Forecast Contracts for U.S. Severe Thunderstorms (Tornadoes, Wind, and Hail)

Severe thunderstorms and their components—namely tornadoes, strong straight-line winds, and hail—can occur in every state, though they are most prevalent in the Midwest and Southeast during spring. The damage caused by these phenomena is substantial and represented the second most frequent type of disaster in the now-discontinued billion-dollar disaster database. Severe thunderstorms are also the weather category responsible for the second-most-insured losses globally, according to SwissRe (in both datasets, hurricanes are responsible for the most damage). 

Rising Losses, Exposure Growth, and Climate Change

Historical data indicate that damages from severe thunderstorms have risen substantially over the past few decades. A recent insurance analysis found that from 1990 to 2022, the inflation-adjusted losses from severe thunderstorms grew about 9% per year. However, the U.S. has experienced rapid development in tornado-prone and hail-prone regions, and over 80% of this upward trend can be explained by exposure growth (more assets of value and higher property values in harm’s way)​ rather than clear changes in storm frequency or intensity​. Analyses of U.S. storm trends find no consistent long-term increase in the occurrence of tornadoes or large hail events once improvements in reporting practices are accounted for. 

Thus, the influence of climate change on driving higher losses seems to be secondary and remains scientifically contested. While warmer temperatures can theoretically enhance atmospheric moisture and instability, other factors (like wind shear) also play critical roles, and no definitive shift in the balance of these elements has been detected​. The bottom line is that the surge in damages is largely due to more targets for storms to hit (expanding bullseye), rather than storms becoming dramatically stronger or more numerous. 

Seasonal Variability and Forecasting Opportunities

Even without a strong long-term climate trend, severe convective storm activity varies greatly from year to year. For example, the U.S. experienced 2,240 tornadoes in 2011, but only 943 two years later, in 2013. This variability is partly associated with larger-scale climate patterns like the El Niño–Southern Oscillation (ENSO) cycle. El Niño years tend to suppress springtime tornado and hail activity, while La Niña years often foster more frequent and intense outbreaks​. Other climate oscillations and antecedent environmental conditions (such as rainfall or soil moisture patterns) can also influence the upcoming severe weather season.

This substantial natural variability, along with its partial predictability, makes severe thunderstorms an interesting use case for forecasting and prediction markets. Because the activity in any given season depends on a mix of atmospheric factors that can be monitored (and sometimes anticipated weeks to months in advance), there is potential for skillful prediction beyond chance. In short, even if climate change is not a dominant factor, informed forecasters can utilize their knowledge to participate in (and inform) a prediction market. 

Severe Storm Forecast Contracts

With these considerations in mind, we introduce severe thunderstorm contracts that will cover tornadoes, hail, and high wind events across the United States. These contracts are based on the frequency of storm reports as tabulated by the NOAA Storm Prediction Center.

Contracts can relate to specific U.S. states as well as the national total, covering both monthly (reflecting the activities of a calendar month) and annual (reflecting a calendar year) periods. For instance, Event Questions might be relatively precise, such as “Will the April 2025 Tornado Count in Oklahoma exceed X?” or they could be broader, like “Will the 2025 Hail Event Count in the U.S. surpass the recent average?”

Baseline Probabilities

To generate baseline probability estimates and initial pricing guidance for these contracts, we employ a straightforward statistical time-series modeling approach using an ARIMA (Auto-Regressive Integrated Moving Average) model trained on the historical record of event counts for each region and time period of interest. Separate ARIMA-based models are developed for each contract’s underlying series. For example, one model is fit to the time series of June tornado counts in Kansas, while another is fit to the series of annual U.S. hail counts, and so forth. These models capture the typical patterns and variability in the data, yielding a baseline forecast distribution that reflects what one might expect based on past occurrences.

Data from NOAA, plotted using Matlab.

For time periods that have already begun and have thus started to accrue data, we augment these predictions conditional on what has transpired in the season thus far in a bootstrap “analogue-year” approach. So, an active start to the 2025 season will increase the odds of 2025 ending above average.

Data from NOAA, plotted using Matlab.

Of course, these baseline forecasts are not intended to be the final word; rather, they serve as an initial anchor, providing a relatively neutral starting point grounded in the historical frequency of severe storms. From there, it is the push-and-pull of the market, informed by all kinds of information, that will refine the probabilities and ultimately produce dynamic forecasts for storm activity. 

Use Cases as Insurance Analogs

Beyond speculation and information discovery, severe thunderstorm forecast contracts have practical applications for risk management. In many ways, these contracts can function like parametric insurance instruments for those exposed to thunderstorm losses. In parametric insurance, payouts are triggered by an objective index (for instance, wind speed) rather than by actual loss claims, allowing for quicker and more transparent settlements. Similarly, a tornado or hail number contract is settled based on event counts, providing a form of hedge against high storm activity. For example, an insurance company that underwrites homeowners’ policies in Oklahoma might be interested in investing in “Yes” contracts for annual tornado counts being above a particular number in Oklahoma in order to offset some of the claims they would need to pay in that scenario. Similarly, an agribusiness conglomerate concerned about hail damage to crops in the Midwest could use hail contracts to financially protect against an unusually severe hail season. Government disaster management agencies, or even regional emergency funds, could also take positions that yield extra funds in the event of an active severe weather season, helping to cover relief expenditures.

It is important to acknowledge basis risk in this setup – the risk that the contract payout does not perfectly correlate with the user’s actual losses. For instance, a company could suffer heavy losses from a single catastrophic tornado hitting a major city, even if the overall state tornado count for the year remains moderate. Conversely, one could imagine a series of small hailstorms that trigger a high count without causing severe damage in aggregate. This mismatch between the index and loss is a classic trade-off in any parametric-like product. However, for users with broad exposure, such as insurers covering thousands of properties across a state – the number of events should serve as a strong proxy for total losses. Statistically, if there are significantly more hailstorms or tornadoes than usual in a region, it is very likely that overall damage will be higher as well. Thus, for large, diversified portfolios of risk, the event-count index correlates well with financial impact, keeping basis risk at a manageable level. 

In summary, the tornado, wind, and hail Forecast Contracts should enable the market to continuously assess the likelihood of severe storm activity and provide a way to transfer or hedge the associated risk. This endeavor invites a broad community, from meteorologists to insurers and investors, to engage with and ultimately improve the forecasting of these impactful events.

—

Originally Posted on May 21, 2025