to the internal capital adequacy assessment process (ICAAP).
About the author
Part of an ongoing series on capital and credit-risk modelling — more about the author.