Tier 3 · Internal Strategy

Banking Inverse Model (BankingPack)

M-420 · lifecycle: development · RAT-420-v1.0.0 · conditionally-approved scaffold

Intended Use

Banking Inverse Model (BankingPack) Recover a bank's implied assumptions (PD, LGD, deposit beta, NIM spread, cost/income, RWA density) from public filings for cross-firm banking assumption recovery and valuation.

ANALYTIC SCAFFOLD (Decision 059) — not a booked-number engine. The thing that genuinely exists is the tenk2win BankingPack: an inverse model that inverts an analytic banking forward map (NII via deposit beta + asset spread; provision via PD×LGD; RWA/CET1 via risk density) through the SAME calibrate / run_recovery / surrogate / counterfactual substrate the insurance vertical uses (Decision 037 — vertical-agnostic; the only new content is the pack's θ/y schemas + forward map). It recovers a peer's implied θ from disclosure and supports cross-firm comparison + valuation; it does NOT compute regulatory CECL allowance, RWA/CET1, or NII for the books. The Phase-6 seam test (tenk2win docs/phase6_banking_seam.md, 2026-06-02) found PD, cost/income and RWA density tightly identified; deposit beta and LGD weakly identified; NIM spread under-determined (an honest collinearity, eased by disclosed sensitivities). Real banking engines (CECL/RWA/NII — M-421/M-422/M-423) replace the analytic forward() in Phase 2c; the θ/y schemas and the inverse loop around them are unchanged.


Components

Inputs, processing, outputs

dimensions
D3

Validation Evidence

Validation Evidence

The scaffold's evidence is the Phase-6 seam / identifiability test (tenk2win docs/phase6_banking_seam.md, 2026-06-02): the pack recovers its implied θ through the same calibrate/run_recovery used for insurance — no solver changes.

  • Identified (tight): pd, cost_income, rwa_density.
  • Weakly identified: deposit_beta, lgd.
  • Under-determined: nim_spread (an honest collinearity — PD×LGD and the two NII-spread terms share signal, the banking analogue of the insurance L0 case, eased by disclosed sensitivities).

No production validation pack exists, and none is expected for a scaffold.


Model Documentation

Upgrade Path

Replace the analytic forward() with the real banking engines — M-421 CECL allowance (ASC 326), M-422 RWA / CET1, M-423 NII / IRRBB — and source bank EDGAR (different schedules) + a banking proxy-book analog. The θ/y schemas and the whole inverse loop around them are unchanged. Until then those three are status: planned stubs in the registry (Phase 2c), not built engines.


Change Log

Change Log

  • 2026-06-28 — Created. Banking enters the governance lifecycle as an analytic scaffold (Decision 060 / insightalm #3389). Honest tier (3) / status (under_development); real CECL/RWA/NII engines registered as planned stubs.

References

References

  • tenk2win src/tenk2win/packs/banking/pack.py (the BankingPack forward map).
  • tenk2win docs/phase6_banking_seam.md (the seam / identifiability test).
  • Decision 059 (banking forward map = analytic scaffold), Decision 060 (banking into the governed lifecycle).

2L Inventory Review

Open findings (1)

Independent 2nd-line review (INV-2026-06) — implemented capability vs registered scope. Each carries a recommended fix and is tracked in insightalm-mrm until closed.

HIGH INV-029 · P5 · validation-gap

Validation evidence + change logs missing across most of the inventory

Only M-001/M-020/M-050 carried full documentation packs before this pass. Most models record validation_evidence: missing and change_log: missing with peer_review: pending. Gold tests freeze behaviour but many assert only structural invariants (e.g. reserve>0), not correctness against external truth. The flagship T0-vs-10-K match is circular (BV-032).

Recommendation: For each Tier-1 model: produce a validation-evidence pack (back-test vs disclosed results once BV-032 re-calibration lands, sensitivity suite, challenger comparison), a change log, and a 2L ratification. Sequence behind BV-032 (firm-data) for anything needing 10-K reconciliation.


Ratification

Conditionally-approved scaffold — RAT-420-v1.0.0

2L issued RAT-420-v1.0.0 as a conditionally-approved Tier 3 analytic scaffold. The model exists, runs, and matches its card — but its approval is intelligence-use-only: its outputs must never be read as a booked, regulatory, capital, or reserving figure. Booked use (and tier promotion) is gated on the PLANNED engines, which are not built. The conditions below are on record.

id deadline condition
C-420-01 INTELLIGENCE-USE-ONLY. M-420 is approved ONLY for cross-firm banking assumption recovery + valuation (decision-support intelligence). It is NOT approved for any booked, regulatory, capital, or reserving use. Its outputs (recovered θ, analytic nii/provision/rwa/cet1_ratio/net_income) must never be read as a CECL allowance, an RWA/CET1 capital ratio, or a booked NII figure.
C-420-02 REAL ENGINES ARE THE PRECONDITION FOR BOOKED USE + TIER PROMOTION. Any booked/regulatory banking use requires the PLANNED (not built) engines — M-421 CECL allowance (ASC 326, Tier-1), M-422 RWA/CET1 (Tier-1), M-423 NII/IRRBB (Tier-2) — to be built, registered with real bindings, and independently ratified. M-420 must not be promoted above Tier-3, and its purpose must not be widened beyond internal_strategy, until such an engine replaces the analytic forward() and a use beyond internal strategy is asserted.
C-420-03 RE-RATIFY ON ENGINE REPLACEMENT. When the analytic forward() is replaced by a real banking engine (the Phase-2c upgrade path), M-420 (or its successor) re-enters ratification at the new tier with the full Tier-1/2 effective- challenge battery (validation evidence, gold snapshot) — this RAT does not carry forward to a booked-number engine.