Ifrs9 Scenarios

IFRS 9 SCENARIO FRAMEWORK REQUIREMENTS

IFRS 9.5.5.17 requires: reasonable and supportable information about future economic conditions, including forward-looking information. This is not optional. Single-scenario ECL is non-compliant with IFRS 9.

SCENARIO STRUCTURE — MINIMUM AND BEST PRACTICE

Minimum (IFRS 9): base + 1 upside + 1 adverse Best practice: 4–5 scenarios with explicit probability weights

| Scenario | Typical Weight | Key Feature | | -------- | -------------- | -------------------------------------------- | | Upside | 10–20% | Above-trend growth, falling unemployment | | Base | 35–50% | Central forecast, moderate conditions | | Adverse | 25–35% | Mild recession, rising unemployment | | Severe | 10–20% | Deep recession, sharply falling asset prices |

Weights must: sum to 1.0; reflect management's genuine probability assessment; be documented and approved by the IFRS 9 Governance Committee. Equal weighting (25% each) is RARELY defensible and will be challenged by auditors.

KEY MACROECONOMIC VARIABLES BY ASSET CLASS

Retail mortgages: House Price Index (HPI), unemployment rate, base rate Consumer loans: Unemployment rate, disposable income index, base rate SME loans: GDP growth, SME default index, unemployment rate Corporate loans: GDP growth, corporate default rates, sector-specific indices Commercial Real Estate: CRE capital value index, vacancy rates, GDP growth

CREDIT CYCLE ADJUSTMENT (CCA) — CONVERTING TTC TO PIT PD

PIT PD = TTC PD x CCA CCA is estimated from a satellite model. Typical satellite model form: ln(CCA) = a + b1(GDPgrowth) + b2(Unemployment) + b3(HPIgrowth) + e

Example CCA values: Severe recession: CCA = 1.8–2.5 (PDs 80–150% above long-run average) Adverse: CCA = 1.2–1.5 Base: CCA ~ 1.0 (by definition — TTC PD already reflects long-run average) Upside: CCA = 0.7–0.9 (PDs below long-run average)

SATELLITE MODEL DETAIL

Model Structure

The satellite model links macroeconomic variables to credit risk parameters. Typical specification for a UK mortgage portfolio:

ln(Default Ratet) = a + b1 Unemploymentt + b2 HPIgrowtht + b3 BaseRatet + b4 ln(Default Rate{t-1}) + et

Key requirements for the satellite model:

• Estimated on a sufficiently long time series (minimum 1 full cycle, ideally 2+)
• Must include at least one recession period in calibration data
• Coefficients must have economically intuitive signs (e.g., higher unemployment

increases default rates)

• Out-of-sample validation required (typically holdout the most recent 2–3 years)
• R-squared typically 0.6–0.85 for well-specified models

Model Validation Requirements

• Annual independent validation by Model Risk Management
• Backtesting: compare predicted vs actual default rates over rolling windows
• Sensitivity testing: how much does ECL change for a 1pp change in each variable?
• Benchmarking: compare against external provider models (Moody's, Oxford Economics)

WEIGHTED ECL CALCULATION

Step 1: Calculate PIT PD for each scenario using scenario-specific CCA Step 2: Calculate ECL for each scenario: ECLs = PDPITs x LGD x EAD (Stage 1) or ECLs = Sumt [PDmarginalts x LGDt x EADt x DFt] (Stage 2/3) Step 3: Weighted ECL = Sums (Weights x ECLs)

NON-LINEAR EFFECTS

Weighted ECL != ECL at weighted-average PD (due to non-linearity in ECL formula). Always calculate ECL for each scenario separately, then probability-weight the results. The difference between these approaches (non-linear adjustment) is material for portfolios with high LGD or long remaining maturities.

Non-Linearity Worked Example

Portfolio: 1,000M gross carrying amount, LGD = 40%

| Scenario | Weight | PIT PD | ECL (PD x LGD x EAD) | | -------- | ------ | ------ | -------------------- | | Upside | 15% | 0.8% | 3.2M | | Base | 40% | 1.5% | 6.0M | | Adverse | 30% | 3.0% | 12.0M | | Severe | 15% | 6.0% | 24.0M |

Correct: Weighted ECL = 0.15 x 3.2 + 0.40 x 6.0 + 0.30 x 12.0 + 0.15 x 24.0 = 10.08M Wrong: Weighted PD = 0.15 x 0.8 + 0.40 x 1.5 + 0.30 x 3.0 + 0.15 x 6.0 = 2.52% ECL at weighted PD = 2.52% x 40% x 1,000 = 10.08M (linear case — same)

For lifetime ECL with compounding and discounting, the non-linear effect becomes material (typically 5–15% higher ECL when correctly scenario-weighted).

SCENARIO EXPLAINABILITY FOR GOVERNANCE COMMITTEE

For each quarterly scenario update, prepare:

1. Scenario name and narrative description
2. Key macroeconomic variables for each scenario (3-year forward path)
3. Scenario weights and rationale for any weight changes from prior quarter
4. ECL under each scenario individually
5. Probability-weighted ECL (reported figure)
6. Sensitivity: ECL if severe scenario were weighted 100% (IFRS 7 required)
7. Changes from prior quarter: which scenarios/weights/variables changed and why

FORWARD-LOOKING HORIZON

Explicit forecast horizon: typically 2–5 years (period with supportable forecasts) Mean reversion: beyond explicit horizon, variables revert to long-run average over a reversion period (typically 2–5 additional years) Perpetuity: beyond reversion period, variables held at long-run average

GOVERNANCE OF SCENARIOS

Scenarios must be:

• Approved by the IFRS 9 Governance Committee before each calculation
• Sourced from a credible economic forecaster (internal Economics team or

external provider: Oxford Economics, Moody's Analytics, Bloomberg)

• Documented in detail in the IFRS 9 Governance Pack
• Disclosed (key variables and weights) in the IFRS 7 notes

Governance Calendar — Typical Quarterly Cycle

| Week | Activity | Owner | | ------- | ---------------------------------------------------- | --------------------- | | Week 1 | Economics team produces draft scenarios | Chief Economist | | Week 2 | Model Risk reviews satellite model outputs | Model Risk Management | | Week 3 | IFRS 9 Governance Committee reviews and approves | CRO / CFO | | Week 4 | ECL calculation run with approved scenarios | Finance / Credit Risk | | Week 4+ | Results reviewed, PMAs assessed, disclosures drafted | Finance |

OUTPUT FORMAT — SCENARIO SUMMARY

IFRS 9 SCENARIO SUMMARY
Reporting Date:     [YYYY-MM-DD]
Entity:             [Bank / Group name]
Approved By:        [IFRS 9 Governance Committee, date]

SCENARIO DEFINITIONS:
| Scenario | Weight | GDP (Y1/Y2/Y3) | Unemployment (Y1/Y2/Y3) | HPI (Y1/Y2/Y3) |
|----------|--------|-----------------|--------------------------|-----------------|
| Upside   | [%]    | [%/%/%]         | [%/%/%]                  | [%/%/%]         |
| Base     | [%]    | [%/%/%]         | [%/%/%]                  | [%/%/%]         |
| Adverse  | [%]    | [%/%/%]         | [%/%/%]                  | [%/%/%]         |
| Severe   | [%]    | [%/%/%]         | [%/%/%]                  | [%/%/%]         |

ECL BY SCENARIO:
| Scenario | ECL (M) | vs Prior Quarter |
|----------|---------|------------------|
| Upside   | [X]     | [+/- Y]          |
| Base     | [X]     | [+/- Y]          |
| Adverse  | [X]     | [+/- Y]          |
| Severe   | [X]     | [+/- Y]          |
| Weighted | [X]     | [+/- Y]          |

SENSITIVITY (IFRS 7.35G):
  100% Severe scenario ECL:   [Amount] ([+X%] vs reported)
  100% Upside scenario ECL:   [Amount] ([-X%] vs reported)

NEVER DO THESE

• NEVER use a single macroeconomic scenario for IFRS 9 ECL — this is non-compliant with IFRS 9.5.5.17 and will result in a qualified audit opinion
• NEVER apply equal scenario weights (25% each) without documented justification — auditors treat equal weighting as a rebuttable presumption of inadequate governance
• NEVER calculate ECL at the weighted-average PD instead of weighting scenario-level ECLs — the non-linear effect is material for lifetime ECL portfolios
• NEVER use a satellite model calibrated on data that excludes a recession period — the model will systematically underestimate adverse and severe scenario PDs
• NEVER change scenario weights between quarters without documented rationale approved by the Governance Committee — unexplained weight changes are a common audit finding

ALL OUTPUTS REQUIRE REVIEW BY A QUALIFIED PROFESSIONAL BEFORE USE IN REGULATORY FILINGS OR BUSINESS DECISIONS.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: panaversity
• Source: panaversity/agentfactory-business-plugins
• License: Apache-2.0
• Homepage: https://agentfactory.panaversity.org/docs/Business-Domain-Agent-Workflows

Install and usage instructions live in the source repository linked above.