rwa-calc 0.3.4

Basel 3.1 Credit Risk RWA Calculator compliant with PRA PS1/26

[!IMPORTANT] This package is still in active development and is not production ready - release 1.0.0 will confirm when ready for current CRR and 2.0.0 for Basel 3.1

UK Credit Risk (CR) & Counterparty Credit Risk (CCR) RWA Calculator

A high-performance Risk-Weighted Assets (RWA) calculator for UK CR & CCR, supporting both current regulations and future Basel 3.1 implementation. Built with Python using Polars for vectorized performance.

Documentation: https://OpenAfterHours.club/rwa_calculator/

Installation

Install with uv (recommended)

uv add rwa-calc

Or with pip

pip install rwa-calc

The web UI and REST API ship with the base package — no extra required.

Optional Dependencies

Extra	Description
ui	Legacy alias — UI deps are now included in the base install
dev	Development tools (pytest, ruff, ty, zensical)
all	All optional dependencies

Quick Start

from datetime import date
from rwa_calc.api import CreditRiskCalc

response = CreditRiskCalc(
    data_path="/path/to/data",
    framework="CRR",
    reporting_date=date(2026, 12, 31),
    permission_mode="irb",
).calculate()

if response.success:
    print(f"Total RWA: {response.summary.total_rwa:,.0f}")
    df = response.collect_results()

IRB Mode — set permission_mode="irb" and provide a model_permissions input table to route exposures to FIRB, AIRB, or slotting based on per-model approvals. Exposures without a matching model permission fall back to SA. See the Data Model docs for the model permissions schema.

Interactive UI — server-rendered web interface (calculator, results, CRR vs Basel 3.1 comparison, and legacy-engine reconciliation) plus a REST API and an editable Marimo workbench:

uv add rwa-calc
rwa-ui
# Open http://localhost:8000 in your browser (REST API + OpenAPI at /docs)

Regulatory Scope

This calculator supports two regulatory regimes:

Regime	Effective Period	UK Implementation	Status
CRR (Basel 3.0)	Until 31 December 2026	UK CRR (EU 575/2013 as onshored)	Active
Basel 3.1	From 1 January 2027	PRA PS1/26	Implemented

A configuration toggle allows switching between calculation modes for:

• Current regulatory reporting under UK CRR
• Impact analysis and parallel running ahead of Basel 3.1 go-live
• Seamless transition when Basel 3.1 becomes effective

Migrating from an Existing Calculator

The hardest part of adopting a new credit-risk engine is not installing it — it is signing off that the new numbers are correct on your own book. Parallel-run reconciliation operationalises that discipline: run this calculator and your existing/legacy calculator on the same portfolio, then reconcile the two outputs component by component (exposure class, approach, CCF, EAD, PD, LGD, risk weight, RWA, …).

Every exposure is bucketed as an exact match, a within-tolerance difference, a break, or missing on one side — with our engine's reason and input drivers attached so each break can be traced to either a data / mapping issue or an engine difference.

from datetime import date
from rwa_calc.api import CreditRiskCalc

calc = CreditRiskCalc(
    data_path="/path/to/data",
    framework="CRR",
    reporting_date=date(2026, 12, 31),
    permission_mode="standardised",
)

# Map your legacy output columns onto our components in a small TOML file
response = calc.reconcile("reconciliation.toml")

print(response.collect_totals_tie_out())        # does it tie out?
print(response.collect_summary_by_component())  # which components agree
breaks = response.collect_breaks_detail()       # the break worklist, largest first
response.to_excel("reconciliation.xlsx")        # full report, one sheet per view

You can also run it without writing Python from the Reconciliation page in the web UI (rwa-ui, then open http://localhost:8000/reconciliation) or over HTTP via POST /api/reconcile. This is distinct from Framework Comparison (CRR vs Basel 3.1 on the same engine) — here the other side is your external, legacy output.

Details: See the Parallel-Run Reconciliation guide.

Key Features

• Dual-Framework Support: Single codebase for CRR and Basel 3.1 with UK-specific deviations
• High Performance: Polars LazyFrames for vectorized calculations (50-100x improvement over row iteration)
• Complete Coverage: Standardised (SA), IRB (F-IRB & A-IRB), and Slotting approaches
• Credit Risk Mitigation: Collateral, guarantees, and provisions with RWA-optimized allocation
• Complex Hierarchies: Multi-level counterparty and facility hierarchy support
• Audit Trail: Full calculation transparency for regulatory review
• Framework Comparison: Side-by-side CRR vs Basel 3.1 impact analysis
• Parallel-Run Reconciliation: Tie this calculator out against your existing/legacy engine component-by-component (EAD, RWA, risk weight, PD, LGD, …) to sign off a migration with confidence
• COREP Output: Export results to COREP regulatory templates
• Multiple Export Formats: Parquet, CSV, Excel, and COREP

Supported Approaches

Approach	Description
Standardised (SA)	Risk weights based on external ratings and exposure characteristics
Foundation IRB (F-IRB)	Bank-estimated PD, supervisory LGD
Advanced IRB (A-IRB)	Bank-estimated PD, LGD, and EAD
Slotting	Category-based approach for specialised lending

Supported Exposure Classes

Sovereign / Central Bank, Institution, Corporate, Corporate SME, PSE, MDB, RGLA, Retail Mortgage, Retail QRRE, Retail Other, Specialised Lending, Equity, Covered Bond, Residential Mortgage, Commercial Mortgage (Basel 3.1 RE split), High Risk, Defaulted, Other

Documentation

Comprehensive documentation is available at OpenAfterHours.club/rwa_calculator

Section	Description
Getting Started	Installation and first calculation
User Guide	Regulatory frameworks, methodology, exposure classes
Reconciliation	Parallel-run tie-out against a legacy calculator for migration
Architecture	System design and pipeline
Data Model	Input schemas and validation
API Reference	Complete technical documentation
Development	Testing, benchmarks, contributing
Plans	Development roadmap and status

Running Tests

Test fixtures are parquet files generated from Python builders and are not checked into the repo (they are gitignored). Generate them once before running the test suite for the first time, and again whenever a fixture builder under tests/fixtures/ changes:

# Generate all test fixtures (parquet output, 8 groups)
uv run python tests/fixtures/generate_all.py

# Run all tests
uv run pytest -v

# Run with coverage
uv run pytest --cov=src/rwa_calc

# Run benchmarks (10K + 100K, excludes 1M/10M)
uv run pytest tests/benchmarks/ -m "benchmark and not slow" -k "not 1m" -o "addopts=" --benchmark-only -v

Test Results: ~5,500 tests (~4,675 unit + ~500 acceptance + ~165 contract + ~145 integration + ~30 benchmark)

License

Apache-2.0 license

References

Current Regulations (CRR / Basel 3.0)

• PRA Rulebook - CRR Firms
• UK CRR - Regulation (EU) No 575/2013 as onshored

Basel 3.1 Implementation (January 2027)

• PRA PS1/26 - Basel 3.1 Final Rules
• PS1/26 Appendix 1 - Regulations
• PS1/26 Appendix 17 - Template Guidance
• Basel Committee - CRE: Calculation of RWA for credit risk