1. What risk measures actually show up on a desk?

Typical daily stack for a power/commodities trading book:

  • VaR / Expected Shortfall → “How bad is a bad day (or worse) under a model or history?”

  • Stress & scenario P&L → “What happens if X happens again (e.g. 2008, 2022), or if curves move in specific ways?”

  • P&L explain / P&L attribution → “Why did we make or lose money yesterday?”

These are complementary:

  • VaR/ES: model/historical distribution of risk.
  • Stress/scenarios: designed shocks – non-statistical, narrative.
  • P&L explain: post-mortem → checks if risk numbers and booking make sense.

2. VaR (Value at Risk) – desk intuition

Definition (1-day, 99% VaR):

The loss threshold that should only be exceeded on about 1 in 100 days under the model.

Formally:

  • 1-day P&L = $ $ from today to tomorrow.
  • Sort P&Ls; VaR is the 1st percentile (large loss in the left tail).

Flavours:

  • Historical VaR
    • Re-use actual past market moves (e.g. last 1,000 days),
    • Revalue today’s portfolio for each move,
    • Look at loss quantile.
  • Parametric VaR
    • Assume P&L is (multi)normal (delta-normal),
    • VaR = z-score × volatility × position size.
  • Monte Carlo VaR
    • Simulate risk factors from a model,
    • Revalue portfolio over many paths.

Power/commodities flavour:

  • You must model forward curve moves, shape factors, spot/forward basis.
  • A flat “one price, one vol” VaR is meaningless; shape is where the book lives.

3. Stress tests and scenarios

VaR is built from past data or a model. But you also want:

  • Historical stresses:
    • “Replay 2022 gas crisis, 2008 oil collapse, 2017 cold snap.”
  • Hypothetical scenarios:
    • “Winter +10°C colder than normal and low hydro,”
    • “Gas +30%, power forwards +50€ on winter, +10€ summer,”
    • “Curve massively flattens / steepens.”

Stress numbers answer:

“If this specific shock hits tomorrow, what is the P&L on today’s book?”

In systems:

  • Scenario engines shock risk factors on a snapshot:
    • Parallel bumps, slope changes, calendar spreads, basis, FX, etc.
  • Revalue portfolio and report scenario P&L and driver breakdown.

On a power desk:

  • Important static scenarios:
    • Winter strip up/down,
    • Spark spreads widening/narrowing,
    • Location/basis blow-outs,
    • Shape shocks (winter vs summer, peak vs base).

4. P&L explain / P&L attribution

Daily question on a trading floor:

“We made/lost X; was that because of the market, time decay, new trades, or something broken?”

P&L explain decomposes actual daily P&L into buckets, for example:

  • Risk factor P&L (predicted):
    • Yesterday’s Greeks × yesterday→today market moves,
    • e.g. delta × dS, vega × dVol, curve bucket sensitivities × rate/fwd changes.
  • Carry / accrual / theta:
    • Time decay of options,
    • Roll-down on curves (today’s 1Y is tomorrow’s 364d, etc).
  • New business / trade changes:
    • P&L from entering/exiting or changing positions.
  • Model / parameter / data changes:
    • Recalibration,
    • Curve rebuilds, fixing corrections.
  • Residual / unexplained:
    • Anything not fitting the above; should be small relative to total P&L.

Good control principle:

“Model risk P&L is the residual. If residual grows, something is wrong with booking, curves, or the risk model.”

On a structured power/gas book:

  • P&L explain must capture:
    • Curve level and shape moves,
    • Changes in optimal exercise (e.g. storage/swing decisions),
    • Parameter changes (vol, correlation etc.).

5. How they connect

  • VaR / ES:
    • “Normal day risk” under your approximations,
    • Used for limits and capital.
  • Stresses / scenarios:
    • “Edge cases and known disasters”,
    • Used for limits, board reporting, risk appetite.
  • P&L explain:
    • “Does the risk model match reality?”,
    • Frontline control; flags issues fast.

If VaR says “small risk” but stresses show huge loss and P&L explain has massive residuals, you don’t trust VaR.

6. Minimal quantitative mental picture

Think of a small portfolio with P&L:

ΔPV≈ΔPVrisk+ΔPVnew+ΔPVresidual. \Delta \text{PV} \approx \Delta \text{PV}^{\text{risk}} + \Delta \text{PV}^{\text{new}} + \Delta \text{PV}^{\text{residual}}.

  • For VaR, you look at distribution of $ ^{} $.
  • For stress, you evaluate $ ^{} $ under specific shocks.
  • For P&L explain, you compare actual $ $ vs predicted $ ^{} + ^{} $ and call the difference residual.

7. Interview soundbites

You can keep it as:

“On a power/commodities desk we usually have VaR/ES for day-to-day distributional risk, plus a stress and scenario framework to test curve level, shape, and basis shocks. Daily P&L explain then checks whether risk-factor sensitivities actually reproduce observed P&L, with residuals used as a control signal for booking and model issues.

For structured products, the risk is shape- and spread-driven, so the factor set for VaR and scenarios must include seasonal buckets and spreads like winter–summer and spark/dark spreads, not just a single price or parallel curve move.”