Scene summary
- Objective: formalise an operating agreement: what the desk needs, what quant needs, and how changes are governed.
- Products: entire structured book (storage, swing, tolling, hydro) plus hedges and risk reporting stack.
- Key tension: avoid future “mystery PnL” by making ownership, provenance, and deliverables explicit.
Transcript
- Time: Friday, 16:00.
- Location: small meeting room, door half open, end-of-week fatigue in the air.
- Ambient: quieter now. The desk has moved on to the next firefight. This meeting is about preventing the next one.
Anna has a notepad open. Lukas has a single slide on his laptop titled Desk–Quant Operating Agreement (Draft). No one calls it that out loud, but that is what it is.
Anna starts without preamble.
Anna: “Let’s summarise what I expect from you, and what you expect from me.”
She counts on her fingers. Not aggressively. Precisely.
1) Anna’s expectations: what good quant support means on a real book
Anna: “From your side, I need five things.”
First finger.
Anna: “First: clear product specs in the library. Payoffs and constraints must match the term sheets. No close enough. If storage has inventory-dependent rates, the pricer must reflect it. If swing has weekly caps, it must be encoded.”
Lukas nods. He knows this is where most long-tail errors are born.
Second finger.
Anna: “Second: consistent, versioned market snapshots and model IDs. I want to know whether a PnL move is market, data correction, calibration drift, or code change. I don’t want mysteries.”
Third finger.
Anna: “Third: stable, interpretable Greeks and risk factors. I need delta, vega, gamma in hedgeable buckets. And I need aggregation across my book that reflects how we actually hedge. Winter is not one thing. Peak is not base. Basis is not a footnote.”
Lukas writes hedgeable coordinates in his notes.
Fourth finger.
Anna: “Fourth: risk analytics I can explain. VaR/ES, scenarios, stress, and PnL explain. Not a black box. If residual is large, it must be diagnosable: data vs model vs mapping vs lifecycle.”
Fifth finger.
Anna: “Fifth: transparent model risk and controls. Tell me assumptions. Tell me limitations. Show me sanity checks. If a number is fragile, label it as fragile.”
She stops counting and looks at him.
Anna: “That’s it. If you deliver those, I can defend the book.”
Lukas does not respond with “sure”. He responds with what he can operationalise.
Lukas: “That’s a clear spec. I can turn it into deliverables and acceptance tests.”
2) Lukas’s expectations: what good desk partnership requires
Lukas speaks in the same structure. He mirrors her style deliberately.
Lukas: “From my side, I need four things.”
First finger.
Lukas: “First: an honest description of how you actually hedge. Which instruments you really trade. Which ones are theoretical hedges that never get done. Where you accept basis risk, and where you do not.”
Anna nods. No argument.
Second finger.
Lukas: “Second: agreement on the risk factor taxonomy. What buckets matter. How granular we need to go. If you hedge Jan/Feb spreads, we should not stop at Winter.”
Third finger.
Lukas: “Third: feedback that is behavioural, not emotional. Not just wrong. Tell me this doesn’t match how the book behaves. Tell me the regime and the symptom. That’s how we debug.”
Anna smiles slightly. She likes that framing.
Fourth finger.
Lukas: “Fourth: discipline around term sheets and booking. If booking conventions drift, the library cannot stay aligned. And when we impose limits in optimisation, I need support. Limits must be formal constraints, not slogans.”
Anna leans back.
Anna: “That’s fair. Traders love soft limits until the day they don’t.”
Lukas nods.
Lukas: “Exactly.”
3) The key operating rule: no silent changes
Anna taps her pen.
Anna: “There is one more operating rule. No silent releases. If a model or calibration change can move marks, I want to know before it hits production.”
Lukas answers immediately.
Lukas: “Agreed. Pre-release impact run against your reference pack. If marks move beyond tolerance, it triggers a review. And the release notes must include: what changed, which books are impacted, and expected directional effects.”
Anna: “Good.”
5) Deliverables: what this becomes next week
Lukas turns his laptop so Anna can see. It’s a short list, not a programme.
Lukas: “Concrete outputs next week:
- Market snapshot and model ID shown on every report page.
- Delta shown at Winter and month level, with a shape section.
- Basis flags between pricing index and hedge instruments.
- Scenario ladder: parallel shift plus shape twist plus vol move.
- A stability card: what changed, and which numbers are fragile.”
Anna reads it once.
Anna: “That’s the right list. It is small enough to deliver, and meaningful enough to change behaviour.”
Key takeaways
- “Quant support” is product correctness, provenance, hedgeable reporting, and diagnosable residuals.
- “Desk partnership” is truthful hedging practice, agreed taxonomy, behavioural feedback, and booking discipline.
- The highest leverage control is governance of change: no silent releases, impact runs, and explicit diffs.