Scene 5 — Risk reports: VaR, stress, scenarios, sensitivities

1) The unifying trick: common risk factors

Lukas: “They force everything into a common factor set. Typically:

• gas forwards by month,
• power forwards by block (base/peak),
• CO₂ futures,
• FX,
• vol buckets.”

Anna: “So all products become deltas and vegas to those buckets.”

Lukas: “That’s the goal. But there’s a difference between producing a sensitivity and that sensitivity being stable enough to net. For path-dependent products, the sensitivity is policy-dependent.”

Anna points at an aggregated delta chart.

Anna: “This says we are net flat Winter-26 gas delta.”

Lukas reads the fine print.

Lukas: “Net flat under today’s policy and today’s snapshot. If you move the curve and the exercise boundary flips, the net flat statement can become false in a stress. That’s what we meant yesterday by regime shifts.”

Anna: “So netting can be dangerous.”

Lukas: “It can be falsely comforting.”

2) Netting: when offsets are real versus cosmetic

Anna scrolls to a section showing net exposures by month.

Anna: “So a long swing and a short vanilla can offset in the same forward bucket?”

Lukas: “Yes. That is the netting logic: sum deltas by bucket. The total factor exposure drives VaR and explain.”

Anna does not let it pass.

Anna: “But my swing delta is not linear. In tight markets it becomes much more directional. In loose markets it behaves like optionality.”

Lukas: “Exactly. The netting is clean for vanillas. For flex, the netting is conditional.”

He points to a small chart, buried deep in the report.

Lukas: “See this? Delta distribution, not just delta point estimate. They have a hint of it. But they present you one number on the front page.”

Anna: “So the front page is a simplification.”

Lukas: “Yes. Sometimes acceptable. Sometimes not.”

3) Scenario analysis: the desk-friendly layer, if done right

Lukas clicks the scenario tab.

Lukas: “Scenario analysis is usually the most actionable piece for a desk.”

He reads a few lines.

Lukas: “They take a base snapshot, apply shocks, revalue. Examples:

• ‘Power Winter-26 +20 €/MWh’
• ‘Gas +30%, Power +10%, CO₂ +10%’
• ‘TTF–PEG basis widens’ Then they reprice the whole portfolio.”

Anna nods.

Anna: “This is closer to how I think. But the shock set is generic. I want scenarios that match our book: winter shape, peak–base, low inflows, slow reversion.”

Lukas agrees and reframes it as governance.

Lukas: “A flex book needs book-specific scenario ladders. Generic scenarios miss the boundary effects. We should propose a ladder like:

• Winter level ladder (±10 / ±20 / ±40)
• Winter–summer spread ladder
• Peak–base ladder
• Vol up / skew steepening
• Basis blowout
• Cold + low inflows combined stress And crucially: reprice with policy recomputed, not frozen.”

Anna underlines policy recomputed on her printed copy.

4) Stress testing: where the politics live

Anna flips to a page titled 2022 Gas Crisis Replay.

Anna: “Here we go. They love this one.”

Lukas reads the assumptions.

Lukas: “Historical stress is usually: replay factor moves from a historic window. Hypothetical is: invent a coherent disaster.”

Anna: “What’s the trap?”

Lukas answers with two traps risk teams often miss.

Lukas: “Trap one: mapping. 2022 had structural breaks: liquidity, margin, bid–ask, constraints. If they only replay mid curves and vols, they understate real loss.”

Anna: “And trap two?”

Lukas: “Trap two: portfolio evolution. In 2022 you would have changed hedges, reduced risk, been stopped out, faced collateral calls. A static replay assumes you sit still. It’s useful as a worst-case mark-move number, but it’s not a realistic trading path.”

Anna looks satisfied. This is defensible and not combative.

5) VaR and ES: what the report actually computes

Anna points at the VaR page.

Anna: “Now VaR. Explain it in one sentence I can defend.”

Lukas gives her the sentence, then the caveat.

Lukas: “VaR is: given a model of one-day factor moves, what loss level do we exceed only 1% of days. ES is: average loss in that worst 1%.”

Anna: “And how do they compute it here?”

Lukas scrolls.

Lukas: “Looks like historical simulation with a filtered volatility model on factors. So:

• build a set of historical one-day shocks on curves and vol surfaces,
• apply each shock to today’s snapshot,
• revalue the whole portfolio,
• take the 99% quantile for VaR and tail average for ES.”

Anna: “Revalue means full repricing?”

Lukas hesitates, then checks a footnote.

Lukas: “This is the key. They are not fully repricing everything. They’re using a hybrid: linear Greek approximation for most trades and full repricing for a subset.”

Anna sits up.

Anna: “So the flex book might be linearised.”

Lukas: “Parts of it, yes. That explains why VaR can miss policy regime shifts. The approximation assumes small moves around today. Stress breaks that assumption.”

Anna: “Then the VaR number is not wrong. It’s conditional.”

Lukas: “Correct. It’s VaR under local linearisation.”

6) Sensitivities: what drives explain, hedges, and limits

Anna clicks the sensitivities table.

Anna: “And these are the Greeks we were fighting about.”

Lukas: “Yes. They map each trade’s sensitivities into buckets and sum. Those aggregates drive three processes:

1. daily PnL explain ($\Delta \times \text{shock}$),
2. hedge sizing,
3. limit monitoring.”

Anna: “So if the Greeks are unstable, the whole stack becomes shaky.”

Lukas: “Exactly. And the report doesn’t show any trust indicator for Greeks. It assumes stability.”

Anna leans back.

Anna: “That is what I need to fix with you. I want the report to tell me: today’s Greeks are near a boundary; interpret offsets with caution.”

Lukas nods.

Lukas: “We can implement a boundary proximity flag for storage and hydro, plus a Greek noise metric from bump stability.”

7) The forward-looking step: risk measures as optimisation problems

Anna points to the ES page and smiles slightly.

Anna: “So if we take ES seriously, we could pick hedges to minimise it.”

Lukas: “Yes. Formally:

• choose hedge vector (x),
• generate a scenario set of factor moves,
• compute portfolio PnL under each scenario,
• optimise (x) to minimise ES, or enforce chance constraints like ‘loss worse than (X) occurs less than 5%.’”

Anna: “That’s exactly how I think about hedging, but informally.”

Lukas: “And the difference is: we can make it systematic. But only if the scenario engine is credible and policy-aware for flex.”

8) What Anna can say to Risk today

Anna turns the laptop slightly away, thinking about the call.

Anna: “Give me the talk track. Short. Defensible. Not combative.”

Lukas writes three bullets on a sticky note and reads them aloud.

Lukas: “1) ‘We accept VaR/ES as a local tail metric, but for flex books it must be complemented with policy-aware scenarios.’ 2) ‘We want the scenario ladder expressed in hedgeable terms: winter–summer, peak–base, basis, vol and skew, and combined cold plus low inflows.’ 3) ‘We need a stability and provenance layer: snapshot ID, model and calibration ID, and a trust indicator for Greeks. Otherwise netting and explain residuals become misleading.’”

Anna nods.

Anna: “Good. That’s not a complaint. That’s an improvement plan.”

Lukas closes the report.

Lukas: “And it’s aligned with governance. Risk doesn’t want surprises. Neither do you. The only way to avoid surprises is to version the engine and measure stability.”

Anna stands up.

Anna: “Alright. Join the call. If they push ‘why did VaR miss Friday’, you say the part about local linearisation and regime shifts. Calmly.”

Lukas: “Understood.”

Anna: “And no jargon.”

Lukas: “Also understood.”

They walk to the meeting room, not as trader versus quant, but as two owners of the same problem: a number that moved, and a process that must explain it.