A practitioner-first framework for governing quantitative and machine-learning models used in investment decision-making.
This repository contains a comprehensive Model Governance Framework covering:
- model classification and lifecycle control,
- declared operating context requirements,
- evidence and validation standards,
- change management and scope preservation,
- deployment approval and capital allocation controls (governance, not execution),
- monitoring, escalation, and retirement,
- roles, accountability, and exception handling.
It is written for investment firms, risk committees, and ML platform teams who need governance that is operational, auditable, and enforceable—not aspirational.
This framework exists to prevent a common failure mode in model-driven investing:
Models are validated under one context and quietly used under another.
The framework provides a structured way to prevent this drift while enabling responsible, context-aware use of models at scale.
The framework enforces:
- Contextual integrity (validity is conditional),
- Scope control (no implicit promotion or ungoverned reuse),
- Evidence discipline (claims cannot exceed tested conditions),
- Risk proportionality (governance rigor follows impact),
- Transparency and accountability (clear ownership and audit trails).
- Investment committees and fiduciaries
- Risk management and model risk functions
- Quant research teams
- ML/AI platform engineering teams
- Compliance and operations partners (where applicable)
- Model Governance Framework (PDF)
Model_and_Governance_Framework.pdf(primary artifact)
Optional supporting artifacts you may add later:
- Summary/checklist extracts
- Templates (Model Context Summary, Deployment Approval Checklist)
- Worked examples mapped to real model types
This framework explicitly distinguishes three model categories:
- Diagnostic models: informational evidence only (not deployable claims)
- Translational models: implementability analysis under constraints
- Deployable models: can influence capital; highest governance burden
It also prohibits:
- implicit scope expansion,
- retroactive justification,
- single-metric "best model" approval,
- capital scale-up driven solely by short-term performance.
A practical adoption path:
- Treat this framework as the starting policy baseline.
- Define your firm's Model Registry and ownership model.
- Require a Declared Operating Context for any model referenced in decisions.
- Apply classification (Diagnostic / Translational / Deployable).
- Use the checklists and worked examples as governance templates.
- Iterate through controlled amendments (do not weaken controls via ambiguity).
This framework is provided for adaptation and internal use by investment firms and governance teams.
If you find it useful, attribution is appreciated but not required. When referencing this framework externally, please cite:
Colin Alexander, CFA, CIPM — Model Governance Framework
This framework grew out of practical experience applying quantitative models in institutional settings where clarity of intent, operating context, and accountability are essential.
The goal is to enable organizations to use models more effectively—by making assumptions explicit, preserving decision integrity across time and regimes, and supporting informed human judgment alongside increasingly capable systems.
If you have seen MandateOS, it is a reference implementation of governance-first, mandate-bound judgment.
This framework is broader: it governs model lifecycle and deployment across a firm, independent of any
specific software system.
This repository provides a governance framework template for educational and internal policy development purposes. It does not constitute legal, compliance, or investment advice. Firms should adapt it to their regulatory environment and internal control standards.