Banking NII / IRRBB

NII over a repricing balance sheet; ΔNII is driven by the asset/liability repricing gap:

UP   (Δr ≥ 0):  ΔNII(Δr) = net_gap_fraction · EarningAssets · Δr
DOWN (Δr < 0):  ΔNII      = a·x + b·x²     (x = |Δr| in 100bps units)

evaluated at the standard IRRBB shocks (±100bps, ±200bps). Disclosed anchors:

1. Level — base earning-asset yield calibrated so base-case NII reconciles to disclosed FY2025 NII.
2. Up-shock sensitivity — the net repricing-gap fraction calibrated to the disclosed +100bps 12-month NII sensitivity. The off-anchor up point (+200bps) stays a non-circular back-tested prediction.
3. Down-shock deposit-floor convexity (C-423-02) — the quadratic down leg a·x + b·x² is calibrated to the disclosed −100/−200bps points: b = (d2 − 2·d1)/2, a = d1 − b. So both disclosed down points reconcile, and b<0 reports the deposit-floor optionality a linear gap structurally omits. With only one disclosed down point (USB) the down leg is linear (b=0, flagged); with none, it falls back to the symmetric linear gap (flagged).

Reuses the parallel-shift rate-scenario framing of finlib's curve/DFAST machinery for the rate leg (REUSE, not re-derive).

• Sourced: total assets (bankview adapter); disclosed FY2025 NII and the disclosed ±bps NII-sensitivity table (FY2025 10-Ks, citations in banklib/disclosed_anchors.yaml).
• Modelled & flagged: the earning-asset base (proxy off total assets) and the repricing-share / deposit-beta-by-bucket split (MD&A narrative, not XBRL). Where no disclosed sensitivity exists for a firm, the gap is set to an asset-sensitive prior and flagged calibrated_sensitivity=False.

bankview/validation_evidence/M-423/ — per-firm modelled-vs-disclosed reconciliation. The +100bps up point reconciles exactly (anchor). C-423-02 closed: the −100bps and −200bps down points now reconcile (err ≈ 0) for the whole disclosed cohort (JPM/BAC/WFC/C/COF) via the per-firm deposit-floor convexity term — previously a −14% to −71% under-prediction (JPM −200bps −30%). The fitted deposit_floor_convexity b is the disclosed down-shock steepening itself. Residual: the off-anchor +200bps up point and USB's single-down-point linear leg stay surfaced limitations. Unit tests: tests/test_banklib_nii.py (incl. convexity reconciliation + the single-/no-down-point fallbacks).

• 2026-06-28 — Created as a planned backlog stub (Decision 060 / insightalm #3389).
• 2026-06-28 — 1L build (insightalm #3414): real top-down NII + repricing-gap IRRBB engine (banklib.L3.nii_irrbb), calibrated to disclosed NII + +100bps sensitivity for JPM/BAC/WFC; back-test, validation evidence + engine_registry binding added. Status → under_development. Ready for 2L re-review.
• 2026-06-28 — 2L verdict (RAT-423-v1.0.0): CONDITIONALLY APPROVED. Registry lifecycle.ratification_ref set to RAT-423-v1.0.0 via the reconciler promote channel. Status held at under_development (not promoted to active) pending the open condition: the −200bps convexity point back-tested against a disclosed downside-shock sensitivity.
• 2026-06-29 — 1L: condition C-423-02 addressed (issue #3432). Extended banklib.nii_irrbb with a per-firm deposit-floor convexity down-shock term a·x + b·x² calibrated to the disclosed −100/−200bps NII sensitivities (free public 10-K rate-risk tables, Decision 062). The down-shocks now reconcile (err ≈ 0) for the disclosed cohort — closing the linear-gap under-prediction (JPM −200bps −30% → 0). deposit_floor_convexity b<0 reports the disclosed steepening; the UP leg stays the linear +100bps-calibrated gap (non-circular preserved). Residual (+200bps up point, USB single-point) surfaced. New unit tests. Ready for 2L re-review.