METHODOLOGY

IFRS 9 Classification Methodology

This document describes LoanStage's ECL calculation engine, SICR detection logic, and audit trail architecture. It is intended for credit risk managers, internal auditors, and regulators reviewing the platform's methodology.

1. Regulatory framework

LoanStage implements the IFRS 9 Financial Instruments impairment model as adopted by the EU. The classification engine follows IFRS 9 §5.5 (Recognition of expected credit losses), §B5.5.17 (SICR indicators), and §B5.5.37 (Default presumption). ECL methodology is aligned with EBA Guidelines on ECL estimation (EBA/GL/2017/06).

2. Three-stage impairment model

Stage 1 (Performing): No significant increase in credit risk since origination. 12-month ECL recognised. PD = annualised lifetime PD × 15% factor. Stage 2 (SICR): Significant increase in credit risk detected. Lifetime ECL recognised. PD = lifetime PD. Stage 3 (Credit-impaired): Objective evidence of impairment (default). Lifetime ECL recognised. Interest revenue calculated on net carrying amount.

3. SICR detection logic

LoanStage checks the following SICR indicators at each classification: • DPD > 30 days: Rebuttable presumption per §B5.5.28 • DPD ≥ 90 days: Default presumption per §B5.5.37 (Stage 3) • Borrower rating downgrade ≥ 2 notches: §B5.5.17(c) • Borrower rating downgrade + DPD: Combined indicator per §B5.5.17 • LTV > 90%: Collateral coverage insufficient per §B5.5.17(g) • Negative sector outlook + weak rating: Forward-looking SICR per §B5.5.16 • Borrower rating E or F: Near-default / credit-impaired Every triggered indicator is logged with the specific IFRS 9 paragraph reference.

4. ECL calculation formula (illustrative defaults)

ECL = PD × LGD × EAD Stage 1: ECL = PD_lifetime × 15% × LGD × EAD (12-month) Stage 2/3: ECL = PD_lifetime × LGD × EAD (Lifetime) PD parameters by loan type: • Mortgage: 0.8% (A) to 35% (F) • Corporate: 1.0% (A) to 45% (F) • SME: 1.2% (A) to 45% (E/F) • Consumer: 1.5% (A) to 45% (E/F) • Trade Finance: 0.5% (A) to 45% (E/F) • Leasing: 0.8% (A) to 45% (E/F)

5. LGD and collateral treatment (institution-configurable)

LGD is determined by collateral type and LTV ratio: • Real estate collateral: Base LGD 25–35%. Haircut 20%. Net collateral = gross × 80%. Dynamic LGD reduced based on LTV coverage. • Equipment / machinery: Haircut 40–50%. • Receivables: Haircut 30–40%. • Unsecured: LGD = loan type default (45–65%). Adjusted LTV = Outstanding balance / Net collateral value. If adjusted LTV > 1.0: collateral uncovered, LGD not reduced.

6. Classification method

LoanStage uses a hybrid classification approach: • Deterministic rules: All loans are first assessed against the SICR rule set. If any rule triggers, the stage is assigned deterministically with the triggering rule logged. • AI-assisted: For edge cases where deterministic rules do not clearly classify (borderline DPD, partial collateral coverage with mixed indicators), an analyst assistant provides a non-binding recommendation for human review. The recommendation is logged in the audit trail. The final classification decision remains with the analyst. AI output is never final authority.

7. Audit trail architecture

Every classification event generates an immutable audit record containing: • loan_id and portfolio_id • Input data used: amount, DPD, loan_type, borrower_rating, previous_rating, collateral_value, sector_outlook • SICR indicators triggered (with IFRS 9 §B5.5.17 references) • Stage result and ECL result • ECL formula applied • Classification method (deterministic / AI-assisted) • Timestamp (UTC) Audit records are append-only. They append-only — tamper-evident and traceable. Full audit log is exportable as CSV.

8. Macro scenario engine

LoanStage includes a macro scenario overlay engine. Sensitivity factors by loan type are applied to base ECL for three scenarios: • Base scenario: example base scenario aligned with ECB-style macroeconomic assumptions • Adverse scenario: GDP -2%, unemployment +3%, rates +150bps • Severely adverse scenario: GDP -5%, unemployment +6%, rates +300bps Macro-adjusted ECL = Base ECL × scenario coefficient. Coefficients are configurable per institution.

9. Data security and governance

• Data stored in Supabase EU region (Frankfurt) • AES-256 encryption at rest • TLS 1.3 in transit • Tenant isolation: portfolio data is never shared between users • Data retention: active while account is open; deleted within 30 days of cancellation • GDPR compliant: full data export and deletion on request • DPA available on request

9b. Sample audit trail record

The following is an example of a single audit trail record generated by LoanStage: loan_id: LN-2024-0001 days_past_due: 0 | borrower_rating: F | previous_rating: A collateral_value: EUR 1,224,285 | collateral_type: Real Estate SICR indicators triggered: • Rating downgrade A to F (5 notch) — IFRS 9 §B5.5.17(c) • High LTV 122% — IFRS 9 §B5.5.17(g) Stage: 2 | ECL: EUR 19,828.30 Formula: PD(6.00%) x LGD(22.09%) x EAD(EUR 1,496,023) = EUR 19,828.30 Method: Deterministic | Timestamp: 2026-05-10T20:06:09Z (UTC) This record is immutable and append-only — tamper-evident and traceable.

10. Input data specification

Required CSV columns: • loan_id (string): Unique loan identifier • amount (number): Outstanding principal balance in EUR • days_past_due (integer): Days since last missed payment • loan_type (string): mortgage / term_loan / revolving_credit / trade_finance / leasing • borrower_rating (string): A / B / C / D / E / F • previous_rating (string): Prior period rating for downgrade detection • collateral_value (number): Gross collateral value in EUR • collateral_type (string): real_estate / equipment / receivables / inventory / none • sector_outlook (string): positive / neutral / negative • maturity_date (date): YYYY-MM-DD format Download demo CSV template: loanstage.app/demo_portfolio.csv

Questions about methodology?

Contact our credit risk team for technical documentation, DPA requests, or methodology review calls.