Catastrophe Model Blending: Build Your Own View of Risk

Catastrophe Model Blending: Building Your Own View of Risk Before Buying Cat Cover

Last reviewed: November 2026

Catastrophe model blending is the practice of combining the loss outputs of two or more catastrophe models, using explicit weights, into a single exceedance probability curve. It is how a cedent turns several vendor opinions into its own view of risk.

That view then drives the biggest outwards decisions of the year: the cat XoL retention, the top limit and how hard to push on price. If the blending method is wrong, every one of those decisions inherits the error.

This guide is for chief actuaries, heads of catastrophe modelling, CROs and heads of outwards reinsurance. It explains how blending works, why the method matters as much as the weights, and how to present a blended view that reinsurers will take seriously.

Key Takeaways

• Catastrophe model blending combines vendor model outputs with explicit, documented weights to form the cedent's own view of risk.
• Frequency blending averages probabilities at a given loss, while severity blending averages losses at a given return period.
• The two methods can produce materially different 1-in-200 losses from the same models and the same weights.
• Model weights should come from validation evidence, not from which answer makes cover cheaper.
• Regulators expect firms to own and validate any vendor model they use, including the blend.
• A transparent blended view gives cedents a stronger position at renewal than a single vendor output.

What Is Catastrophe Model Blending?

Catastrophe model blending combines two or more modelled loss distributions into one curve that represents the cedent's view. The inputs are usually event loss tables or year loss tables from vendor models, sometimes alongside an in-house or open-source model.

The output is a blended occurrence exceedance probability (OEP) curve, an aggregate exceedance probability (AEP) curve, or both. Those curves then feed cat XoL structuring, capital models and risk appetite metrics.

Blending is different from adjusting. Adjustments change one model's output, for example by loading for non-modelled perils or poor exposure data, while blending decides how much each model counts.

Most mature programmes do both. They adjust each model for known weaknesses first, and then blend the adjusted outputs.

Why Do Cedents Need Their Own View of Risk?

Cedents need their own view because every vendor model is one opinion about hazard, vulnerability and financial loss. Buying cover against a single opinion leaves the cedent exposed to that model's blind spots.

Vendor models can disagree substantially on the same portfolio. Differences typically come from event catalogues, hazard intensity assumptions, vulnerability curves for local construction, and how secondary uncertainty is treated.

A single-model approach also creates version risk. When a vendor releases a new version, the cedent's modelled 1-in-200 can move sharply overnight, and its reinsurance strategy moves with it.

An own view of risk smooths that volatility and forces a documented judgement. It also changes the renewal conversation, because the cedent can explain why its view differs from the reinsurer's rather than simply accepting the reinsurer's model.

For the data and platform side of building that capability, see our guide to catastrophe modelling infrastructure.

How Does Frequency Blending Differ From Severity Blending?

Frequency blending averages exceedance probabilities at a fixed loss amount, while severity blending averages loss amounts at a fixed return period. They answer subtly different questions and can give different tail results.

Frequency blending is equivalent to treating each model as a possible "true" world, chosen with probability equal to its weight. Severity blending is closer to averaging the experts' answers to the same return period question.

Neither is wrong in principle. The danger is choosing one by habit, or switching between them from year to year without disclosure.

How Much Does the Blending Method Change the Answer?

The blending method alone can move the 1-in-200 loss by more than ten percent, even with identical models and weights. The example below shows how.

Worked Example: Same Models, Same Weights, Different 1-in-200

The illustration below is simplified and uses USD millions of occurrence loss. It assumes two vendor models, with Model A weighted 60% and Model B 40% after validation against the cedent's own event history.

Model outputs and severity blend:

Severity blending gives a 1-in-200 loss of 240. A cat programme with a top limit of 240 would appear to protect to 1-in-200.

Frequency blend at the same loss levels:

Exceedance probabilities between table points are interpolated linearly for simplicity.

Summary of the two methods:

Both answers use the same models and the same 60/40 weights. The only difference is whether the cedent averaged losses or averaged probabilities.

The lesson for cedents: the blending method is a material assumption, not a technical detail. If the board's risk appetite is "protect to 1-in-200", the cedent must state which blended curve that refers to, and test the top limit against both.

The lesson for reinsurers: when a cedent presents a blended 1-in-200, ask for the method, the weights and the individual model curves. A severity-blended view can make a programme's exhaustion point look more remote than a frequency-blended view of the same models.

How Should You Set Model Weights?

Model weights should be set from documented evidence about how well each model represents the cedent's portfolio. They should not be reverse-engineered from the reinsurance budget.

The evidence usually falls into five areas:

1. Back-testing against own losses. Compare each model's modelled loss for historical events with the cedent's actual indexed losses for those events.
2. Hazard credibility for the region. Review each model's event catalogue, hazard data and recent scientific updates for the main perils.
3. Vulnerability fit. Check how each model treats local construction, occupancy, year built and insurance terms such as deductibles and sub-limits.
4. Treatment of secondary perils. Assess whether flood, storm surge, convective storm, wildfire and fire following are modelled, partially modelled or ignored.
5. Stability across versions. Note how much each model has moved between versions and whether the changes were explained.

Weights can differ by peril and region. A cedent might give one model more weight for windstorm and another more weight for earthquake, as long as the reasoning is recorded.

Weights should change only through a governed model change process. If weights move every year in the direction of cheaper cover, reinsurers and supervisors will notice.

When Should You Blend, Adjust or Use a Single Model?

The right approach depends on how much the models disagree, how much evidence the cedent holds and how material the peril is. The table below maps common situations to the approach that usually fits.

These are starting points rather than rules. The final approach should be tested against the cedent's event history and reviewed by someone independent of the reinsurance purchase.

Which Blending Mistakes Distort Cat Buying?

Six mistakes cause most of the problems with blended catastrophe views. Each one can push a programme's limit or retention in the wrong direction without anyone noticing.

How Should Cedents Present a Blended View to Reinsurers?

Cedents should present the blended view alongside the underlying model outputs, with the method, weights and adjustments stated plainly. Reinsurers will run their own models, so the goal is to narrow the debate rather than win it with a single number.

What Should the Submission Include?

Which Questions Will Reinsurers Ask?

• Which model do you weight most heavily for each peril, and what evidence supports it?
• Is your blended 1-in-200 based on frequency or severity blending, and what does the other method show?
• How did your actual losses from recent events compare with each model's estimate?
• What loads do you apply for secondary perils and non-modelled exposures?
• How much of the year-on-year change comes from the portfolio, and how much from model versions?

Treating the renewal as a scenario discussion rather than a single-number negotiation helps here. Our article on treaty renewal scenario rooms covers how to connect exposure change, claims and capital in one view.

How Do Regulators and Lloyd's Treat a Blended View?

Regulators allow vendor models and blending, but they expect the firm to own the result. A blended view is a modelling choice that must be governed, validated and explained.

Solvency II and Solvency UK

Article 126 of the Solvency II Directive states that using a model or data obtained from a third party does not exempt a firm from the internal model requirements. The PRA Rulebook carries the same principle in its external models and data chapter, and requires firms to monitor limitations arising from external models.

In practice, that means the blend itself needs validation, documentation and a clear model change policy. A firm cannot point to the vendor's own validation as a substitute.

Lloyd's

Lloyd's Principles for Doing Business expect managing agents to define and maintain an appropriate catastrophe view of risk methodology. They also expect risk-based validation of that view and a materially complete representation of catastrophe risk.

Changes to the cat view of risk, including vendor model version changes, typically fall within Lloyd's model change governance. Syndicates should check the current Lloyd's guidance for their own reporting obligations.

Rating Agencies and Internal Stakeholders

Rating agency capital models also rely on modelled catastrophe losses, and agencies typically ask how the company's own view compares with vendor outputs. The exact treatment varies by agency, so confirm current criteria directly.

Internally, the board's risk appetite statement should specify which blended curve its catastrophe tolerances refer to. Otherwise the appetite and the reinsurance strategy can quietly measure different things.

What Does the Market Look Like for 2027 Renewals?

Catastrophe pricing has softened from the peaks that followed the 2023 reset, and capacity has returned. Softer markets put more pressure on terms, retentions and model assumptions than on headline rate.

Secondary perils remain the area of greatest model disagreement and of greatest scrutiny. Severe convective storm, flood and wildfire continue to drive questions about whether vendor models capture current hazard and vulnerability, a theme explored in our piece on how climate change acts as a reinsurance multiplier.

In this environment, cedents with a well-governed blended view can argue for lower retentions or better terms with evidence. Cedents relying on one model's output are more likely to accept whatever the reinsurer's model says.

Checklist for Chief Actuaries and Heads of Cat Modelling

Use this checklist before your blended view is used to set retentions, limits or capital.

• Exposure data has been cleansed and quality-scored before any model is run
• The blending method (frequency or severity) is chosen deliberately and applied consistently to OEP and AEP
• Model weights are supported by back-testing and hazard and vulnerability reviews
• Loads for non-modelled perils, inflation and data quality are documented and applied in a stated order
• The cat programme's top limit has been tested against both blending methods
• Model versions, weights, adjustments and exposure snapshots are version-controlled together
• The board risk appetite states which blended curve its catastrophe tolerances refer to
• Treaty layers, retentions and limits derived from the view are held in a controlled system

For the last point, a centralised reinsurance treaty management platform keeps the treaty terms that flow from your view of risk in one controlled place.

Key Terms

Catastrophe model blending: Combining outputs from two or more catastrophe models, using explicit weights, into one view of risk.

Frequency blending: Averaging each model's annual probability of exceeding a given loss. It is equivalent to a weighted mixture of the models' distributions.

Severity blending: Averaging each model's loss at a given return period. It is equivalent to a weighted average of the models' quantiles.

OEP (occurrence exceedance probability): The annual probability that the largest single event loss in a year exceeds a given amount. It drives per-event covers such as cat XoL.

AEP (aggregate exceedance probability): The annual probability that the total of all event losses in a year exceeds a given amount. It drives aggregate covers, earnings and capital metrics.

Year loss table (YLT): A simulated set of years with the event losses in each year. Mixing YLTs by weight is a practical way to frequency-blend.

Secondary uncertainty: The uncertainty in the loss for a given event, around the model's mean damage estimate.

Own view of risk: The cedent's documented, validated catastrophe loss view after adjustments and blending, used for decisions in place of any single vendor output.

Sources

• Directive 2009/138/EC (Solvency II), Article 126: use of external models and data
• PRA Rulebook: Solvency Capital Requirement – Internal Models, External Models and Data
• Lloyd's Principles for Doing Business: Catastrophe Exposure