Output Schemas¶

This page documents the schemas for calculation results and output data produced by each calculator and the final aggregated output.

Source of truth: All schemas are defined in src/rwa_calc/data/schemas.py. Result bundles are defined in src/rwa_calc/contracts/bundles.py.

SA Result Schema¶

Results from Standardised Approach calculation. Each exposure gets a risk weight looked up from regulatory tables by exposure class and CQS, then RWA = EAD × risk weight.

Source: SA_RESULT_SCHEMA in data/schemas.py

IRB Result Schema¶

Results from IRB approach calculation (F-IRB and A-IRB). Includes the full IRB formula breakdown: PD → LGD → correlation → capital requirement K → maturity adjustment → RWA.

Source: IRB_RESULT_SCHEMA in data/schemas.py

Slotting Result Schema¶

Results from slotting approach calculation for specialised lending. Risk weights are assigned by supervisory category, lending type, and (under CRR) remaining maturity.

Source: SLOTTING_RESULT_SCHEMA in data/schemas.py

Calculation Output Schema¶

The master output schema covering all exposure-level results with full audit trail. Designed so that users can investigate why results occurred and replicate calculations from the output data alone.

Source: CALCULATION_OUTPUT_SCHEMA in data/schemas.py

Identification and lineage¶

Counterparty hierarchy (rating inheritance)¶

Lending group hierarchy (retail threshold)¶

Exposure hierarchy (facility structure)¶

CRM inheritance (from hierarchy)¶

Exposure classification¶

Original exposure values¶

CCF application (off-balance sheet conversion)¶

CRM — collateral impact¶

CRM — life insurance collateral (Art. 232)¶

Captures life-insurance-policy collateral recognised under the Other Funded Credit Protection Method (PS1/26 Art. 232 / CRR Art. 232). Unlike financial collateral, life-insurance policies do not reduce EAD on the SA side — they map the secured portion to a preferential risk weight via the Art. 232(3) insurer-RW table:

The SA calculator then blends the mapped RW across the secured / unsecured portions of the exposure (no Art. 222 floor applies, unlike FCSM). On the IRB side, life-insurance collateral feeds the LGD waterfall with LGD_S = 40% (Art. 232(2)(b)) — the columns below describe the SA-side mechanic only.

References: PRA PS1/26 Art. 232 (Other Funded Credit Protection Method, including the new 7-tier mapping with 30% / 65% / 135% bands); CRR Art. 232 (4-tier 20% / 50% / 100% / 150% mapping); Art. 200(b) (eligibility); Art. 212(2) (operational requirements). Source: engine/crm/life_insurance.py::compute_life_insurance_columns.

CRM — guarantee impact (substitution approach)¶

Pre/post CRM counterparty tracking¶

CRM — provision impact¶

CRM — currency mismatch (Basel 3.1 only)¶

Captures the unhedged-FX retail uplift for residential mortgages and qualifying retail exposures whose servicing currency differs from the borrower's primary income currency.

Reference: PRA PS1/26 Art. 123B / BCBS CRE20.93 — applies a 1.5× risk-weight multiplier (capped at 150%) to the otherwise-applicable risk weight when there is no natural or financial hedge in place. Under CRR there is no equivalent retail FX uplift.

EAD calculation¶

Risk weight determination — SA¶

Risk weight determination — IRB¶

Specialised lending and equity¶

Real estate specific¶

Final RWA calculation¶

Post-model adjustments (Basel 3.1 only)¶

Captures the IRB post-model adjustment (PMA) overlays required by PRA PS1/26 to compensate for known IRB model deficiencies. Sequencing is regulatorily prescribed: the mortgage RW floor (Art. 154(4A)(b)) is applied first to establish the post-floor RWEA base, then general PMA scalars (Art. 146(3) / Art. 153(5A) / Art. 154(4A)(a)) and unrecognised-exposure adjustments are applied on top of the floored amount.

References: PRA PS1/26 Art. 146(3) (PMA root obligation), Art. 153(5A) (corporate / institution / sovereign sequencing), Art. 154(4A) (retail sequencing and 10% mortgage RW floor), Art. 158(6A) (EL monotonicity — PMA can only increase EL, never decrease it).

Double default (CRR only)¶

CRR Art. 153(3) provides a double-default treatment for IRB exposures with eligible unfunded credit protection (Art. 202 — eligible protection providers; Art. 217-218 — operational requirements). Under PRA PS1/26 Art. 153(3), Art. 202, and Art. 217 are all "[Note: Provision left blank]" — the double-default treatment has been removed and these columns will be False / 0 / null for Basel 3.1 runs.

References: CRR Art. 153(3) (RWEA formula with double-default scalar), Art. 202 (eligible protection providers), Art. 217-218 (operational requirements).

Expected loss (IRB)¶

EL is compared against Pool B of Art. 159 — provisions plus additional value adjustments (AVA, Art. 34) plus other own-funds reductions. The Pool B formula determines whether IRB exposures generate a CET1 deduction (el_shortfall, Art. 36(1)(d)) or a T2 credit (el_excess, capped at 0.6% of IRB credit-risk RWA, Art. 62(d)).

References: CRR Art. 159(1)(a) (provisions), Art. 159(1)(c) / Art. 34 (AVA), Art. 159(1)(d) (other own-funds reductions). When ava_amount or other_own_funds_reductions are absent from the input bundle, they default to zero so that pre-Pool-B implementations behave the same as before.

Supporting factors (CRR only)¶

The supporting-factor columns capture both the SME (CRR Art. 501) and infrastructure (CRR Art. 501a) RWA reductions. Under Basel 3.1 / PRA PS1/26 these factors are not available and the columns are populated for schema parity only (supporting_factor = 1.0, supporting_factor_applied = False). The *_applied flag is generic — it is True whenever any qualifying factor was applied, regardless of whether the binding factor was SME or infrastructure.

Warnings and validation¶

Framework-Specific Output Additions¶

CRR (Basel 3.0) additions¶

Source: CRR_OUTPUT_SCHEMA_ADDITIONS in data/schemas.py

Basel 3.1 (PRA PS1/26) additions¶

Source: BASEL31_OUTPUT_SCHEMA_ADDITIONS in data/schemas.py

Result Bundles¶

AggregatedResultBundle¶

The final result from the output aggregator. Combines SA, IRB, slotting, and equity results with output floor application and supporting factor adjustments.

Source: AggregatedResultBundle in contracts/bundles.py

@dataclass(frozen=True)
class AggregatedResultBundle:
    """Final aggregated output from the output aggregator."""

    results: pl.LazyFrame                               # Final RWA results
    sa_results: pl.LazyFrame | None = None               # Original SA results
    irb_results: pl.LazyFrame | None = None              # Original IRB results
    slotting_results: pl.LazyFrame | None = None         # Original slotting results
    equity_results: pl.LazyFrame | None = None           # Equity results
    floor_impact: pl.LazyFrame | None = None             # Output floor impact analysis
    supporting_factor_impact: pl.LazyFrame | None = None # Supporting factor impact (CRR)
    summary_by_class: pl.LazyFrame | None = None         # RWA by exposure class
    summary_by_approach: pl.LazyFrame | None = None      # RWA by approach
    pre_crm_summary: pl.LazyFrame | None = None          # Gross view by original class
    post_crm_detailed: pl.LazyFrame | None = None        # Split rows for guarantees
    post_crm_summary: pl.LazyFrame | None = None         # Net view by effective class
    el_summary: ELPortfolioSummary | None = None          # EL summary with T2 credit cap
    errors: list = field(default_factory=list)            # All pipeline errors

ELPortfolioSummary¶

Portfolio-level expected loss summary with T2 credit cap (CRR Art. 62(d)).

@dataclass(frozen=True)
class ELPortfolioSummary:
    total_expected_loss: float        # Sum of EL across IRB exposures
    total_provisions_allocated: float # Sum of provisions allocated to IRB
    total_el_shortfall: float         # Sum of max(0, EL − provisions) per exposure
    total_el_excess: float            # Sum of max(0, provisions − EL) per exposure
    total_irb_rwa: float              # Total IRB RWA (denominator for T2 cap)
    t2_credit_cap: float              # 0.6% of total IRB RWA
    t2_credit: float                  # min(total_el_excess, t2_credit_cap)
    cet1_deduction: float             # 100% of total_el_shortfall (Art. 36(1)(d))
    t2_deduction: float               # Always zero (no T2 deduction for shortfall)

Other result bundles¶

Export Formats¶

Parquet¶

result.results.collect().write_parquet("rwa_results.parquet")

Full precision, efficient compression, schema preservation.

CSV¶

result.results.collect().write_csv("rwa_results.csv")

Human-readable, Excel-compatible.

Excel (via API export module)¶

from rwa_calc.api.export import ResultExporter

exporter = ResultExporter(result)
exporter.to_excel("rwa_results.xlsx")

Next Steps¶

• Regulatory Tables
• API Reference
• Configuration Guide