Insight to action: making seasonal hurricane forecasts pay off in risk transfer markets 

By Ian Bolliger, Reask, and Francesco Comola, LGT ILS Partners 

Key points 

• 2026 impacts. Seasonal weather variability carries real, decision-relevant information. 2026’s strong El Niño points to a below-average Atlantic season, yet warm coastal waters keep a live risk of an intense US landfall. 

• Integrating seasonal information. Probabilistic hurricane event sets conditioned to seasonal weather forecasts can be built into systematic pricing and investment strategies that take advantage of seasonal risk variability. 

• Measuring the value. In a 40-year, out-of-sample back-test, a seasonally conditioned ILS strategy delivered up to 30% higher returns, up to 40% higher risk-adjusted returns, and up to 50% smaller drawdowns in high-loss years than a strategy built on a static view of risk. 

In this post, we will put insights from the previous three articles in Inigo’s 2026 Atlantic hurricane season series into practice. But before we get there, a reminder of where we have come. In the first article, Inigo placed 2026 in an unusual corner of the risk map: a strong-to-very-strong El Niño that should suppress Atlantic activity, sitting atop an anomalously warm Atlantic and Gulf that should boost it. Dr. Daniel Swain then explained why a lower storm count need not mean proportionally lower risk, due to a high thermodynamic ceiling on storm intensity. Finally, Dr. Tom Philp described the industry’s asymmetric view of seasonal risk, in which high-risk years trigger alarm and higher rates while low-risk years rarely bring a commensurate decrease. He argued this pattern could raise the cost of cover for insurers and policyholders alike. 

We believe that asymmetry persists partly because the industry has lacked a rigorous, quantitative, validated way to act on seasonal signals, whether those signals push for a higher or lower view of risk. To address that gap, we recently published a working paper that brings the type of seasonal insights provided in this series into a single quantitative investment framework, and we used it to test how far a seasonally varying view of risk can outperform the static view more commonly used across (re)insurance. Taking industry loss warranties (ILWs) — a form of insurance-linked security — as our case study, we set out to answer two questions: 

1. Given the inherent noise in hurricane insured losses, is there enough signal in seasonal risk variability to act on? 

2. If so, how can seasonal forecast information be incorporated into an investment framework to capture that signal? 

The short answer: by using probabilistic event sets conditioned to seasonal oceanic and atmospheric patterns, a strategy investing in a portfolio of ILWs would have substantially outperformed one built on a static risk model over the past four decades. 

The problem with a single season 

Seasonal forecasting can feel futile because of noise. In any year, a single landfalling hurricane can dominate the entire loss experience, and basin-wide named-storm counts swing from as few as two to as many as fifteen against a long-term average of about six. Insured losses are noisier still, because they depend not on how many storms form but on whether one tracks over exposed coastline. 2025 was a classic example: near-average counts, a record share of storms reaching major intensity, yet not a single hurricane landfall in the continental US — and correspondingly low losses on US policies. 

This is why “it only takes one” is such a well-worn cliché, and why many conclude that seasonal forecasts are interesting but not actionable. But a forecast does not have to predict the single outcome to be valuable. It only must shift the distribution of outcomes in a way you can price and act on, consistently, over many seasons. The insurance industry is built to manage uncertainty; it just needs a model for how that uncertainty should change with the state of the ocean and atmosphere. 

Turning the signal into a loss distribution 

The framework we built does exactly this. Reask’s Unified Tropical Cyclone (UTC) model (Loridan and Bruneau, 2025) generates large ensembles of stochastic hurricane seasons conditioned on seasonal weather forecasts. Reask’s Climate-Based Risk Adjuster (CBRA) then reshapes the output of a standard, off-the-shelf catastrophe model — adjusting event frequency, intensity, and landfall patterns — to reflect the distribution of synthetic storms produced for the forecasted climate state of the coming season. The result is a dynamically updated loss distribution. In other words, it is the same underlying catastrophe model, with its events re-weighted each year. Crucially, the re-weighting is symmetric — it pulls the distribution down in suppressed years and up in active ones, with no unintentional excess bias toward caution in low-risk years. Instead, a tunable risk-aversion parameter lets an investor optimize across the portfolio according to their appetite for volatility. 

A 40-year, out-of-sample test 

The heart of the paper is a back-test with no hindsight: a 40-year evaluation of investment performance from 1985 to 2024. Each year, we force UTC with a June seasonal forecast from the European Centre for Medium-Range Weather Forecasts (ECMWF). Critically, the model never sees the weather that materialized that year, nor anything about realized industry losses. 

We then simulated a systematic strategy allocating capital across 36 regional ILWs, comparing allocations guided by the climate-conditioned model against an identical strategy guided only by static climatology. The inputs were all real: loss tables from a widely used commercial catastrophe model, historical insured losses from Property Claim Services, and broker-quoted ILW prices. Across the full four decades, the conditioned strategy consistently outperformed on mean and compound returns, risk-adjusted returns, and drawdowns. 

Improvement of the climate-conditioned strategy over the static benchmark, 1985–2024. Source: Comola et al. (2026). 

The behavior in individual loss years illustrates the model’s approach. When the forecast flagged elevated risk, the strategy pulled back or reallocated, and that cushioned the portfolio when a major storm landed. In 2017 (Harvey, Irma, Maria), with risk elevated across the whole coast, an aggressive static allocation drew down around 40%, while the conditioned version, having shifted toward higher-attaching ILWs, lost roughly 10%. In 2012 the signal was regional rather than basin-wide, so the model moved capital out of the Northeast and into the Gulf. When Sandy struck New Jersey, the conditioned allocation eked out a small gain (+1%) where the static one lost (−7%). Even 2022 (Ian) — a La Niña year in which the model saw higher-than-average risk — came out ahead by trimming exposure (0% versus −10%). 

Seasonal hurricane risk by coastal gate in 2012, shown as a ratio to the long-term view. Blue gates ran below the long-term risk and red gates above it, so the strategy moved capital out of the elevated Northeast toward the suppressed Gulf. This is roughly an inverse pattern to what we see in 2026. Source: Comola et al. (2026), Figure 2c. 

Aggressive-profile annual returns in loss-heavy years. In most, the conditioned model trimmed exposure and cut losses; 1992 is the exception. Source: Comola et al. (2026), Table 1. 

So where does leaning in pay off? Mostly in the many quiet seasons that make up the bulk of the record. In low-risk years, a static view leaves capital under-deployed or inefficiently allocated; the conditioned strategy recognizes the calmer state and allocates more to harvest the available premium. Compounded over four decades, that is where much of the outperformance accrues. For an aggressive risk profile, invested capital ends up well over 100% ahead of the static benchmark by the end of the period. 

This isn’t to say the seasonally conditioned approach always wins. Its edge is statistical, not a yearly guarantee, and 1992 is the reminder. That year the forecast pointed to below-average risk along the entire coast, so the strategy leaned in. But Hurricane Andrew hit South Florida, and the conditioned allocation lost around 25% against the static benchmark’s 10%. In any single season, noise can overwhelm the signal. But the case for conditioning is that acting on a seasonal signal consistently and symmetrically will win in the long run. This of course would imply an alpha opportunity for the investor. But by avoiding the asymmetry bias described by Dr. Philp, it also helps keep cover affordable for the insured. 

What this means for 2026 and beyond 

2026 is a useful test of nuance, because the headline and the detail diverge. With a strong El Niño, the forecaster consensus points to a clearly below-average season — roughly eleven named storms, five hurricanes, and two major hurricanes, against a 1991–2020 climatology of about fourteen, seven, and three. On counts alone, this looks like a season to lean into. 

But, as Dr. Swain pointed out, basin counts are not landfall risk. Using Reask’s forecast-conditioned event sets, a June 2026 outlook puts the probability of at least one Category 4 or 5 US landfall at 27%. This is down from 39% in 2025, but far from negligible. The reduction is uneven: the Gulf of Mexico, where El Niño’s fingerprint is strongest, falls sharply (around 10% versus 19% last year), while Florida moves much less (about 14% versus 19%). The warm subtropical Atlantic is partly offsetting the basin-scale suppression, keeping a meaningful intensity tail alive even in a strong El Niño year. In other words, 2026’s signal varies by geography, and its effect on your portfolio depends on where your exposure sits. 

None of this makes the noise go away. 2026 could still produce a damaging US landfall despite a suppressed basin, as that live Florida tail reminds us. But the right response to irreducible single-season noise is not to ignore the signal. It is to act on it consistently and symmetrically by incorporating quantitative risk forecasts. The 2026 signal points to lower overall frequency, a sharply reduced Gulf threat, and a more stubborn intensity risk along parts of the Florida and southeast coast. A static view sees none of that, while a conditioned view prices each exposure for the season we are currently in. The story of the 2026 season begins with El Niño. We would argue it should go further: toward a more honest, more symmetric, and more quantitative conversation about what to do with the information the weather is giving us. 

Findings are drawn from Comola et al., “Climate-Conditioned Catastrophe Modeling for Dynamic Risk Assessment,” a preprint posted to Research Square in March 2026 (CC BY 4.0) and currently undergoing peer review; the figures above are original visualisations of its results. Anonymised loss tables and ILW premia are publicly available. Ian Bolliger is an employee of and shareholder in Reask, which provides the seasonal event sets and CBRA package used in the study; Francesco Comola is at LGT ILS Partners. 2026 landfall figures are from MS Amlin’s June 2026 forecast, which uses Reask data. For information only; not investment advice. Past performance, including backtested results, is not a guide to future returns.