Blog

The Models That Price Risk No Longer Recognize the Climate They Are Pricing

Jan Gunash · Co-Founder & CGO · April 27, 2026

The best questions are not always the comfortable ones. Like: what happens to the financial system designed to absorb climate shocks when the data that system is built on stops being a reliable guide to the future? And when the models fail quietly, who bears the cost?

These are not hypothetical questions. The climate insurance gap is widening in real time, and the data problem sitting underneath it is the same one running through every story in this newsletter series. The inputs are getting worse. The decisions built on them are getting more consequential. And the institutions responsible for managing that gap are only now beginning to acknowledge it.

A System Built on the Past

Insurance is fundamentally a data business. Premiums are priced on historical loss distributions. Capital reserves are sized by catastrophe models calibrated to past events. Reinsurance markets transfer risk based on probability estimates derived from decades of claims data. The entire architecture rests on a single assumption: the past is a reliable guide to what is coming.

That assumption is breaking down.

The EIOPA Chairperson said it plainly last week in Bratislava: historical data alone is no longer a reliable predictor of future losses. We are operating in a new risk paradigm (source: EIOPA). Aon's Head of Climate Risk Advisory put it in almost identical terms in the firm's 2026 Climate and Catastrophe Insight report: traditional views of risk, based only on historical experience, are no longer sufficient (source: Aon CCI 2026). These are not activist statements. They are assessments from the institutions that price and regulate the global insurance market.

Global insured losses from natural catastrophes have exceeded $100 billion annually for several consecutive years, growing at roughly 5 to 7% per year in real terms. Aon estimates global economic losses from flooding alone exceeded $42 billion in 2025, contributing to approximately $2 trillion since 2000. Drought caused around $13 billion in economic losses in 2025. Swiss Re projects insured losses could reach $148 billion in 2026 if the year aligns with long-term trends, or $320 billion if major events occur (source: Swiss Re / Carbon Brief).

But the model failure is not primarily about the size of recent losses. It is about the shape of future ones.

Where the Models Are Failing

Traditional catastrophe models rely on historically calibrated statistical distributions, asking what past storm data implies about future loss probabilities. That question made sense in a relatively stable climate. It makes progressively less sense as the climate shifts faster than the historical record can capture.

Research published this year by climate risk firm Reask identifies the specific mechanism. North Atlantic sea surface temperatures have been significantly warmer than long-term averages, increasing the intensity and unpredictability of tropical cyclone tail events. Traditional models that rely on historical data systematically underestimate them because the historical record does not reflect today's environmental conditions. Many risk professionals may be underestimating their capital at risk as a result (source: Reask).

California recognised this explicitly. New regulations now allow insurers to use forward-looking catastrophe models rather than historical loss data to seek price increases. That change was driven by the straightforward observation that historical data from a cooler, less fire-prone California was producing systematically wrong price signals for a hotter, drier one (source: HBS). The same problem exists across every peril and every geography. The models are calibrated to a climate that no longer exists.

The Gap That Results

In the EU, only 25% of natural catastrophe losses have been insured over the past decades. Only 17% of Europeans hold coverage for property damage caused by natural catastrophes (source: EIOPA). The German Insurance Association has warned that premiums could double within a decade. In high-risk regions, insurers are retreating from areas where accurate risk pricing would require premiums that households cannot afford.

In the U.S., the share of homeowners without insurance has almost doubled from 7.4% in 2021 to 13.4% in 2024. Over 1.9 million home insurance policies have been non-renewed since 2018, concentrated in California, Florida, and Louisiana (source: WWF). Only 2.6% of residential properties in U.S. counties receiving National Flood Insurance Program payouts for 2025 flood events were actually covered by NFIP policies (source: Aon CCI 2026). Properties that cannot get insurance cannot get mortgages, leading to collapsing property values and a fiscal burden that falls on already-stretched public budgets.

Globally, half of all economic losses from natural catastrophes remained uninsured in 2025. In Asia, Africa, and Latin America the share is 80 to 90% (source: Swiss Re / Mapfre). Those are also the regions facing the sharpest increases in climate risk. Hurricane Melissa struck Jamaica in 2025 causing losses equivalent to more than 40% of the country's GDP, of which the insured fraction was negligible (source: Aon CCI 2026).

When the Energy Transition Becomes Uninsurable

The insurance data failure is not confined to households. It is now appearing inside the economics of the energy transition itself.

Renewable energy infrastructure is being built at scale in exactly the geographies where insurance markets are under the most strain. Offshore wind farms face intensifying storm exposure. Utility-scale solar faces wildfire and extreme heat risk. Coastal grid infrastructure faces flood and storm surge exposure that catastrophe models are systematically underpricing. The same models failing to accurately price household flood risk are the models underwriting clean energy project financing.

Several offshore wind projects in the U.S. and Europe have cited rising insurance costs as a material factor in delays, financing gaps, or cancellations over the past eighteen months. Insurance costs for offshore wind assets have risen between 30 and 50% in some markets since 2022, driven by reinsurance market tightening and model uncertainty about tail risk (source: Reinsurance News / GCube Underwriting). For projects already operating on thin margins, that increase directly affects whether a project gets financed, at what cost, and on what timeline.

There is also no public database of insurance coverage levels for critical energy infrastructure. Substations, transmission lines, and battery storage facilities carry insurance through bilateral arrangements that generate no public disclosure. The systemic exposure of energy infrastructure to climate risk is essentially invisible until something fails.

Nature connects all three threads. WWF's research found that every dollar invested in nature-based climate resilience can save up to $13 in avoided losses. Wetlands buffer storm surge from coastal power plants. Forests reduce wildfire spread along transmission corridors. But these services are almost never valued in insurance models or integrated into energy infrastructure planning (source: WWF). Without a standardised framework for quantifying what a functioning wetland is worth as protection against a grid substation flood, nature-based solutions remain systematically underfunded and insurance models continue to underestimate what is being lost.

The Data Loop That Makes It Worse

Climate risk models learn from loss data. The better the data, the better the models. But the regions where climate risk is growing fastest have the sparsest loss data, the least insurer presence, and the lowest investment in risk modelling. The same data inequality that characterises the weather forecasting gap examined in this newsletter's March 16 edition is now visible in the financial infrastructure designed to respond to the disasters those forecasts are meant to predict. The insurance gap is simultaneously largest where the data is worst and where the consequences of being uninsured are most severe.

What Is Being Built

The institutional response is beginning to move. EIOPA and the European Stability Mechanism published a joint proposal on April 9 for an EU-wide catastrophe pool, a risk-sharing framework designed to pool private risks across the EU and provide a public backstop for extreme tail events (source: EIOPA / ESM). The proposal requires participating insurers to use forward-looking, climate-adjusted risk models rather than historical loss distributions. EIOPA has also released its CLIMADA open-source catastrophe modelling platform to allow smaller regulators and emerging market insurers to build risk assessment capacity without bearing the full cost of proprietary model development (source: EIOPA).

AI and satellite tools are generating new flood and wildfire exposure datasets for regions where ground-level data has never existed. Parametric insurance products, which pay out based on measurable physical triggers rather than assessed damage, are being deployed in developing markets precisely because they do not require the dense loss history that traditional products depend on (source: Aon CCI 2026 / WWF). These are genuine advances, but they are scaling piecemeal, in the same way that data infrastructure for critical minerals, fertilizer markets, and climate monitoring is being repaired, one institution and one dataset at a time.

The Path Forward

The insurance protection gap is not primarily a market failure. It is a data infrastructure failure that produces market outcomes nobody designed and nobody wants. Insurers cannot price risk accurately without data reflecting current climate conditions. Regulators cannot mandate appropriate capital reserves without models capturing actual exposure. Households, energy developers, and governments cannot plan resilience investments without honest information about the risk they face.

What is needed is the same infrastructure that every other critical data system eventually requires: standardised open-access loss databases covering developing as well as developed regions, forward-looking risk models incorporating real-time climate data, and a governance framework that treats the data underpinning insurance markets as public infrastructure rather than proprietary advantage.

The German Insurance Association's warning that premiums could double within a decade is not a prediction about profit margins. It is a warning about what happens when the data that prices risk stops working and the system compensates by charging more for the uncertainty it can no longer quantify. The households, communities, and clean energy developers that will pay that price are not the ones that produced the uncertainty.

Sources:

Want to contribute your thoughts or findings in data and AI with climate, energy, or nature? Message Jan Gunash on LinkedIn or send an email to jgunash@projectcoeus.com.