Scene 2 — Afternoon whiteboard: Start from payoffs, please

1) Inventory: what’s actually in the book

Lukas writes a list.

Lukas: “You trade:

1. Vanilla options on power and gas forwards.
2. Structured: storage, swing, tolling.
3. Hydro-linked optionality in the power book. Plus the hedges around them.”

Anna: “Correct. For vanilla: European calls and puts on delivery months and seasons. For structured:

• Storage: inject/withdraw with min/max rates, total capacity.
• Swing: flexible offtake around a contract quantity, daily/weekly constraints.
• Tolling: fuel-in vs power-out with heat rate, min run, start costs.
• Hydro: dispatch against prices and inflows; water value is a real constraint.”

Lukas writes two columns:

PATH-INDEPENDENT | PATH-DEPENDENT + DECISIONS

He puts vanilla under the first, everything else under the second.

Lukas: “Good. That split is the entire story of why PnL explain and VaR can lie to you if they treat everything like vanilla.”

Anna points with her marker.

Anna: “And the constraints. Every deck glosses over the constraint logic.”

Lukas: “Exactly. Constraints are the product.”

He writes: Structured = payoff + feasible set + decision rule.

2) Vanilla options: your language vs my language

Anna leans back.

Anna: “Can we recap vanilla options the way you think versus the way I trade? Because risk reports read like your world, but hedging is in mine.”

Lukas draws a simple payoff.

Lukas: “In quant language: a European call on a forward has payoff $$\max(F_T - K, 0)$$ Path-independent. Price is discounted expectation under a pricing measure. Greeks are derivatives of that price with respect to the inputs.”

Anna answers immediately in trader terms.

Anna: “In desk language: a call is long upside without owning the forward. A put is insurance. I watch delta and vega because those are what I can hedge.”

Lukas nods and writes:

Desk hedgeables: delta buckets, vega buckets, basis, shape, optionality.

Lukas: “So we should express exposures in the same hedgeable coordinates you use. If the system says ‘vega per tenor’ but you hedge ‘winter shape and skew’, we’re already misaligned.”

Anna taps the board.

Anna: “That’s why Friday hurt. The report told me a story I couldn’t trade.”

3) The conventions that quietly change everything

Lukas draws three boxes labelled Gas, Power, FX/Rates.

Lukas: “Commodities are not one asset class. Conventions differ, and those differences leak into PnL.”

He writes bullets under Gas:

• Gas day vs calendar day
• Location indices (TTF, PEG, NBP, etc.)
• Delivery: month, season, within-month shape

Under Power:

• Peak/offpeak blocks
• Weekdays/weekends
• Shaping and profile risk

Under FX/Rates:

• Discount curve choice
• Collateral / CSA assumptions
• Currency conversion timing

Anna interrupts on purpose.

Anna: “Stop. That first one. Gas day. If your system timestamps wrong, I get phantom PnL.”

Lukas: “Yes. And it can show up as residual because the explain assumes ‘today shock’ applied to ‘yesterday sensitivities’, but the underlying cashflows shifted by one gas day.”

He writes: Residual can be “calendar mismatch”. He circles it.

4) Put–call parity, bounds, and why they matter on a desk

Anna watches Lukas write formulas with mild suspicion. He notices and translates.

Lukas: “You don’t need the measure theory. But you do need the no-arbitrage checks because they’re our first defence against bad market data.”

He writes:

• Parity: calls and puts must be consistent on the same forward
• Bounds: option price must sit within obvious limits
• Monotonicity/convexity: price decreases with strike; convex in strike

Lukas: “If these fail, either the quote is wrong, or we built the surface wrong. And then your Greeks are unreliable. That is exactly how you end up hedging ghosts.”

Anna nods.

Anna: “So the sanity checks are not academic. They are a guardrail for hedge reliability.”

Lukas: “Yes. And when we patch the surface fitter, we should show you an impact report: which deltas and vegas changed and why.”

Anna smiles, but it’s a hard smile.

Anna: “You mean like you should have done before Thursday night’s release.”

Lukas accepts it.

Lukas: “Yes.”

5) Vol surface: trader quotes vs library object

Anna flips her risk pack open.

Anna: “Risk speaks vol. We speak vol. Yet I don’t trust the surface you show. What do you treat as primary: prices or vols?”

Lukas answers quickly.

Lukas: “Primary object is the implied vol surface. But we do not accept raw points. We build a smooth surface with constraints.”

He writes on the board:

Input: market quotes $\sigma(K,T)$ (or delta-quoted vols) Process: fit + enforce constraints + fill gaps Output: surface + stable Greeks

Anna presses on the practical failure mode.

Anna: “And if the fit shifts because you changed smoothing or interpolation, my vega changes and PnL explain residual appears.”

Lukas: “Yes. That is why the vol build must be versioned and reproducible. We need: snapshot ID, surface model ID, calibration dataset ID.”

He writes: Market snapshot is not just curves. It’s curves + vols + fixings + build config.

Anna points to that line.

Anna: “That’s the sentence risk needs.”

6) Transition to structured: why Greeks are not the same thing

Lukas steps back and draws a fork:

Vanilla: price = closed form, Greeks = analytic Structured: value = optimisation over time, Greeks = response of optimal policy

He emphasises the second line.

Lukas: “For storage, swing, hydro, tolling: you are not pricing a payoff. You are pricing a decision problem. The value is: maximum discounted value you can extract subject to constraints.”

Anna nods hard.

Anna: “Yes. That’s what traders mean by optionality. It’s not a formula.”

Lukas continues.

Lukas: “That changes risk. For these products, delta is not ‘dPrice/dForward holding policy fixed’. The policy itself changes when the curve or vol moves. So you get kinked exposures, regime shifts, and instability near constraint boundaries.”

Anna points to the tolling deal.

Anna: “So if power rallies, the plant runs more, which changes fuel burn, which changes gas exposure. That’s endogenous.”

Lukas: “Exactly. And if the risk report linearises it, it will be wrong in stress.”

He writes: Linear VaR + flex book = false comfort.

7) Making it desk-real: what the outputs should look like

Anna takes the marker from Lukas and writes her own headings:

Hedges I can actually do:

• Winter vs summer spreads
• Month-to-month shape
• Location basis
• Peak/offpeak shape
• Option skew (25-delta risk)
• Operational constraints risk (min run/start costs)

She turns to Lukas.

Anna: “This is what I hedge. If your report gives me vega in tenors but not skew, I can’t action it. If it gives me delta by month but ignores peak/offpeak shape, I get surprised.”

Lukas nods and writes underneath:

Quant output should be expressed as:

• bucketed deltas in hedge instruments
• vega in quoted market coordinates (ATM, 25Δ, skew)
• scenario ladders (winter spike, basis blowout, vol jump) with policy-aware repricing
• residual diagnosis flags: snapshot change, model change, mapping change, lifecycle change

Anna caps the marker.

Anna: “Good. Now link this back to Friday. If the residual was driven by a surface rebuild or a policy shift, I want to see it as ‘your implied winter skew moved the storage policy boundary’—not ‘other / unexplained’.”

Lukas: “Agreed. That ‘other’ label is a failure. It’s a design failure.”

Anna’s phone buzzes again. She glances. Doesn’t pick up.

Anna: “One more thing. Risk asked me: ‘Why did VaR not capture this?’ I want a sentence that is true and defensible.”

Lukas writes a single line in the middle of the board, clean and blunt:

VaR missed it because the portfolio is non-linear and policy-driven; linear factor explain underestimates jumps when the optimal exercise regime changes.

Anna reads it twice.

Anna: “That’s usable. Not an excuse. A diagnosis.”

Lukas nods.

Lukas: “And it leads to the fix: policy-aware scenario VaR for flex books, and strict versioning so residuals can be attributed.”

Anna points at the original stack.

Anna: “Next: walk me through how you implement storage and swing in practice. Not the whole PhD. Just: what state variables, what constraints, what calibration targets, and what can break.”

Lukas puts the marker down.

Lukas: “Good. We’ll do storage next. And we’ll end with one thing: a checklist that tells you, before you trade, whether the risk numbers are stable enough to trust.”

Anna nods once.

Anna: “Perfect. Start from state variables, please.”