Scenario Design Granularities: Narrative, Factors, Geographies, Time

Climate scenario analysis fails most often not because the scenarios are wrong, but because the granularity is wrong for the decision. A 30-year systemic scenario run at country level cannot inform a postal-code underwriting appetite change. A high-resolution physical hazard run at property level cannot inform a 15-year capital plan. The scenario is only as useful as the granularity match.

The NGFS catalogue, the IEA pathways, the PRA’s biennial exploratory scenarios — they are built at specific granularities for specific purposes. Using them for a different purpose, without adjusting granularity, is where most scenario work breaks.

1. The Six Granularity Dials

Scenario design has six granularity dials that can be tightened independently:

Scenario narrative. The story that holds the scenario together. At the loosest end, “orderly transition” or “hot house”; at the tightest, a year-by-year policy, geopolitical and technology narrative that answers specific questions about sequencing.

Climate factors. Physical variables — temperature, precipitation, sea level, extreme event frequency — linked to climate research. Resolution ranges from global mean temperature to gridded high-resolution physical hazard.

Broad economic factors. Macroeconomic variables — GDP, inflation, interest rates, unemployment, exchange rates — derived from the macro model linked to the climate narrative. Resolution ranges from global aggregates to country-level paths.

Sectoral granularity. Economic output, emissions and technology deployment broken out by sector. At the loosest, three or four sectors; at the tightest, NACE level-3 or equivalent with sub-sector technology splits.

Firm granularity. Corporate-level impact on counterparties — credit risk, underwriting exposure, investment valuation. Ranges from sector proxies to name-level forward-looking metrics.

Activity granularity. Impact at the level of specific economic activities — a policyholder’s operations, a borrower’s physical assets, a specific insured location. The tightest granularity, where physical hazard meets individual exposure.

2. Finer Is Not Always Better

There is a temptation to push all six dials tighter. Finer is not always better. Tightening granularity multiplies data requirements and computational cost. It narrows the confidence interval on any individual number and widens the number of numbers that need to be produced, calibrated, governed and explained.

The pragmatic question is always: what decision is the scenario supporting, and what granularity does that decision require?

flowchart TB
    A[Scenario Purpose?] --> B[Regulatory ORSA<br/>loose granularity]
    A --> C[Strategic Planning<br/>medium granularity]
    A --> D[Book Underwriting<br/>tight granularity]
    A --> E[Product Pricing<br/>very tight granularity]

    B --> F[Country-level<br/>macro factors<br/>Sector aggregates]
    C --> G[Country-level<br/>Sector detailed<br/>Narrative explicit]
    D --> H[Sub-national geography<br/>Activity level<br/>Physical hazard grid]
    E --> I[Address-level<br/>Activity level<br/>Year-by-year]

    style A fill:#0A0F1E,stroke:#C9A84C,color:#F0F0F0
    style F fill:#0A0F1E,stroke:#00C2CB,color:#F0F0F0
    style G fill:#0A0F1E,stroke:#00C2CB,color:#F0F0F0
    style H fill:#0A0F1E,stroke:#00C2CB,color:#F0F0F0
    style I fill:#0A0F1E,stroke:#00C2CB,color:#F0F0F0

3. A Worked Example: Tightening Geography for a Homeowners Book

Take a South African homeowners insurance book. The insurer is revisiting underwriting appetite in coastal and flood-exposed postal codes following the KZN floods of 2022 and the repeated Cape flooding of 2023-2024.

Running NGFS scenarios at national level produces a single temperature and rainfall path for South Africa. This is useless for the decision — the decision is about postal-code pricing and coverage terms.

Tightening geography from country to province gets closer but still conflates Western Cape coastal exposure with Karoo drought exposure. Tightening again, from province to district or postal code, lets the scenario actually discriminate between a property in an elevated Johannesburg suburb and a property in the Cape flats flood line.

But tightening geography to postal code while leaving sectoral granularity at national aggregates creates a different problem: the macro context (labour market, inflation, insurance affordability) is invariant across postal codes, which is clearly wrong. So the geography dial pulls sectoral and firm dials with it.

The data cost escalates accordingly. Postal-code rainfall projections require downscaled climate models. Postal-code affordability proxies require census and economic data at the same resolution. This is why most scenario analysis stops at district or province level in practice — the data either does not exist or is too expensive to acquire and maintain.

4. Calibrating the Dials to the Decision

A working rule I use with clients:

• Regulatory ORSA scenarios — loose granularity across all six dials is usually acceptable. NGFS or equivalent, country-level macro, sector aggregates. Regulators want to see that the methodology exists and the governance is in place more than they want postal-code precision.
• Strategic business planning scenarios — medium granularity. Country-level macro is fine; sector detail matters because the question being answered is which lines of business grow or shrink; narrative must be explicit because boards make decisions on stories, not gradients.
• Book-level underwriting scenarios — tight on geography, sector, activity. This is where the physical hazard work matters. Narrative may be kept relatively simple because the decision is a three-year underwriting appetite, not a 30-year thesis.
• Product pricing scenarios — very tight on geography and activity; narrative can be suppressed to a single pathway because the decision is about relative pricing. Frequency and severity distributions at policy level.

5. The Over-Engineering Risk

The opposite failure mode, less commonly discussed, is over-engineering scenarios beyond what the decision warrants. This shows up in actuarial teams who build high-resolution physical hazard infrastructure to answer a regulatory disclosure question that could be answered with NGFS references and narrative text.

The symptoms are familiar: multi-month build cycles, heavy vendor contracts for catastrophe data, sophisticated geospatial tooling, and at the end a one-page ORSA chapter that any external reviewer could have produced with public scenarios in a week.

The right check before building: what changes in the decision if the scenario is more precise? If the answer is “nothing” or “we still end up at the same underwriting appetite”, the precision is not paying for itself.

6. Granularity As a Governance Topic

Once you accept that scenarios have six granularity dials, calibration decisions become a governance topic in their own right. The risk committee should approve the granularity matrix for each scenario use case — not because the risk committee cares about downscaled climate models, but because granularity choices are inputs to materiality, to scenario credibility, and to how much budget is being allocated to scenario analysis rather than other risk management priorities.

That conversation is not usually happening at the board level today. When it starts happening, scenario analysis moves from a compliance exercise run by the climate team to an operational capability owned by the whole risk function. Which is where it needs to end up.