Comparison & Impact Analysis¶

The RWA Calculator supports dual-framework comparison, capital impact attribution, and transitional floor schedule modelling. These features help firms prepare for the CRR to Basel 3.1 transition by quantifying the impact on their credit risk capital requirements.

Why Comparison Matters¶

UK firms must operate under both CRR (until 31 Dec 2026) and Basel 3.1 (from 1 Jan 2027) during the transition period. Understanding the RWA impact at exposure, portfolio, and timeline levels is essential for capital planning, board reporting, and ICAAP stress testing. The comparison module runs both frameworks against the same portfolio from a single codebase, ensuring consistent methodology and eliminating reconciliation issues.

Dual-Framework Comparison (M3.1)¶

DualFrameworkRunner runs the same portfolio through both CRR and Basel 3.1 pipelines and joins the results on exposure_reference.

Usage¶

from datetime import date
from rwa_calc.contracts.config import CalculationConfig
from rwa_calc.engine.comparison import DualFrameworkRunner

runner = DualFrameworkRunner()

crr_config = CalculationConfig.crr(reporting_date=date(2026, 12, 31))
b31_config = CalculationConfig.basel_3_1(reporting_date=date(2030, 6, 30))

comparison = runner.compare(data, crr_config, b31_config)

Output: ComparisonBundle¶

Exposure Deltas¶

The exposure_deltas LazyFrame contains per-exposure comparison columns:

Delta convention: Positive delta = Basel 3.1 is higher (increased capital).

The join uses a full outer join, so exposures unique to one framework (e.g., equity exposures removed from IRB under Basel 3.1) still appear with nulls filled to 0.0.

Capital Impact Analysis (M3.2)¶

CapitalImpactAnalyzer decomposes the RWA delta into four regulatory drivers using a sequential waterfall methodology. The four drivers sum exactly to the total delta.

Usage¶

from rwa_calc.engine.comparison import CapitalImpactAnalyzer

analyzer = CapitalImpactAnalyzer()
impact = analyzer.analyze(comparison)

Waterfall Drivers¶

The waterfall applies sequentially, with each driver capturing a specific regulatory change:

Step 1 — Scaling factor removal (1.06x): CRR multiplies IRB capital by 1.06. Basel 3.1 removes this. Always negative for IRB exposures, zero for SA.

Step 2 — Supporting factor removal: CRR applies SME (0.7619/0.85) and infrastructure (0.75) factors as RWA discounts. Basel 3.1 removes them. Typically positive (removing the discount increases RWA).

Step 3 — Methodology & parameter changes: The residual after accounting for the other three drivers. Captures PD/LGD floor changes, SA risk weight table revisions, F-IRB supervisory LGD changes, correlation formula updates, and all other rule differences.

Step 4 — Output floor impact: Basel 3.1 floors IRB RWA at a percentage of SA RWA. Additional RWA from the floor binding. Zero for SA exposures.

Additivity Invariant¶

For every exposure:

scaling_factor_impact + supporting_factor_impact + output_floor_impact + methodology_impact == delta_rwa

This invariant is enforced by computing the methodology impact as the residual.

Output: CapitalImpactBundle¶

Portfolio Waterfall¶

The portfolio_waterfall LazyFrame has 4 rows:

The final cumulative_rwa equals the total Basel 3.1 RWA.

Transitional Floor Schedule (M3.3)¶

TransitionalScheduleRunner models the output floor phase-in from 2027 to 2032 by running the Basel 3.1 pipeline at each transitional reporting date.

PRA Output Floor Phase-In Schedule¶

Usage¶

from rwa_calc.domain.enums import PermissionMode
from rwa_calc.engine.comparison import TransitionalScheduleRunner

runner = TransitionalScheduleRunner()

schedule = runner.run(
    data=raw_data_bundle,
    permission_mode=PermissionMode.IRB,
)

# Access the timeline
timeline = schedule.timeline.collect()
print(timeline)

# Access individual year results
result_2027 = schedule.yearly_results[2027]
result_2030 = schedule.yearly_results[2030]

Custom Reporting Dates¶

from datetime import date

# Model specific dates instead of the default 4-year schedule
schedule = runner.run(
    data=raw_data_bundle,
    permission_mode=PermissionMode.IRB,
    reporting_dates=[date(2027, 12, 31), date(2030, 12, 31)],
)

Output: TransitionalScheduleBundle¶

Timeline Columns¶

Regulatory References¶