Hierarchy & Classification Specification¶

Counterparty hierarchy resolution, rating inheritance, and exposure class determination.

Counterparty Hierarchy¶

Organisation Mappings¶

The calculator resolves parent-child relationships between counterparties using org_mappings:

• Child counterparties inherit ratings from their parent when they lack their own
• The hierarchy is traversed upward until a rated entity is found

Multi-Rating Resolution (CRR Art. 138)¶

When a counterparty has more than one external rating from nominated ECAIs, the resolver applies CRR Art. 138 rather than picking the most recent:

1. Per-agency dedup — multiple assessments from the same agency are first reduced to the most recent (one assessment per agency).
2. Art. 138 selection across the remaining distinct-agency assessments:
   • 1 assessment → use it.
   • 2 assessments → use the higher risk weight (worse of the two).
   • ≥ 3 assessments → take the two assessments generating the two lowest risk weights, then use the higher of those two (i.e. the "second-best" rating).

Resolution is performed on CQS rather than RW because the CQS → RW mapping is monotone non-decreasing within every SA exposure class, so ranking by CQS ascending gives the same outcome as ranking by RW. External ratings are not inherited from the parent — only the counterparty's own assessments participate in Art. 138 resolution.

Lending Group Aggregation¶

Lending groups aggregate exposure across related counterparties for threshold calculations (e.g., SME turnover, retail exposure limits).

• Members are defined via lending_mappings
• The parent counterparty is automatically included as a member
• Duplicate membership is resolved (a counterparty appearing in multiple groups keeps only the first assignment)
• Residential property exposures are excluded from retail aggregation per CRR Art. 123(c)

Facility-to-Exposure Mapping¶

The facility_mappings table links facilities to their underlying exposures (loans and contingents):

• Facility undrawn amount = max(facility_limit - sum(drawn_amounts) - sum(contingent_nominals), 0)
• Supports multiple exposure types under a single facility
• Pro-rata allocation of facility-level attributes

Multi-Level Facility Hierarchies¶

Facilities can form their own hierarchies (e.g., a master facility with sub-facilities beneath it). The resolver handles this via facility root lookup — an iterative traversal that mirrors the counterparty hierarchy pattern:

• Facility-to-facility edges are identified from facility_mappings where child_type = "facility"
• The hierarchy is traversed upward (up to 10 levels) to find the root facility for each sub-facility
• Output columns: child_facility_reference, root_facility_reference, facility_hierarchy_depth

Undrawn Amount Aggregation¶

For multi-level facility hierarchies, drawn amounts from loans under sub-facilities are aggregated up to the root facility. The root undrawn equals the parent's headroom net of every drawn loan/contingent across the descendant tree:

Root Facility (limit = 1,000,000)
├── Sub-Facility A
│   ├── Loan 1 (drawn = 200,000)
│   └── Loan 2 (drawn = 100,000)
└── Sub-Facility B
    └── Loan 3 (drawn = 150,000)

Root undrawn = 1,000,000 - (200,000 + 100,000 + 150,000) = 550,000

Key rules:

• Root/standalone facilities produce undrawn exposure records
• Sub-facilities are excluded from producing their own undrawn records (avoids double-counting)
• Negative drawn amounts are clamped to zero before aggregation (negative balances do not increase headroom)
• Only facilities with undrawn_amount > 0 and committed = True generate exposure records

Multi-Option Facility (MOF) Waterfall Allocation¶

A Multi-Option Facility is any facility with at least one child_type='facility' mapping. Each sub-facility carries its own risk_type (and therefore its own SA CCF), so the parent's undrawn headroom is split across sub-facility CCF buckets — not collapsed onto the worst-case CCF as in earlier versions of the engine.

Allocation rule (per MOF parent):

1. Compute per-sub headroom: sub_headroom = max(0, sub_limit − sub_drawn), where sub_drawn sums loans and contingents directly mapped to that sub (no roll-up).
2. Compute parent headroom: parent_headroom = max(0, parent_limit − total_drawn − total_contingent) rolled up across the whole descendant tree.
3. Sort committed sub-facilities by descending SA CCF, with deterministic tie-break on risk_type then facility_reference.
4. Walk subs in order. Allocate min(sub_headroom_i, max(0, parent_headroom − cum_{i-1})) to each sub, where cum_{i-1} is the running sum of higher-CCF allocations.
5. Emit one facility_undrawn row per sub with allocation > 0. Each row carries the sub's risk_type and counterparty_reference; provenance is captured in mof_risk_type_source.
6. If parent_headroom > sum(allocations), emit one residual row at the parent's own risk_type and counterparty_reference for the leftover headroom.

Worked example — sub-limits exceed parent limit (waterfall caps):

FAC_01    limit £100m,  parent's risk_type = MR
├── FAC_SUB_01  limit £60m, risk_type = MR  (50% CCF)
└── FAC_SUB_02  limit £60m, risk_type = MLR (20% CCF)

Worked example — drawn loans nets per sub, residual emerges:

FAC_01    limit £100m,  parent's risk_type = FR (100% CCF)
├── FAC_SUB_01  limit £50m, MR (50% CCF), Loan_A drawn £20m
└── FAC_SUB_02  limit £30m, MLR (20% CCF)

per-sub: sub_01 headroom = 30m, sub_02 headroom = 30m
parent headroom = 100m − 20m = 80m

Output schema:

• exposure_reference = "{parent_ref}_UNDRAWN_{sub_ref}" for sub waterfall rows; "{parent_ref}_UNDRAWN_RESIDUAL" for the parent residual row.
• source_facility_reference = parent_ref on every row, so facility-level collateral allocation and downstream rollups still group by the MOF parent.
• mof_risk_type_source = sub_ref on sub rows; null on the residual row.

Edge cases:

• Uncommitted sub (committed=False): skipped entirely from the waterfall — no row, no headroom consumption. The bank can refuse to lend, so the sub carries no commitment EAD.
• Uncommitted parent (committed=False): no rows emitted at all (existing behaviour preserved). Loans/contingents already mapped under the parent flow as their own exposure rows unaffected.
• Sub with zero headroom (fully drawn): drops out of the waterfall — emits no row.
• Sub-limits sum below parent limit: residual at parent's own risk_type covers the gap.
• Standalone facility (no facility-typed children): single undrawn row with parent's own risk_type (non-MOF path unchanged).

Type Column Handling¶

The engine's contract is a single discriminator column child_type, guaranteed to exist by the loader edge seal — the resolver itself carries no presence guards. The loader handles the three accepted input shapes exactly once at load:

A file carrying both child_type and node_type is structurally ambiguous: the loader's alias rename collides at schema resolution and the load wrappers surface it (required table → DataLoadError; optional table → DQ007).

Exposure Classification¶

Entity Type to Exposure Class¶

Counterparty entity type determines the base SA exposure class:

Basel 3.1 Exposure Class Priority (Art. 112)¶

PRA PS1/26 Art. 112 Table A2 defines 16 exposure classes with a strict priority ordering. When an exposure could belong to multiple classes, the highest-priority class takes precedence:

Calculator coverage: The calculator currently implements classes for: central govt/CB, RGLA, PSE, MDB, institution, corporate, specialised lending, retail, equity, real estate, ADC, and default. Securitisation, CIU (beyond 1,250% fallback per Art. 132(2) / Art. 132c), and covered bonds are tracked as future enhancements.

SME Detection¶

Corporate counterparties are reclassified as CORPORATE_SME when:

• CRR: Group turnover < EUR 50m (GBP converted to EUR at configured rate)
• Basel 3.1: Group turnover < GBP 44m (direct GBP threshold per PRA PS1/26, calculated on highest consolidated accounts of the group; no FX conversion needed)

Retail Qualification¶

Individual counterparties qualify for retail treatment when:

• CRR: Aggregate exposure < EUR 1m (GBP ~873k at default EUR/GBP rate of 0.8732)
• Basel 3.1: Aggregate exposure < GBP 880k
• QRRE limit (IRB Art. 147(5A)(c)): Largest aggregate nominal exposure to any single individual in the QRRE sub-portfolio ≤ EUR 100k (CRR) / GBP 90,000 (Basel 3.1). This is a portfolio-level constraint, not a per-facility check.

If retail thresholds are breached, the exposure is reclassified as CORPORATE.

Basel 3.1 Retail Qualifying Criteria (Art. 123A)¶

Under Basel 3.1, Art. 123A has a two-path structure:

• Art. 123A(1)(a) — SME retail: Exposures to SMEs automatically qualify as regulatory retail without further conditions.

• Art. 123A(1)(b) — Natural person retail: Three conditions must all be met:

• Product type (Art. 123A(1)(b)(i)): The exposure takes the form of revolving credits/lines of credit (credit cards, overdrafts), personal term loans/leases (instalment loans, auto loans, student loans), or small business facilities. Must not be a derivative, bond, or equity instrument. Mortgages are excluded (separate class).

• Granularity (Art. 123A(1)(b)(ii)): Total exposure to the obligor (or connected group) does not exceed GBP 880,000. No single exposure represents more than 0.2% of the retail portfolio.
• Pool management (Art. 123A(1)(b)(iii)): The exposure is part of a significant number of similarly managed exposures with similar characteristics. This is a qualitative/attestation requirement, not a calculated check.

Large Corporate Revenue Threshold (Basel 3.1)¶

Under Basel 3.1, corporates with consolidated annual revenue exceeding EUR 500 million (GBP 440 million) are classified as large corporates and are restricted to F-IRB only (cannot use A-IRB). This threshold is distinct from the SME firm-size adjustment threshold (EUR 50m / GBP 44m).

FSE Classification Requirements¶

Financial sector entity (FSE) classification is required under Basel 3.1 for:

• Applying the correct F-IRB LGD (45% for FSE vs 40% for non-FSE, per Art. 161(1))
• Applying the FI scalar (1.25x correlation multiplier)
• Determining approach restrictions under Art. 147A(1)(e) (all FSEs → F-IRB only, not just large FSEs)

The cp_is_financial_sector_entity flag is sourced from the counterparty schema and propagated through the classifier. It is used for the Art. 147A(1)(e) F-IRB-only block and for applying the 45% FSE LGD floor.

Defaulted Exposures¶

Exposures flagged with a default status are identified and tracked throughout the calculation. Defaulted status affects risk weighting (e.g., 150% SA risk weight for defaulted unsecured).

Approach Assignment¶

Basel 3.1 Approach Restrictions (Art. 147A)¶

Under Basel 3.1, PRA PS1/26 Art. 147A mandates specific approaches by exposure sub-class. These are not optional — firms cannot choose an alternative even if they have model approval:

Art. 112 Table A2 Priority Ordering¶

When an exposure qualifies for multiple SA exposure classes, the highest-priority class takes precedence. See the priority table in the Exposure Classification section above. The classifier should apply this ordering systematically rather than relying on entity type alone.

Dual-Approach Split¶

Based on IRB permissions in the configuration, exposures are routed to:

1. SA - Standardised Approach
2. IRB - Foundation IRB or Advanced IRB
3. Slotting - Specialised lending categories
4. Equity - Equity exposures (pass-through, no CRM applied)

Real estate classification derivation¶

The Basel 3.1 SA real-estate framework (Art. 124 to Art. 124L) relies on a number of classification flags — is_adc, is_materially_dependent, is_mixed_use, exposure_subclass for corporate granularity — that the current pipeline consumes as pre-tagged inputs from the caller. None of the derivations below are performed by the classifier today; each subsection records the regulatory expectation for full compliance.

ADC classification trigger (PS1/26 App 1, Art. 124K)¶

Verbatim PDF definition (PS1/26 App 1, p.3, Glossary entry):

"ADC exposure means an exposure to a corporate or special purpose entity financing any land acquisition for development and construction purposes, or financing development and construction of any residential real estate or commercial real estate."

Verbatim PDF quote (PS1/26 App 1, p.58, Art. 124K):

"Subject to paragraph 2, an institution shall assign a risk weight of 150% to an ADC exposure. An institution may assign a risk weight of 100% to an ADC exposure financing any land acquisition for the development and construction of residential real estate, or financing the development and construction of residential real estate if: (a) the exposure is subject to prudent underwriting standards …; and (b) at least one of the following conditions is met: (i) legally binding pre-sale or pre-lease contracts … amount to a significant portion of total contracts; or (ii) the borrower has substantial equity at risk."

Expected derivation. ADC classification should be derived from the conjunction of:

• product_type ∈ {construction_loan, development_finance} OR acquisition_financing == True OR the exposure funds ADC activity as defined in the Glossary;
• counterparty_entity_type ∈ {corporate, spv, special_purpose_entity} (natural-person ADC lending is out of scope of Art. 124K by definition); and
• the property purpose is development/construction of RRE or CRE (not a completed income-producing or owner-occupied building).

Current implementation. The classifier consumes is_adc as a pre-set boolean on the exposure schema. The 100% concessionary branch of Art. 124K(2) (pre-sale/equity-at-risk) also depends on caller-supplied flags.

Art. 124(4) mixed-use splitting¶

Verbatim PDF quote (PS1/26 App 1, p.50, Art. 124(4)):

"An institution shall split a mixed real estate exposure into a residential real estate exposure and a commercial real estate exposure according to the ratio of the values of the residential real estate and the commercial real estate that the exposure is secured by. An institution shall assign the relevant risk weights set out in Article 124J to each part of the exposure …"

Expected derivation. When an exposure's collateral comprises both RRE and CRE portions, the exposure must be split pro-rata by property value. The RRE portion is risk-weighted under Art. 124F–124I (as RRE), and the CRE portion under Art. 124J (as CRE) — each with its own RRE/CRE-specific LTV and materially-dependent logic. A single blended risk weight is not permitted.

Current implementation. Real-estate exposures are classified as either RRE or CRE on the basis of a single is_residential_real_estate flag. Mixed-use splitting is not performed.

Art. 124E three-property limit for "materially dependent"¶

Verbatim PDF quote (PS1/26 App 1, p.54, Art. 124E(1)–(2)):

"A residential real estate exposure is materially dependent on the cash-flows generated by the property unless it is: (a) to one or more natural persons and the exposure is secured by a single property that is the obligor's primary residence; (b) to one or more natural persons that individually meet the three property limit in accordance with paragraph 2; …

A natural person meets the three property limit referred to in point (b) of paragraph 1 if they have no more than three qualifying properties. A qualifying property is a property that is residential real estate, is not the primary residence of the natural person and that is either: (a) security for a residential real estate exposure to the natural person …; or (b) security for a residential real estate exposure to an entity which is created specifically to finance and/or operate immovable property, where the natural person acts as a guarantor …"

Expected derivation. Natural-person obligors with four or more simultaneous buy-to-let / investment RRE exposures (i.e. breaching the three qualifying-property limit in paragraph 2) should automatically flip from Art. 124F (non-materially-dependent RRE — lower risk weights via Art. 124L counterparty RWs) to Art. 124G (materially-dependent RRE — higher property-LTV-based weights). The three-property count must be computed across all lenders' RRE exposures to the obligor (per the "regardless of which lender has the residential real estate exposure" qualifier), but in practice a single-firm calculator can only count exposures it sees.

Current implementation. The classifier consumes is_materially_dependent as a caller-supplied flag on the exposure schema. No automatic counting of qualifying properties is performed.

Art. 147A(1)(e)/(f) subclass reporting for corporates¶

Verbatim PDF quote (PS1/26 App 1, p.92, Art. 147A(1)(e)–(f)):

"(e) for point (c)(ii) of Article 147(2) (financial corporates and large corporates): (i) the Standardised Approach for exposures where permission has been granted under Article 148 or Article 150; (ii) the Foundation IRB Approach for all other exposures within that exposure subclass;

(f) for point (c)(iii) of Article 147(2) (other general corporates): (i) the Standardised Approach for exposures where permission has been granted under Article 148 or Article 150; (ii) the Advanced IRB Approach for exposures where permission has been granted under Article 143(2A) or (2B) to use the Advanced IRB Approach; (iii) the Foundation IRB Approach for all other exposures within that exposure subclass …"

Expected derivation. Corporate exposures subject to Art. 147A should carry three distinct exposure_subclass values corresponding to the Art. 147(2)(c)(ii)/(iii) split:

Even though large corporate and FSE share the same approach restriction (F-IRB only), they should be reported separately because COREP templates split these sub-classes on distinct rows and PRA Pillar 3 disclosures (Annex XX / XXII) use sub-class-level granularity. Likewise, "other general corporates" differs in that A-IRB remains available with the Art. 143(2A)/(2B) permission.

Current implementation. The classifier enforces the approach restrictions (F-IRB block on large corporates and FSEs) but does not emit a distinct exposure_subclass value for each of (d), (e) and (f). All three flow through as a generic CORPORATE class with approach-level distinction only.

FX Conversion¶

All monetary values are converted to the base currency (GBP) using provided FX rates before calculation.

Key Scenarios¶

Hierarchy Resolution (HIER)¶

Exposure Classification (CLASS)¶