Decision Accounting and Bank Model-Risk Regimes: A Clause-Level Crosswalk to SR 11-7 and BCBS 239

DA records add decision reasoning without satisfying SR 11-7 or BCBS 239

The paper argues that Decision Accounting is complementary to two banking regimes: SR 11-7 governs model risk management, and BCBS 239 governs risk data aggregation and reporting. DA adds a structured record of why a model or data architecture decision was made.

• SR 11-7 focuses on model definition, development, implementation, validation, monitoring, independent review, and governance.
• BCBS 239 focuses on governance, data architecture, aggregation, reporting, supervisory review, remedial actions, and home-host cooperation.
• DA records document a decision, not a model inventory, validation package, data architecture, or aggregation control.
• The analysis is proposed and not yet peer-reviewed.

The 17-field record captures who decided, what evidence mattered, and what system costs were expected

Decision Accounting uses a fixed decision record with fields for authority, evidence, alternatives, assumptions, limitations, dissent, implementation, review, private payoffs, and system welfare.

• Fields 1 and 2 identify the decision-maker and approver.
• Fields 5 through 10 capture risk assessment, alternatives, model-selection rationale, data sources, assumptions, and limitations.
• Field 11 records dissent and how objections were handled.
• Field 17, SYSTEM WELFARE, is the structural addition that bank model and data regimes do not require.

System welfare is a separate coordinate, not a function of A and B payoffs

The Missing System Theorem states that no function f can express system welfare W from only the two bargaining parties' payoffs A and B. Two transactions can have identical private payoffs and different system effects.

• The proof uses T1 and T2 with A(T1)=A(T2) and B(T1)=B(T2), while W(T1) differs from W(T2).
• The system includes non-transacting stakeholders, future generations, and the natural environment.
• A loan can benefit both borrower and lender while imposing different systemic risk on third parties.
• Field 17 records the missing coordinate independently.

A Hollow Win is private gain with negative system welfare

The Private Pareto Theorem defines the Hollow Win as A=1, B=1, C=0: both transacting parties have non-negative payoffs, while system welfare is negative.

• A=1 means party A's payoff is non-negative.
• B=1 means party B's payoff is non-negative.
• C=0 means system welfare W is negative.
• The paper names the 2008 financial crisis, Enron, and the 2023 bank failures as Hollow Win cases.

β_W measures system welfare change against annual revenue, never profit

System-welfare beta is defined as β_W = ΔW / Π, where Π is annual revenue and ΔW is annual system welfare change measured in the same numeraire.

• β_W = 1 means welfare destruction equals revenue generation.
• β_W > 1 means welfare destruction exceeds revenue generation.
• β_W < 1 means welfare destruction is below revenue generation.
• Revenue is used because profit can be changed through accounting choices.

DA helps with model-selection reasoning, not model validation

The paper maps DA records against seven SR 11-7 domains. DA partially supports domains where examiners need evidence of reasoning, and is orthogonal where SR 11-7 requires technical model-risk controls.

• Model definition and inventory: orthogonal, because a DA record is not a model inventory.
• Development documentation: partial support through Fields 7, 8, 9, and 10, but not coefficient estimates, testing, benchmarking, or code versioning.
• Validation: orthogonal, because DA does not replace independent validation by qualified personnel.
• Ongoing monitoring: orthogonal, because a DA record is static unless later decisions are separately recorded.

DA records give reviewers evidence, while SR 11-7 still requires the review function

For independent review and governance, DA records provide a structured artifact that reviewers can inspect. They do not create the bank's governance structure or replace internal audit, validation, or board oversight.

• Independent review: partial support because the five-minute/stranger test lets a reviewer reconstruct the decision without relying on memory.
• Governance and roles: partial support because Fields 1 and 2 name the decision-maker and approver.
• Implementation and use documentation: partial support through Fields 12 and 13, but not system integration specs or user procedures.
• The Conflictoring Protocol is presented as a mechanism for challenging system-harming decisions, not as an SR 11-7 substitute.

DA is orthogonal to data architecture and aggregation principles 1 through 7

BCBS 239 Principles 1 through 7 require governance, data architecture, IT infrastructure, accuracy, completeness, timeliness, and adaptability in risk data aggregation. DA records can reference data sources, but they do not build or control those systems.

• Principle 1: DA does not constitute a governance framework.
• Principle 2: DA does not constitute data architecture or IT infrastructure.
• Principles 3 and 7: DA does not ensure data accuracy or integrity.
• Principles 4 through 6: DA does not aggregate complete, timely, or adaptable risk data.

DA partially supports clarity, comprehensiveness, and supervisory review

The paper finds partial support only for BCBS 239 Principles 8, 9, and 12. DA records can make a decision report clearer and more useful to supervisors, but they remain decision-specific.

• Principle 8: partial support because a DA record is a clear decision report under the five-minute/stranger test.
• Principle 9: partial support because DA can cover a material risk decision, but multiple records would be needed across risk areas.
• Principle 12: partial support because examiners can use DA records to see why data sources, aggregation methods, or reporting formats were chosen.
• Principles 10, 11, 13, and 14 remain orthogonal.

DA scoring models fall back under SR 11-7

The paper warns that any model used to compute β_W, detect Hollow Wins, or run CDO Dashboard metrics is itself a model under SR 11-7 when it processes data into quantitative estimates.

• A DA scoring model must be documented, validated, monitored, and governed under SR 11-7.
• The decision to use the DA scoring model can be recorded in DA.
• The scoring model itself still needs independent validation and technical documentation.
• This creates recursive governance: DA documents the decision, while SR 11-7 governs the DA measurement model.

SVB and First Republic are used to show missing decision records around model and risk-data choices

The paper uses the 2023 failures as examples of banks with model and risk-data issues where a structured record of the decision process could have exposed assumptions, limitations, dissent, and system welfare effects.

• SVB is described as having internal risk models that flagged rising interest rate risk without a decision mechanism that forced escalation.
• The paper frames this as a non-decision pattern: Mode 2 external conditions should have triggered Mode 1 incident-response decisions.
• First Republic is paired with SVB as a case where model choices, risk-data aggregation, and private-gain trade-offs were inadequately documented.
• DA would not have prevented failure by itself; it would have made the reasoning record inspectable.

The thesis fails if DA duplicates or replaces the bank regimes

The complementarity claim is testable. DA should add a decision-reasoning layer while leaving SR 11-7 and BCBS 239 obligations intact.

• Falsified if a DA record contains all SR 11-7 development, validation, and monitoring documentation.
• Falsified if a DA record contains all BCBS 239 data lineage, aggregation rules, controls, and reporting infrastructure.
• Falsified if a bank claims DA replaces model validation or data accuracy requirements.
• The paper's supervisory proposal is narrower: add DA record review to exams for high-risk model and data-architecture decisions.