creditriskengine 0.3.0

Production-grade open-source credit risk analytics: Basel III/CRR3 RWA, IFRS 9/CECL ECL, IRB modeling, and model validation

CreditRiskEngine

Production-grade open-source credit risk analytics library.

The scikit-learn of credit risk.

Features

• RWA Calculation -- Basel III/IV Standardized Approach and IRB (F-IRB / A-IRB) with output floor phase-in, double default (CRE32), and equity IRB (CRE33)
• ECL Engines -- IFRS 9, US CECL (ASC 326), and Ind AS 109 with staging, SICR, lifetime PD, and scenario weighting
• PD / LGD / EAD Modeling -- Scorecard development, calibration (anchor-point & Bayesian), TTC-to-PIT conversion, term structures, Merton structural model, Altman Z-score, and transition matrix estimation
• Model Validation -- Discrimination (AUROC, Gini, KS, IV), calibration (binomial, Hosmer-Lemeshow, traffic-light), stability (PSI, CSI, migration)
• Portfolio Risk -- Vasicek ASRF, Gaussian copula Monte Carlo, parametric VaR, economic capital, and stress testing (including reverse stress)
• Concentration Risk -- Single-name, sector-level, and granularity adjustment analytics
• Capital Adequacy -- Capital buffers (CConB, CCyB, G-SIB/D-SIB), leverage ratio (CRE80), and MDA framework
• CVA Risk -- BA-CVA (CVA25) and SA-CVA delta risk charge (CVA26) with supervisory parameters
• Market Risk -- FRTB credit spread SbM (MAR21), Default Risk Charge (MAR22), and RRAO (MAR23)
• Securitisation -- SEC-SA, SEC-ERBA, and SEC-IRBA per CRE40-45
• Operational Risk -- Standardised Measurement Approach (SMA) per OPE25
• Credit Risk Mitigation -- Comprehensive and simple approaches, haircut framework per CRE22
• Multi-Jurisdiction -- EU CRR3, UK PRA, US Basel III Endgame, India RBI, Singapore MAS, Hong Kong HKMA, Japan JFSA, Australia APRA, Canada OSFI, Saudi Arabia SAMA, and BCBS baseline
• Regulatory Reporting -- COREP, Pillar 3 disclosure templates (CR1/CR3/CR4/CR6), FR Y-14 (CCAR), and model inventory
• Stress Testing -- EBA, BoE ACS, US CCAR/DFAST, RBI frameworks, and reverse stress testing

Installation

pip install creditriskengine

From source

git clone https://github.com/ankitjha67/baselkit.git
cd baselkit
pip install -e ".[dev]"

Requires Python 3.11+.

Quick Start

IRB Risk Weight

from creditriskengine.rwa.irb.formulas import irb_risk_weight

# Corporate exposure: PD=1%, LGD=45%, maturity=2.5y
rw = irb_risk_weight(pd=0.01, lgd=0.45, asset_class="corporate", maturity=2.5)
print(f"Risk Weight: {rw:.2f}%")

Standardized Approach

from creditriskengine.rwa.standardized.credit_risk_sa import assign_sa_risk_weight
from creditriskengine.core.types import SAExposureClass, CreditQualityStep, Jurisdiction

rw = assign_sa_risk_weight(
    exposure_class=SAExposureClass.CORPORATE,
    cqs=CreditQualityStep.CQS_2,
    jurisdiction=Jurisdiction.EU,
)
print(f"SA Risk Weight: {rw:.0f}%")

IFRS 9 ECL

from creditriskengine.ecl.ifrs9.ecl_calc import calculate_ecl
from creditriskengine.core.types import IFRS9Stage

ecl = calculate_ecl(
    stage=IFRS9Stage.STAGE_1,
    pd_12m=0.02,
    lgd=0.40,
    ead=1_000_000,
    eir=0.05,
)
print(f"12-month ECL: {ecl:,.2f}")

PD Scorecard

from creditriskengine.models.pd.scorecard import scorecard_to_pd, assign_rating_grade, build_master_scale
import numpy as np

scores = np.array([537, 587, 640, 706])
pds = scorecard_to_pd(scores)  # Convert scorecard points to PD

master_scale = build_master_scale([0.0003, 0.001, 0.005, 0.01, 0.02, 0.05, 0.10, 0.20, 1.0])
grades = [assign_rating_grade(pd, master_scale) for pd in pds]
# grades: ['Grade_7', 'Grade_5', 'Grade_2', 'Grade_1']

Model Validation

from creditriskengine.validation.discrimination import auroc, gini_coefficient, ks_statistic
import numpy as np

y_true = np.array([0, 0, 1, 0, 1, 1, 0, 0, 1, 0])
y_score = np.array([0.1, 0.2, 0.7, 0.3, 0.8, 0.6, 0.2, 0.15, 0.9, 0.05])

print(f"AUROC: {auroc(y_true, y_score):.4f}")
print(f"Gini:  {gini_coefficient(y_true, y_score):.4f}")
print(f"KS:    {ks_statistic(y_true, y_score):.4f}")

Project Structure

src/creditriskengine/
    core/               # Exposure model, portfolio container, enums, audit trail, logging
    regulatory/         # Jurisdiction YAML configs (17 jurisdictions) and loader
    rwa/
        standardized/   # SA risk weight tables (CRE20)
        irb/            # IRB formulas (CRE31) -- correlation, K, RW
        output_floor.py # Output floor phase-in by jurisdiction
        capital_buffers.py # CConB, CCyB, G-SIB/D-SIB, MDA (RBC40)
        leverage_ratio.py # Basel III leverage ratio (CRE80)
        cva.py          # BA-CVA (CVA25) and SA-CVA (CVA26) capital charges
        market_risk.py  # FRTB SbM, DRC, RRAO (MAR21-23)
        securitisation.py # SEC-SA, SEC-ERBA, SEC-IRBA (CRE40-45)
        operational_risk.py # Standardised Measurement Approach (OPE25)
        crm.py          # Credit risk mitigation, haircuts (CRE22)
    ecl/
        ifrs9/          # Staging, SICR, lifetime PD, scenarios, TTC-to-PIT, ECL calc
        cecl/           # PD*LGD, loss-rate, vintage, DCF, qualitative factors
        ind_as109/      # India-specific wrapper over IFRS 9
    models/
        pd/             # Scorecard, calibration, master scale, Vasicek PD, Merton, Z-score, transition matrices
        lgd/            # Workout LGD, downturn LGD, term structure, floors, cure rate
        ead/            # CCF estimation, supervisory CCF, EAD term structure
        concentration/  # Single-name, sector, granularity adjustment
    portfolio/          # Copula simulation, VaR, economic capital, stress testing, Vasicek ASRF
    validation/         # Discrimination, calibration, stability, backtesting, benchmarking
    reporting/          # COREP, Pillar 3 (CR1/CR3/CR4/CR6), FR Y-14 (CCAR), model inventory

Testing

# Run full test suite
pytest

# Quick run without coverage
pytest -q --no-cov

# Specific module
pytest tests/test_rwa/ -v

1,786 tests across all modules with 100% line coverage. Type-checked with mypy --strict and linted with ruff.

Performance

Benchmarked on a single core (run python benchmarks/bench_portfolio.py):

Operation	Throughput
IRB risk weight (single)	~100k calc/sec
IRB portfolio (10k exposures)	~10k exp/sec
SA risk weight (10k exposures)	~100k exp/sec
IFRS 9 ECL (10k calculations)	~100k calc/sec
Stress test (10k × 3yr)	< 0.01s
Monte Carlo (10k × 10k sims)	< 2s

Dependencies

Package	Version	Purpose
NumPy	≥1.26, <3.0	Numerical computation
SciPy	≥1.12, <2.0	Statistical distributions (norm CDF/PPF)
pandas	≥2.2, <3.0	Data manipulation and audit trail export
Pydantic	≥2.6, <3.0	Data validation and exposure model
PyYAML	≥6.0.1, <7.0	Regulatory configuration loading
scikit-learn	≥1.4, <2.0	AUC, logistic regression, model validation
statsmodels	≥0.14, <1.0	Statistical tests for calibration
Jinja2	≥3.1, <4.0	Regulatory report templating

Documentation

Full documentation: https://ankitjha67.github.io/baselkit/

Build locally:

pip install -e ".[docs]"
mkdocs serve

Governance

• Regulatory Mapping: Every function traced to its Basel/IFRS paragraph
• Regulatory Disclaimers: Important caveats for production use
• Config Versioning: How regulatory config changes are managed
• Audit Trail: AuditTrail class records every calculation with inputs, outputs, timestamps, and regulatory references
• Input Validation: Schema validation for YAML configs and sanitization for exposure inputs

Contributing

See CONTRIBUTING.md for full guidelines. Quick start:

git clone https://github.com/ankitjha67/baselkit.git
cd baselkit
pip install -e ".[dev]"
pytest                   # Run tests
ruff check src/ tests/   # Lint
mypy src/                # Type check

License

Apache 2.0 -- see LICENSE for details.

Disclaimer: This library is provided for educational and analytical purposes. It has not been reviewed or endorsed by any regulatory authority. See Regulatory Disclaimers for details.