Scene summary
- Objective: translate desk intuition about shape and regimes into modelling objects and risk coordinates.
- Products: storage, swing, tolling, hydro, and their dependence on curve structure.
- Key tension: the desk experiences shape and policy boundaries; risk reports compress into “gas factor”.
Transcript
- Location: same meeting room. Whiteboard now half full.
- Time: 14:35.
- Ambient: Lukas has drawn “STORAGE” in a box, then erased it and replaced it with “CURVES”. Anna has started talking faster, because this is where the desk intuition lives.
Anna stands up and points at the board.
Anna: “Okay. Now move to power and gas specifics. Forwards, shape, mean reversion. How do you model these curves?”
Lukas does not answer with “we assume lognormal”. He draws two curves: one for gas, one for power, both with a heavy winter hump.
Lukas: “First: we do not model a curve. We model a market snapshot of many curves and profiles. Power and gas curves are full of structure. Seasonality is not a rounding error. It is the trade.”
He writes in large letters:
LEVEL RISK vs SHAPE RISK
And then underneath:
- Level: everything up/down
- Shape: winter–summer, peak–base, location basis, profile within month
Anna nods quickly.
Anna: “Right. I think in:
- Winter-26 versus Summer-26.
- Base versus Peak inside the season.
- And within a month: profile versus flat.”
Lukas draws three nested layers on the board:
- Delivery buckets: months / quarters / seasons
- Blocks: base / peak / offpeak
- Profile: within-month shaping (weekday/weekend, hourly if needed)
Lukas: “Good. In the library we represent forwards as prices per delivery bucket, but we keep explicit mapping from trade cashflows into those buckets. Otherwise you cannot explain PnL.”
Anna lifts the risk pack.
Anna: “That’s the missing bit. Risk treats everything as ‘gas curve move’. But the loss came from shape. Winter moved relative to summer and it hit my storage and tolling book differently.”
Lukas circles shape.
Lukas: “Exactly. And for structured products, shape is often the first-order risk. A storage deal is basically a portfolio of seasonal spreads plus optionality. A tolling deal is a spread machine with operational constraints.”
He adds a blunt line under it:
Structured PnL ≈ shape PnL + optionality PnL + constraint boundary effects
Anna gestures at him with the marker.
Anna: “Now mean reversion. You always say ‘spot mean reverts’. I need to know what that actually means in your models. And how it impacts decisions.”
Lukas draws a simple timeline: prompt month at the left, Winter-26 far right.
Lukas: “We treat time like two regimes.”
He labels the left side short end and the right side long end.
Lukas: “Short end: day-ahead, week-ahead, prompt month. That part can spike and mean revert. You have weather, outages, constraints. The process is not a random walk. It snaps back towards a structural level.”
Anna interrupts immediately.
Anna: “Yes. One cold week, system tight, prices explode. Then it normalises.”
Lukas: “Exactly. Long end: seasons and years. That behaves like expectations of fundamentals: supply, demand, storage levels, policy, fuel switching. It moves, but it is less spiky. More persistent shifts.”
He pauses, then makes it concrete.
Lukas: “So we do two things in practice:
- We take the forward curve as given for valuation and risk. That’s the tradable object.
- For scenario generation and some structured valuation, we use a dynamic model where short-term deviations mean revert around a seasonal mean curve.”
Anna’s eyes narrow.
Anna: “So for storage and hydro: do you actually simulate spot with mean reversion? Or do you just treat forwards as a set of deterministic prices and slap a vol on top?”
Lukas answers carefully.
Lukas: “Both, depending on the product and the risk purpose.
- For vanilla and many structured approximations: we use the forward curve plus implied vol as the pricing backbone.
- For operational optionality where exercise is truly dynamic—like storage and hydro—we often need a model that produces a path and a policy.”
He writes:
Pricing backbone: (F(t,T) + (T,K)) Decision backbone: stochastic spot/prompt dynamics + constraints
Anna sits back.
Anna: “So the desk intuition is: mean reversion affects exercise timing. If prices revert quickly, waiting is less valuable. If shocks persist, storage optionality increases.”
Lukas nods and draws two cartoons:
- Fast mean reversion: spikes decay quickly → less value in holding inventory
- Slow mean reversion: shocks persist → inventory becomes leverage
Lukas: “Yes. And that is where residual PnL appears when the model changes. If we tweak mean reversion speed or the volatility term structure, the optimal exercise boundary shifts. The mark can jump even if the curve does not.”
Anna points at the board.
Anna: “That sentence is key. Risk thinks in ‘market moved’. But sometimes ‘policy boundary moved’.”
Lukas: “Exactly.”
Convenience yield and the storage logic question
Anna pivots to a concept she knows risk people like to drop into meetings.
Anna: “And convenience yield and storage logic—do you actually model it or just treat forwards as given?”
Lukas smiles slightly.
Lukas: “Convenience yield is a good explanation. It’s a dangerous input.”
Anna raises an eyebrow.
Lukas: “In theory, for a storable commodity, forward relates to spot via storage costs and convenience yield. That’s clean in textbooks.”
He taps the desk.
Lukas: “In practice, we don’t infer true convenience yield and then build our curve. We calibrate to the observed forward curve. Convenience yield becomes a derived interpretation of why the curve looks the way it does.”
Anna tests the practical implication.
Anna: “So if I ask: ‘what drives my storage value’, your answer is not ‘convenience yield’. It’s ‘the forward spread structure and its uncertainty’.”
Lukas: “Exactly. For storage valuation, what matters is:
- the seasonal spread shape in the forward curve (winter premium vs summer),
- the volatility and correlation across months,
- and the constraints: injection/withdrawal limits, costs, and inventory bounds.”
Anna nods.
Anna: “Fine. For me the important bit is: term structure and seasonality are real risks. Mean reversion and storage logic are the intuition behind why gas storage and hydro have big optionality.”
Lukas underlines seasonality and constraints.
Lukas: “And that’s also why your risk outputs should be reported in those terms. Not ‘gas factor’. We should show:
- winter–summer spread delta,
- peak–base spread delta,
- and a policy sensitivity indicator: how much the optimal exercise boundary shifts under small curve or vol changes.”
Anna looks at him.
Anna: “Can you actually produce that last one?”
Lukas nods, but keeps it grounded.
Lukas: “Yes. Not perfectly. But we can produce something operational: run two re-pricings with small perturbations and measure boundary movement and value change. If that metric jumps after a library release, we know we moved the engine under your car.”
Anna smiles. The earlier tension eases.
Anna: “Good. That’s the bridge I need between your world and mine.”
She picks up her risk pack.
Anna: “Next, I want to connect this to VaR and scenario. Risk wants one-day 99%. I want a scenario ladder that matches how we think: winter spike, basis blowout, vol up, and slow reversion regime.”
Lukas caps the marker.
Lukas: “That becomes Scene 4: risk metrics that respect shape and policy. And we’ll be explicit about where linear VaR breaks.”
Key takeaways
- In power and gas, shape is often first-order risk, not a second-order detail.
- Mean reversion is a decision variable in disguise; it shifts exercise timing and policy boundaries.
- Residual PnL can be “policy boundary moved”, not “market moved”.