Exposure Classification¶

This document provides comprehensive documentation of the exposure classification system used to determine regulatory exposure classes and calculation approaches.

Overview¶

The classifier (src/rwa_calc/engine/classifier.py) is responsible for:

1. Mapping counterparty entity types to regulatory exposure classes
2. Determining the calculation approach (SA, F-IRB, A-IRB, Slotting)
3. Applying SME and retail classification rules
4. Identifying defaulted exposures
5. Determining FI scalar eligibility for IRB correlation adjustment

Entity Type: The Single Source of Truth¶

The entity_type field on counterparties is the authoritative source for exposure class determination. This design eliminates ambiguity from overlapping boolean flags and ensures consistent classification.

Valid Entity Types¶

The system supports 18 entity types organised into logical groups:

VALID_ENTITY_TYPES = {
    # Sovereign class
    "sovereign",
    "central_bank",

    # RGLA class (Regional Governments/Local Authorities)
    "rgla_sovereign",      # Has taxing powers or government guarantee
    "rgla_institution",    # No sovereign equivalence

    # PSE class (Public Sector Entities)
    "pse_sovereign",       # Government guaranteed
    "pse_institution",     # Commercial PSE

    # MDB/International org class
    "mdb",
    "international_org",

    # Institution class
    "institution",
    "bank",
    "ccp",
    "financial_institution",

    # Corporate class
    "corporate",
    "company",

    # Retail class
    "individual",
    "retail",

    # Specialised lending
    "specialised_lending",

    # Equity
    "equity",
}

Dual Exposure Class Mapping¶

Each entity type maps to both an SA exposure class and an IRB exposure class. These can differ for certain entity types based on regulatory requirements.

SA Exposure Class Mapping¶

Used for SA risk weight table lookups:

IRB Exposure Class Mapping¶

Used for IRB formula selection:

Why Classes Can Differ¶

The SA and IRB exposure classes differ for RGLA, PSE, and MDB entity types because:

• SA treatment: Uses specific risk weight tables for RGLA, PSE, and MDB
• IRB treatment: Uses the underlying IRB formula (central govt/central bank or institution) based on the nature of the entity's credit support

For example, a government-guaranteed PSE (pse_sovereign) uses the PSE risk weight table under SA but the central govt/central bank IRB formula because its credit risk is backed by the government.

Classification Pipeline¶

The ExposureClassifier.classify() method processes exposures through a defined sequence of steps:

Step 1: Add Counterparty Attributes¶

_add_counterparty_attributes(exposures, counterparties)

Joins exposure data with counterparty attributes needed for classification: - entity_type - Single source of truth - annual_revenue - For SME check - total_assets - For large FSE threshold - default_status - For default identification - apply_fi_scalar - For FI scalar determination - is_managed_as_retail - For SME retail treatment

Step 2: Derive Independent Flags¶

_derive_independent_flags(exposures, config, schema_names)

Derives all independent classification flags in a single .with_columns() call for LazyFrame plan optimisation. This batch covers exposure class mapping, default identification, infrastructure classification, and FI scalar classification.

Exposure class mapping — maps entity_type to exposure classes using the constant mappings:

# Result columns:
exposure_class_sa   # SA class for risk weight lookup
exposure_class_irb  # IRB class for formula selection
exposure_class      # Unified class (SA class for backwards compatibility)

Default identification — checks default_status flag: - Sets is_defaulted = True - Sets exposure_class_for_sa = DEFAULTED (SA treatment) - IRB exposures keep their class but use default LGD

Infrastructure classification — identifies infrastructure exposures per CRR Art. 501a: - Checks product_type for "INFRASTRUCTURE" pattern - Sets is_infrastructure = True - Eligible for 0.75 supporting factor under CRR (not Basel 3.1)

FI scalar classification — determines FI scalar eligibility per CRR Art. 153(2):

• requires_fi_scalar — derived directly from the user-supplied apply_fi_scalar flag on counterparties
• Effect: 1.25x multiplier on IRB asset correlation

Slotting enrichment — derives slotting metadata from patterns in reference fields:

Slotting Category (from counterparty_reference): - *_STRONG* → strong - *_GOOD* → good - *_SATISFACTORY* → satisfactory - *_WEAK* → weak - *_DEFAULT* → default

Specialised Lending Type (from product_type): - *PROJECT* → project_finance - *OBJECT* → object_finance - *COMMOD* → commodities_finance - IPRE → ipre - HVCRE → hvcre

HVCRE Flag: - is_hvcre = True if sl_type == "hvcre"

Step 3: SME and Retail Classification¶

_classify_sme_and_retail(exposures, config)

Applies SME and retail classification in a single .with_columns() call.

SME criteria per CRR Art. 501: - Entity must be classified as CORPORATE - annual_revenue < EUR 50m (converted to GBP using config FX rate) - Revenue must be > 0 (excludes missing data)

If criteria met: - Sets is_sme = True - Updates exposure_class to CORPORATE_SME

Retail eligibility per CRR Art. 123:

1. Mortgage detection: Identifies mortgages via product_type pattern matching
2. Threshold check: Aggregated exposure to lending group < EUR 1m
3. Residential exclusion: Residential property collateral excluded from threshold (CRR Art. 123(c))

Step 4: Corporate to Retail Reclassification¶

_reclassify_corporate_to_retail(exposures, config, schema_names)

Reclassifies exposures based on retail threshold outcomes: - Mortgages to individuals → RETAIL_MORTGAGE - Retail exceeding threshold + SME revenue → CORPORATE_SME - Retail exceeding threshold + no SME criteria → CORPORATE - Retail within threshold → remains RETAIL_OTHER

Step 5: Resolve Model Permissions¶

_resolve_model_permissions(exposures, model_permissions)

When permission_mode=PermissionMode.IRB and model_permissions data is provided, resolves per-exposure IRB permissions:

1. Joins exposures to model_permissions via model_id (propagated from internal rating via rating inheritance)
2. Filters by exposure_class match
3. Applies geography filter (country_codes) and book code exclusions
4. Sets model_airb_permitted, model_firb_permitted, and model_slotting_permitted boolean columns
5. Exposures without model_id get all flags set to False (fall back to SA)

When model permissions are active, Step 6 uses per-row model_airb_permitted / model_firb_permitted / model_slotting_permitted flags exclusively. There is no org-wide fallback — exposures without a matching model permission use SA.

See Input Schemas — Model Permissions for the data schema.

Step 6: Determine Approach and Finalize¶

_determine_approach_and_finalize(exposures, config, has_model_permissions)

Assigns calculation approach and builds classification audit trail in a single .with_columns() call.

When permission_mode=PermissionMode.IRB and model permissions are present, per-row model_airb_permitted / model_firb_permitted / model_slotting_permitted flags drive all approach routing. When permission_mode=PermissionMode.STANDARDISED, all exposures are assigned SA.

Audit trail — builds a classification reason string for each exposure:

entity_type=corporate; exp_class_sa=CORPORATE; exp_class_irb=CORPORATE;
is_sme=true; is_mortgage=false; is_defaulted=false; is_infrastructure=false;
requires_fi_scalar=false; qualifies_as_retail=true

Step 7: Split by Approach¶

Filters exposures into separate LazyFrames: - sa_exposures - Approach = SA - irb_exposures - Approach = FIRB or AIRB - slotting_exposures - Approach = SLOTTING

Output Schema¶

The classifier adds these columns to the exposure data:

Exposure Classes¶

The system supports these exposure classes (defined in domain/enums.py):

Regulatory References¶

• CRR Art. 112: SA exposure class definitions
• CRR Art. 115: RGLA treatment
• CRR Art. 116: PSE treatment
• CRR Art. 117-118: MDB and international organisation treatment
• CRR Art. 123: Retail exposure criteria
• CRR Art. 147: IRB exposure class definitions
• CRR Art. 153(2): FI scalar (1.25x correlation)
• CRR Art. 501: SME supporting factor
• CRR Art. 501a: Infrastructure supporting factor
• CRE30.6: SME classification (Basel Framework)
• CRE20.65-70: Retail exposure criteria (Basel Framework)