Executive summary
- A catastrophe model has four modules: hazard, exposure, vulnerability and financial.
- Primary uncertainty concerns which events occur; secondary uncertainty concerns the loss given an event, and is often the larger component.
- Exceedance probability curves drive reinsurance structuring and capital requirements.
- Model differences between vendors for the same portfolio can be substantial, making multi-model comparison a control.
- The protection gap widens when modelled risk rises faster than willingness or ability to pay, which risk transfer alone cannot resolve.
The four modules
- Hazard: a stochastic event set of physically plausible events with associated frequencies and intensity footprints, extending beyond the historical record.
- Exposure: the insured portfolio with locations, values, construction, occupancy, age and protection characteristics.
- Vulnerability: functions mapping hazard intensity at a location to a distribution of damage ratios for that exposure type.
- Financial: application of policy structures — deductibles, limits, sub-limits, reinsurance — to convert ground-up damage into insured and net loss.
Each module compounds the uncertainty of the previous one. The financial module is the most mechanical and the most frequently mis-specified in practice, because complex policy structures are simplified during data preparation and the simplification is not always recorded.
Primary and secondary uncertainty
Primary uncertainty is about event occurrence: which events happen, where and with what intensity. Secondary uncertainty is about the loss conditional on an event: given a known hazard intensity at a known location, the damage is still a distribution rather than a point.
Secondary uncertainty is frequently underappreciated. Two identical-looking buildings experiencing identical wind speeds can suffer very different damage depending on maintenance, prior weakness, debris impact and post-event conditions. Models that suppress secondary uncertainty produce tighter loss distributions than reality supports, understating the tail on which capital depends.
| Return period | Occurrence exceedance loss | Aggregate exceedance loss |
|---|---|---|
| 1 in 10 | Single largest event loss at this frequency | Total annual loss at this frequency |
| 1 in 100 | Used for reinsurance layer structuring | Used for annual aggregate covers |
| 1 in 200 | Common solvency capital reference point | Reflects multi-event years |
| 1 in 500 | Tail reference for extreme scenarios | Sensitive to secondary uncertainty assumptions |
Capital and reinsurance implications
Occurrence exceedance probability curves inform per-event reinsurance structuring; aggregate curves inform annual aggregate covers and capital held against multi-event years. Solvency frameworks commonly reference a one-in-two-hundred-year annual loss, which places considerable weight on a part of the distribution that is estimated rather than observed.
- Run more than one vendor model where the exposure is material; differences of tens of per cent at key return periods are not unusual.
- Understand which perils and sub-perils are non-modelled, and hold a documented view of the residual exposure.
- Test the sensitivity of the capital number to secondary uncertainty settings, since these are frequently left at defaults.
- Reconcile modelled losses against actual events promptly, and record what the comparison implies about the model or the exposure data.
The protection gap
The protection gap is the difference between economic loss from catastrophes and the portion that is insured. It widens through several mechanisms, and better modelling contributes to some of them: as risk is measured more accurately in high-hazard locations, technical prices rise, and where those prices exceed what households and businesses will or can pay, cover is withdrawn or declined.
- Accurate pricing is a necessary condition for a sustainable market, but it does not by itself make risk affordable.
- Risk reduction — building standards, defences, land-use planning and retrofit — changes the underlying risk rather than redistributing it.
- Public-private structures, pooling arrangements and parametric products can extend availability where conventional indemnity cover has withdrawn.
- Transparency about which locations are becoming uninsurable, and why, gives policymakers information they otherwise lack.
For an insurer this is a strategic question as much as a technical one. Withdrawal from a region is a decision about long-term market position, regulatory relationships and reputational exposure, not only about the current year's expected loss ratio.
Practical example
An insurer runs the same windstorm portfolio through two models. Model A produces a one-in-two-hundred-year aggregate loss of 240 million; Model B produces 310 million. Reinsurance is structured to a 250 million limit.
Under Model A the programme appears adequate; under Model B it exhausts. Neither model is wrong — they differ in event set construction and vulnerability calibration. The appropriate response is not to select the more convenient result but to understand the source of the difference, decide which assumptions are better supported for this portfolio, and document the basis for the structuring decision taken.
Limitations and caveats
- Stochastic event sets extend beyond the historical record and are therefore partly model-generated rather than observed.
- Vulnerability functions are calibrated on limited claims data and may not represent current construction or repair-cost conditions.
- Non-modelled perils and secondary perils represent a residual exposure that is difficult to quantify.
- Climate change undermines the stationarity assumption underlying historically calibrated event frequencies.
- Exposure data quality constrains all outputs regardless of model sophistication.
Conclusion
Catastrophe models are indispensable and imprecise. Used with multiple models, explicit uncertainty settings and honest treatment of non-modelled perils, they support sound capital and reinsurance decisions.
The protection gap they help reveal is not a modelling failure. It is a signal that in some places the answer is risk reduction and public policy rather than risk transfer at a higher price.
References
- IPCC Sixth Assessment Report — Intergovernmental Panel on Climate Change, 2021–2023
- Enhancing the Insurance Sector's Contribution to Climate Adaptation — OECD
- Insurance Core Principles — International Association of Insurance Supervisors
Author bio
Jonas Mohamed Osman Abdelghafour, known as Yonas Osman is an actuary, FRM and financial risk professional specialising in banking, insurance, model risk, capital modelling and quantitative risk management.