Eight Definitions, One Cast

The Examiner’s Question · RBI’s MRMF 2026, read like evidence. By Dr. Anandkumar Prakasam

The Clause

Paragraph 7 finishes what episode one started. Around the definition of model itself, the draft defines a cast. A model approver, who sanctions use. A model developer, who builds. A model owner, who is accountable for the model through its life. A model validator, who challenges it independently. And around the cast, the draft defines the stakes. Model risk includes not just the risk of a wrong model but the risk of a right model used wrongly, and the risk that a model stops suiting the conditions it was built for as time passes. Decommissioning gets a definition of its own, because retiring a model is a governed act, not an expiry. So does explainability, the degree to which a model’s workings can be understood and articulated.

What It Actually Asks

Definitions in a regulatory instrument are not vocabulary. They are a staffing plan. The moment approver, owner, developer and validator are defined terms, an examiner can ask for names against each of them, per model, and can test the boundaries between them. Was the validator independent of the developer. Did the approver understand what was being sanctioned or merely chair the meeting where it happened. Is the owner someone who can actually answer for the model, or a department mailbox. The time-suitability limb of the risk definition quietly imposes a duty of periodic re-examination, because a risk defined as arising over time cannot be managed by a one-time approval. And the explainability definition arms every later clause that conditions autonomy on understanding. Together the eight definitions convert what most institutions run as a project pipeline into a set of accountable roles that survive the project’s end.

The Incident File

In January 2012 JPMorgan’s Chief Investment Office adopted a new Value at Risk model for its synthetic credit portfolio. The bank’s own Task Force report, published in January 2013, reconstructs how. The model was conceived and largely operated by a single quantitative analyst. It ran through a series of Excel spreadsheets that had to be completed manually, values copied and pasted between workbooks. One formula divided by the sum of two rates where it should have used their average, which cut the volatility estimate roughly in half. Approval came under time pressure, from a review function that flagged concerns and then signed anyway. The operational result was that the office’s risk limit alarms went quiet exactly when the portfolio was growing most dangerous. The trades it permitted became the London Whale losses, more than six billion dollars. Ask the paragraph 7 questions of that history. Who was the developer. The same person as the operator. Who was the validator. A reviewer approving under deadline. Who was the owner once the analyst moved on. The report has no clean answer, and that absence is the finding.

The Trap

The paper compliance move is an organisation chart exercise. Titles get created, a RACI matrix gets filled, and the same three people appear in every box with different hats. The draft’s definitions are satisfied in name and defeated in structure, because independence is a property of incentives and reporting lines, not of labels. The second trap is the ownerless middle age. Models acquire owners at launch, when there is budget and attention, and lose them silently at the first reorganisation. An examiner who picks a five year old model from the inventory and asks to meet its owner will, in most institutions today, watch the question ricochet between departments for a week. That ricochet is the evidence.

The Fix

Treat every definition as a record with a name, a date and a signature. The pattern is a per-model role register kept as part of the inventory entry, with a rule that no model may operate a day without a named, current owner and no validation may count if the validator reports into the developing function. Add succession by default. Ownership transfers are governed events with handover notes, the way key person risk is handled everywhere else that matters. The time-suitability limb gets its own instrument, an assumption register per model, which later episodes will develop, because the draft returns to it in the lifecycle chapter. The book behind this series calls the underlying discipline decision ownership, and its chapter on it argues the point this episode’s incident proves, that the most dangerous state a model can reach is not wrongness but orphanhood.

The Examiner’s Question

Pick the oldest high-materiality model in your inventory and ask its named owner three questions. What assumptions is it built on, when were they last checked, and what would make you switch it off. If the owner cannot answer all three without a meeting, you have a name on a form, not an owner.

Commentary on a draft under consultation, not legal advice. Paragraph numbers may change at finalisation.