1. What stress testing actually asks
Stress testing answers:
“If a really bad but plausible market move hits tomorrow, what happens to today’s portfolio?”
Key differences from VaR:
- VaR/ES: looks at a quantile of a model/historical distribution.
- Stress: focuses on specific, severe scenarios, chosen by humans + governance.
Stress = subset of scenarios with:
- Clear narrative,
- Large shocks,
- Governance approval (limits, reporting).
2. Historical vs hypothetical stresses
2.1 Historical stress
Replay a past crisis on today’s portfolio.
Examples (power/commodities flavour):
- 2022 gas crisis:
- Gas forwards +X%, power winter strips up strongly, basis blows out.
- 2008 oil collapse.
- A specific cold winter with tight capacity.
Mechanics:
- Identify a historical window (e.g. worst 10 days in 2022).
- Extract joint moves of relevant risk factors.
- Apply those moves to today’s snapshot and revalue.
Pros:
- Realistic, internally consistent moves.
- Easy to explain: “What if 2022 happens again?”
Cons:
- Past may understate future extremes.
- Market structure/regulation may have changed.
2.2 Hypothetical stress
Construct a scenario that has not happened exactly, but is plausible.
Examples:
- “Severe cold winter + low gas storage + low hydro levels.”
- “Sudden CO₂ spike + outage of key nuclear fleet.”
- “Regime change: price cap removed + strong demand rebound.”
Mechanics:
- Start with a story (drivers and mechanisms).
- Translate story into joint shocks:
- Gas/power CO₂ curves, shape, spark/dark spreads, basis, FX, vols.
- Check consistency with:
- Physical constraints (merit order, capacity),
- Historical ranges (not completely crazy).
Pros:
- Tailored to current vulnerabilities.
- Can encode new regimes.
Cons:
- More subjective; requires expert judgement.
3. Coherent stories and factor moves
A good stress scenario is not “+50 everywhere”.
It has:
- Narrative driver(s):
- Weather (cold/heat),
- Supply shock (pipeline, plant outages),
- Policy/regulation,
- Macro (recession, embargo).
- Mechanism:
- How does this driver move:
- Gas prices and term structure,
- Power prices (especially winter/peak),
- CO₂, FX, basis, spreads,
- Volatilities.
- How does this driver move:
- Consistent factor moves:
- If gas spikes, power should react consistent with merit order.
- Winter strip likely moves more than summer in a cold-weather story.
- Basis shocks at constrained locations.
Risk + quant role:
- Map narrative → joint shocks in factor space.
- Ensure no internal contradictions (e.g. spark spreads imply plants losing money while the story says they run flat out).
4. How stress testing links to systems
In an analytics framework:
- Start from base snapshot of risk factors.
- For each stress:
- Construct shocked factor vector $ f^{} $.
- Build a stressed snapshot.
- Revalue all trades → stressed PVs.
- Compute:
- Stress P&L = PV_stress − PV_base per:
- Trade, book, desk, entity.
- Stress P&L = PV_stress − PV_base per:
- Aggregate and report:
- Worst stresses per book.
- Contributions by product, factor, region.
Link to governance:
- Board and risk committees approve:
- List of stresses,
- Loss limits per stress,
- Escalation rules when breaches occur.
5. Interview soundbite
You can summarise like this:
“Stress tests are scenario analyses focused on severe but plausible shocks. Historical stresses replay episodes like the 2022 gas crisis using the original joint factor moves. Hypothetical stresses start from narratives – such as a cold winter with low storage – and translate them into coherent shocks on gas, power, CO₂, and basis, consistent with merit-order logic. In the system, we apply these shocks to a market snapshot, revalue the portfolio, and aggregate P&L by desk and factor to check that losses under these stories stay within risk appetite.”