Scenario libraries: why '2008 again' isn't a plan

Everyone's favorite stress test

"What happens if we get another 2008?"

It's the most common stress test request. Also incomplete.

A 2008 replay tells you how your current book would've performed in 2008 conditions. That's useful as a diagnostic. But it's not telling you much about your forward-looking risk, and in my experience people conflate those two things more than they realize.

The problem with pure replays

Starting conditions are different. Vol surface, rate environment, credit spreads, none of it matches September 2008 anymore. Your portfolio wasn't built for those correlations either. And market structure has moved on: liquidity dynamics, counterparty exposure, the algo flow ecosystem.

There's a subtler issue too. Historical replays give you false precision. You get a single P&L number and it feels authoritative. "We'd lose 14.3% in a 2008 repeat." That number is wrong. Not because the math is bad, but because the premise is fiction.

A replay is a sanity check. Treating it as a plan is where things go sideways.

What forward scenarios actually need

Building real forward scenarios is harder and nobody likes doing them because you have to state your assumptions out loud.

Market shock: [SPX -30% over 4 weeks]
Vol response: [VIX peaks at 65, term structure inverts]
Correlation shift: [Equity/bond correlation goes positive]
Liquidity impact: [Bid-ask widens 3x in options, 5x in credit]
Sequencing: [Initial shock → policy response → secondary move]

That sequencing line matters more than anything else in the block. The order in which things break determines where you actually get hurt, and two scenarios with identical endpoint moves can produce wildly different P&L paths depending on whether the vol spike or the correlation shift comes first.

Building something usable

A decent scenario library needs a few layers. Historical anchors first (2008, 2020, 2022) because they're useful for backtesting and they give you common reference points when communicating with stakeholders.

Then forward narratives. Specific hypotheses about what could break and how. Not "recession" but "US IG spreads blow out 200bps over six weeks while equity vol stays suppressed for the first two." Get specific or don't bother.

Sensitivity variants pull a lot of weight too. Take the same shock and run it with different vol responses. Same directional move but different speeds. You learn more from the deltas between variants than from any single scenario in isolation.

Then tail combinations. Multiple bad things happening simultaneously. These feel unrealistic when markets are calm. They feel obvious in hindsight.

The goal isn't prediction. It's exposing the assumptions your portfolio is quietly relying on.

The governance question

Scenarios that can't trigger action are academic exercises. Before investing time in a library, nail down the basics. What would you actually change if a scenario started looking plausible? Who has the authority to act without convening a committee? How often does the library get refreshed?

There's an organizational problem buried in here. The person building the scenarios is rarely the person with authority to act on them. So you end up with this disconnect where the quant team produces good analysis that sits in a shared drive while the CIO makes decisions on gut feel.

Bridging that gap is a governance problem, not an analytics problem.

Where to start

Stop asking "what if 2008 happens again." Instead: what are your most concentrated exposures right now? What market conditions would actually stress each one? And what would you do if you started seeing early signals?

If you want help building a scenario library for your portfolio: Contact