Building a stress scenario narrative: the story before the numbers

What a scenario narrative is#

A stress scenario narrative is a written account of what is happening in the world during the stress: what starts it, how it spreads, how long it lasts, how authorities and markets respond, and what that means for the institution’s customers and counterparties. It is written in prose, before any parameter is chosen, and every number in the scenario is then derived from it.

The order is not stylistic. A parameter set assembled first and given a narrative afterwards is a set of movements someone thought looked severe, retrofitted with a story — and retrofitted stories are exactly what a reviewer is trained to detect.

Why the story comes first#

Writing the story first also changes who can participate. Very few board members can meaningfully challenge a table of variable paths, and most will not try.

Nearly all of them can challenge a paragraph that says a regional slowdown reduces demand for the commodity that half the corporate book depends on, that the currency weakens as export receipts fall, and that the government responds by tightening. That is a discussion about the institution’s actual exposures, conducted in language the people accountable for those exposures can use.

The narrative is, in practice, the only part of a stress test that a governing body can genuinely own.

The five parts of a narrative#

A complete narrative has five parts.

1. The trigger: the identifiable event or turn that starts the sequence, specific enough to be argued with.
2. Transmission: how the trigger reaches this institution, which is the section that separates a scenario about the world from a scenario about the bank.
3. Duration and shape: whether the stress is a sharp shock with a recovery, a prolonged grind, or a shock followed by a second one before the first has passed — a distinction that matters more than peak severity for a multi-year capital projection.
4. Response: what authorities, markets and competitors do, including the responses that make things worse.
5. Second-round effects: what happens once the institution’s own reaction, and everyone else’s, feeds back into conditions.

Severity belongs in the narrative#

Severity belongs in the narrative rather than in the parameter table, and it should be argued in terms of what the institution is willing to survive. The productive framing is not "how bad could it get" — an unanswerable question that invites either timidity or theatre — but "what is the worst set of conditions under which this institution intends to remain viable without extraordinary support?"

That is a statement of appetite, it has an owner, and it converts cleanly into parameters afterwards. It also gives the board something it can decline: a severity proposed and rejected as too mild is evidence of a functioning process, and its absence over several cycles is evidence of the opposite.

Build it from this balance sheet’s concentrations#

The scenario must then be built from this institution’s concentrations, which is the step template packs cannot perform. The method is mechanical enough to be checked: list what the balance sheet is actually concentrated in — sectors, single names, geographies, currencies, collateral types, depositor groups, funding channels, and the operational dependencies that would stop the bank functioning — and confirm that the narrative disturbs the largest of them.

A scenario that leaves the institution’s three biggest concentrations untouched will produce a comfortable result for a reason that has nothing to do with resilience.

Where narratives break#

Internal consistency is where narratives most often break, and the failures are systematic enough to check against a list.

• Variables that move in economically impossible combinations: sharp currency depreciation with no inflation, deep recession with rising employment, collapsing property prices with unchanged household default rates.
• Policy responses that no authority would take in the conditions described, or the absence of any response at all through a multi-year crisis.
• Market prices that fall without any change in liquidity or bid-offer.
• Peers who behave as though nothing is happening while the institution alone is under strain.
• And the most common of all: a narrative describing a severe multi-year downturn alongside a balance sheet projection in which volumes, margins and business mix continue along plan.

From narrative to parameters#

Translating narrative into parameters is a documented act of judgement, not a lookup. Each variable path should carry a written reason tying it to the story — this currency path follows from the export shock described in paragraph two; this deposit attrition follows from the confidence effect described in paragraph five.

Where a historical episode informed the calibration, it should be named along with what was adjusted and why, since no past episode maps onto a present balance sheet unaltered.

The record matters because the reviewer’s question is never "is this number right?", which has no answer, but "how did you arrive at it?", which does.

Who writes it#

Authorship should be deliberately mixed.

A narrative written entirely inside the risk function tends to be economically tidy and commercially naive; one written entirely by the business tends to be well observed and insufficiently severe.

The workable arrangement is that risk drafts, the business lines challenge the transmission and the assumed customer behaviour, treasury challenges the funding and market assumptions, and someone with no stake in the result reads it for coherence.

Where a scenario touches a specialist area — a particular commodity, a particular counterparty structure — the specialist should be in the room, because the transmission section is where a generalist quietly invents things.

Two tests before a reviewer runs them#

Two simple tests catch most defects before a reviewer does.

1. Read the narrative aloud to someone who runs a business line and ask whether they recognise their customers in it; if they say the story is about a different bank, the scenario will not be defended when its results are challenged.
2. Read it against the balance sheet and ask what it fails to disturb; the answer is the scenario’s blind spot, and it should be a deliberate choice recorded in the documentation rather than an oversight discovered later.

Neither test requires a model, and both are more diagnostic than another decimal place in the parameter table.

What a good narrative ends up doing#

A narrative worth the effort ends up doing more work than the exercise it was written for.

• It gives the board a severity decision it can take rather than ratify.
• It gives the modellers a reason for every path they implement.
• It gives validation something to challenge other than arithmetic.
• It gives the recovery plan a description of the conditions its options must work under.
• And it gives the institution, a year later, an answer to the question that closes most stress-testing reviews: not what the numbers were, but why anyone believed them.