Physical Climate Risk: Hazard, Exposure, Vulnerability — and Time

The 2012 IPCC Special Report on managing extreme events introduced a framing that has since become the operational definition of physical risk: the product of hazard, exposure, and vulnerability. Insurers have been using the components — usually under different names — since long before that report. What the IPCC did was formalise the relationship and add the variable most actuarial frameworks treat as a rounding error: time.

Hazard is the physical event — the flood, the cyclone, the heatwave. Exposure is what is in the path of that event — the policyholders, the properties, the infrastructure, the crop under cover. Vulnerability is how badly those exposures fare when the event occurs — the structural integrity of the building, the socio-economic capacity to recover, the catastrophe cover terms on the reinsurance tower. Multiply the three together over a defined period and you get the expected loss. This is the risk equation every actuary works with in some form.

The time dimension is what separates a static catastrophe model from a climate-aware one. Hazard frequency is shifting. Exposure distributions are shifting. Vulnerability is shifting. All three are moving at once, and each is moving for different reasons and on different timescales. A physical-risk model that uses historical averages for all three is modelling a world that no longer exists.

flowchart TB
    H[Hazard:<br/>Climate drivers<br/>frequency and intensity]
    E[Exposure:<br/>What is in the path<br/>policyholders, assets, systems]
    V[Vulnerability:<br/>Susceptibility of<br/>the exposure to harm]
    T[Time:<br/>How each variable<br/>evolves over horizons]

    H --> R[Physical Risk]
    E --> R
    V --> R
    T -.evolves.-> H
    T -.evolves.-> E
    T -.evolves.-> V

    style H fill:#0A0F1E,stroke:#00C2CB,color:#F0F0F0
    style E fill:#0A0F1E,stroke:#00C2CB,color:#F0F0F0
    style V fill:#0A0F1E,stroke:#00C2CB,color:#F0F0F0
    style T fill:#0A0F1E,stroke:#C9A84C,color:#F0F0F0
    style R fill:#0A0F1E,stroke:#C9A84C,color:#F0F0F0

1. Hazard

Hazard is the component most people recognise as “climate risk”. It is the frequency and intensity distribution of harmful events — cyclones, floods, droughts, wildfires, heat stress, hail. For an insurer, the relevant hazards are the ones mapped to the book: KwaZulu-Natal convective storms for a short-term insurer, Western Cape drought for an agricultural portfolio, coastal storm surge for a J-Bay homeowners book.

The time dimension here is well studied by climate scientists and reasonably well communicated. Forward-looking projections from CMIP6 and the NGFS scenario set provide distributions of how hazard frequency and intensity shift under defined emissions pathways. The work for the actuary is translating those physical-science outputs into the specific peril set that matters for the book — which usually requires a vendor catastrophe model, a downscaling step, and a view on the scenario to be used.

The risk is mechanical: hazards the historical record underweights. A one-in-fifty-year flood event in a warmer, wetter climate may be a one-in-twenty-year event by 2040. A standard-formula catastrophe loading that assumes stationary frequency is wrong by a factor that compounds every year.

2. Exposure

Exposure shifts for reasons that have almost nothing to do with climate science and everything to do with demographics, urbanisation, and economic geography. In South Africa, coastal property values in J-Bay, Ballito, and the Cape Town Atlantic Seaboard have grown well above CPI for a decade; exposure in those locations has compounded in rand terms even before any change in hazard. The policyholder base shifts too: more semigration to coastal towns, more second-home ownership in fire-prone fynbos interface zones.

For an insurer, the operational question is how often the exposure dataset is refreshed, and at what granularity. A catastrophe model run on last year’s TIVs is modelling last year’s book. For fast-growing coastal and peri-urban segments, the gap between the modelled exposure and the actual exposure can be material within a single underwriting cycle.

The time dimension on exposure is the one most amenable to direct action: an insurer can update exposure data quarterly, can geocode at a parcel rather than postal-code level, and can flag accumulations against hazard footprints. The vulnerability and hazard components are harder to influence. Exposure discipline is within reach.

3. Vulnerability

Vulnerability is how much damage a given hazard produces in a given exposure. It is socio-economic as well as physical. A formal-sector commercial building with compliant fire protection has different vulnerability from an informal settlement dwelling exposed to the same flood depth. A smallholder farmer without irrigation has different drought vulnerability from a commercial operation with groundwater rights. The vulnerability component is where climate risk meets inequality — and where parametric and microinsurance products find their opening.

The time dimension cuts both ways. Proactive adaptation — flood defences, building-code enforcement, early-warning systems, crop diversification — reduces vulnerability over time. Unchecked growth of informal exposure in hazard-prone locations increases it. South African municipal finance constraints mean adaptation investment is unevenly distributed; insurers with books exposed to municipalities that are unable to fund adaptation are exposed to a slow-moving vulnerability deterioration that no annual repricing cycle fully captures.

The actuarial implication: vulnerability assumptions need to be scenario-dependent. A vulnerability curve calibrated on 2015 damage data for the 2027 book is assuming no change in either the exposure mix or the underlying adaptation capacity. Both are moving.

4. Time — the Fourth Variable

The temptation in physical-risk modelling is to treat each component as static and layer a deterministic trend on top. That is what most capital-model climate overlays do today: a fixed uplift factor applied to the baseline loss. It is defensible as a first pass. It is inadequate as a long-run framing.

A more honest approach decomposes the time dimension by component. Hazard evolves on a physical-science timescale driven by emissions pathway and climate sensitivity. Exposure evolves on a demographic and economic timescale driven by regional growth and migration. Vulnerability evolves on a policy and adaptation timescale driven by infrastructure investment, building codes, and insurance penetration. Under NGFS scenarios these three evolve consistently; under real-world conditions they can diverge substantially.

For IFRS S2 disclosure, the time dimension is not optional. The standard requires forward-looking information on climate risks across short-, medium- and long-term horizons. That cannot be produced from a single historical catastrophe model. It requires decomposing physical risk into its components, parameterising each against defined climate pathways, and reporting the scenario-dependent sensitivity of the insurer’s expected losses.

The practical deliverable is a physical-risk dashboard that shows, for each major book, the current expected loss decomposed into hazard, exposure and vulnerability drivers, and the projected evolution of each over the planning horizon. The dashboard will be wrong. Every scenario-based output is wrong in specifics. Its value is directional: it tells the underwriting, pricing, and capital functions which of the three variables to prioritise as the book evolves. Without the decomposition, physical climate risk remains a single opaque line item in the scenario analysis appendix. With it, the insurer has a running view of which component of risk is moving and why.