Fair Value Weather & Climate — Monday, August 10, 2026

A detailed daily look at some of the most interesting weather and climate prediction markets at ForecastEx and elsewhere.

• 1.25°F, ForecastEx ranks 1st of 12. Average error on daily highs, last 7 days, 5:00 PM ET day before through 5:00 AM ET. Aviation Forecast next best with an average error of 1.48°F
• 13°F, Largest error, San Francisco high. Settled 70°F against 83°F from the European model
• 11.8×, Best return on daily temperatures yesterday, net of fees. No on the Detroit high above 83° at 8¢, priced 31 hours before it resolved and 22 hours before the high itself happened, settled 83°.
• 6°F, Widest disagreement for Tuesday, Philadelphia. ForecastEx below the National Weather Service forecast on the high
• 30%, At least one major Atlantic hurricane in August. A Yes contract at 30¢ pays 3.3× the money put in, net of fees
• Access to live probabilities and the ability to trade is available via IBKR Prediction Markets.
• General primers on how these prediction markets work are available in the IBKR prediction markets FAQ (scroll down) and the ForecastEx FAQ.

3) How ForecastEx temperature markets compared with the weather models on Sunday

Below is a scorecard for the daily temperature forecasts as they stood at 5:00 PM ET on Saturday, August 8th for the target day of Sunday, August 9th. The grid below orders its cities by mean absolute error (MAE) on the daily highs, which is the average size of the miss in degrees, ignoring whether it was too warm or too cold.

ForecastEx is scored against eleven public forecast tools throughout this letter, all listed with links in the FAQ at the end. The grid below pulls out four canonical tools for illustration: the American model, the European model, the National Weather Service, and the Aviation Forecast.

The largest errors on Sunday were in San Francisco, where the five systems averaged 6.7°F on the high. Across all listed cities, the American model finished the day with an average error of 3.9°F on the highs, against 1.7°F for ForecastEx, and its misses ran in both directions, 8 degrees too warm in San Francisco and 7 degrees too cold in Los Angeles.

Postmortem on the San Francisco miss. The high reached 70°F in Sunday San Francisco, whereas all five systems in the grid above were at least three degrees warmer, with the European model 13 degrees warmer. 

The day before, the National Weather Service office in San Francisco reported that the marine layer was about 1,600 feet deep and was forecast to compress to around 500 feet by Sunday morning, but that if a deeper marine layer than forecast persisted, it would suppress the high. It appears that the marine layer was compressing and retreating until about noon, but at that time winds swung back onshore, and the temperature fell for the rest of the day (black observed line vs. the purple dashed Aviation Forecast line below).

Forecast accuracy and its utility

The standings below measure overall forecast accuracy, indicating that ForecastEx has been the most accurate tool of the 12 publicly available tools tracked here over the past 7 days.

This constitutes ongoing evidence that weather prediction markets like ForecastEx may be the most accurate short-range weather forecasts available, and the difference becomes greater as lead time decreases.

Daily weather forecasting is one of the most mature, established, and scientifically principled fields of science and industry. It is not an exaggeration to describe conventional weather forecasting systems as the frontier of applied physics, statistics, and computer science, resting on decades, if not centuries, of scientific inquiry.

As such, conventional weather forecasts should hardly represent a target ripe for a novel forecasting system to improve upon. Despite the formidable challenge, ForecastEx prediction markets seem to be doing just that, not just echoing public forecasts but aggregating dispersed information in a way that improves upon sophisticated established systems.

This is possible because prediction markets are not substitutes for standard forecasts but rather sit downstream of them. They incorporate conventional forecast information as one input and convert it into refined probabilities through direct financial rewards for being accurate and direct financial penalties for being inaccurate. This creates a dual effect of attracting accurate individuals and systems into the market while deterring those who are inaccurate. People or systems that consistently make poor forecasts are heavily motivated to either improve or leave the market.

This improvement in accuracy is valuable for a variety of reasons. For example, wholesale electricity is bought and sold a day ahead in regional markets, and the primary cause of daily variations in electricity demand and prices is temperature.

The figure below pairs two cities’ daily mean temperature with the daily electricity demand of the region serving it, defined by the Energy Information Administration, and with the day-ahead price at the region’s trading hub, over the past 30 days.

In the two examples above, the temperature-electricity demand correlation runs from 0.89 in Houston to 0.94 in Dallas, both positive because in the cooling season a hotter day means more electricity demand.

The dots past the divider are Tuesday’s implied forecast from the ForecastEx market and the demand and price that temperature implies through the fitted relationships over the past 30 days. Tuesday indicates 4 percent above the 30-day average demand in Dallas at $30 per megawatt hour, and 1 percent above it in Houston at $30 per megawatt hour.

Improving forecast accuracy is critical, but just as critically, if not more so, the structure of these contracts allows for direct hedges for industries exposed to high prices. Buying Yes contracts on lower-probability temperature extremes entails disproportionately large return multiples that would compensate for the associated disproportionately large electricity price spikes.

On average, no public tool beats ForecastEx consistently, but the trade board below asks the narrower question of where a single tool could potentially be used to identify edge in particular circumstances.

The most undervalued contracts for Sunday were

• Access to live probabilities and the ability to trade is available via IBKR Prediction Markets.
• General primers on how these prediction markets work are available in the IBKR prediction markets FAQ (scroll down) and the ForecastEx FAQ.

The most undervalued contract above was No on the Detroit high above 83 degrees, 8 cents 31 hours before it resolved and 22 hours before the high itself arrived at 2:53 PM, for an 11.8 times return net of fees. The twenty most undervalued contracts split evenly between Yes and No positions, and the National Weather Service was the closest public tool on three of them, more than any other system. 

2) ForecastEx temperature market forecasts and prices for Tuesday

The widest gap on the map above is the Philadelphia high, where ForecastEx sits ~5°F below the National Digital Forecast Database from the National Weather Service. The National Weather Service office in New Jersey wrote about an approaching cold front, so the winner of this disagreement will come down to the strength and timing of that front.

• Access to live probabilities and the ability to trade is available via IBKR Prediction Markets.
• General primers on how these prediction markets work are available in the IBKR prediction markets FAQ (scroll down) and the ForecastEx FAQ.

If the tool that was closest for each market on Sunday proves exactly right, the most undervalued contract on the high side is No on the Detroit high above 80 degrees at 11 cents, an 8.7 times return net of fees, with the National Blend of Models predicting 80. On the lows it is No on the Miami low below 85 degrees at 10 cents, a 9.5 times return net of fees, with the American model predicting 85.

Temperature Market of the Day, Los Angeles

The highlighted market for Tuesday is Los Angeles (KLAX)‘s daily high temperature contract. The figure above runs through the end of Tuesday, showing two independent hourly forecasts, each model’s forecast high and low for the contract day, and the most recent high ladder at right.

Thirteen models span 16.1°F on Tuesday’s high, from the European model at 89.2 down to the American model at 73.1, and the middle half of them sit 2.9°F apart. The European model is the outlier that creates most of that range, sitting 4.8°F above the next-highest model, and without it the remaining twelve span 11.3°F. The National Weather Service office serving Los Angeles describes a more robust marine layer expanding into most coastal valleys by tomorrow, so as with recent days in Los Angeles and as with San Fransico described above, the high will depend on how far inland the marine air holds through the afternoon.

3) Atlantic hurricane forecast contracts, three areas to watch

The National Hurricane Center’s Tropical Weather Outlook highlights three areas to watch across the basin this morning. 

An area of low pressure a few hundred miles south of the Cabo Verde Islands, shows a 60 percent chance of cyclone formation over the next week.

Additionally, a tropical wave midway between West Africa and the Lesser Antilles has a 10 percent chance of cyclone development over the next seven days, as does a trough of low pressure northwest of Bermuda, which would not threaten the US. 

The highest-priced landfall market on the map above is Florida, where the probability of a major hurricane landfall stands at 10.2%, or 11 cents for the Yes contract. The probability of a Category 4 United States landfall (by November 30th) stands at 8%, which represents the seasonal forecasts and is not influenced by any current storm. The same question trades at Polymarket at around 18 cents. The season’s two named storms, Arthur at 46 mph and Bertha at 58 mph, both peaked as tropical storms.

The probability of a major hurricane making landfall within 50 miles of Miami-Dade, Florida currently stands at around 5%. Thus, a purchase of a “Yes” at 5 cents is an analog to parametric insurance, where the occurrence of a disaster based on a simple weather parameter, with no claims process and no need to prove a loss, would pay out, in this case 18X of the purchase price.

Two named storms have formed against the 2.2 the season would normally have produced by now if the season finishes at the total forecast by the hurricane research team at Colorado State University.

Frequently asked questions

What is a weather prediction market?

A market in which each contract pays one dollar if a stated weather outcome occurs and nothing if it does not, so the contract’s price is the market’s probability of that outcome. ForecastEx lists these on daily high and low temperatures at individual weather stations, on Atlantic named storm and hurricane counts, and on major hurricane landfall by location, among many others.

What are the four canonical forecast systems compared here?

All four commonly referenced forecast systems rely on physical numerical weather prediction models that solve the equations of the atmosphere forward in time. ECMWF (the European model) and GFS (the American model) are shown here in their raw form. Raw physical models are flexible and can handle weather situations they have never seen (they are not explicitly trained on historical data but rather adhere to the laws of physics), but forecasts apply to a relatively large discrete grid box that contains the weather station rather than the single point where the station sits, causing them to carry systemic biases relative to the stations.

The Aviation Forecast is a model output statistics system. It compares historical forecasts from physical models with what actually occurred and uses the errors to statistically correct systemic biases at the level of individual weather stations. The tradeoff is that a statistical fit is anchored in past situations and bends less readily to a genuinely unusual one. The National Weather Service’s National Digital Forecast Database is undergirded by the National Blend of Models but layers on human forecasters’ judgment, which is more flexible in dynamic weather situations but introduces subjective judgment (more on all of this here).

What are all the forecast tools used here?

• National Weather Service — the official public forecast, human judgment layered on model guidance; that judgment is subjective and can lag a fast-changing day.
• Aviation Forecast — statistically corrected to each station and strong inside a day; reaches only about 25 hours and is anchored in past situations.
• National Blend of Models — averaging many models cancels much of their individual error; a blend can be slow to commit when the models genuinely split.
• European model — consistently among the most accurate global physical models; raw grid-box output still carries station-level bias.
• American model — a global physical model updated four times a day; prone to systematic warm or cold stretches in conditions it resolves poorly.
• German model — a modern global physical model with strong mid-latitude performance; the same grid-box limitations as its peers.
• German statistical model — station-calibrated output built on multiple physical models; anchored in past relationships between forecast and outcome.
• Canadian model — an independent global physical model whose errors differ usefully from the American and European models; raw station-level accuracy is middling.
• UK model — a long-developed global physical model; coarser at United States stations than the domestic systems.
• French model — an independent global physical model that adds diversity to the panel; raw station-level accuracy trails the leaders.
• Japanese model — an independent global physical model; often the weakest of the panel on the daily-high standings, which is itself informative about model diversity.

How does a ForecastEx ladder become a single forecast temperature?

ForecastEx does not publish a single forecast temperature. It lists a ladder of contracts at different strikes, and each price is the ForecastEx probability that the day’s extreme passes that strike. Comparing ForecastEx with the other systems therefore requires converting that ladder into a central estimate. The 50 percent level is the ForecastEx implied median, so where two adjacent listed strikes bracket that crossing, the central value is interpolated between them. Where the ladder does not bracket 50 percent, because every listed strike sits far in or far out of the money, no central value is recorded and ForecastEx is left unscored rather than extrapolated beyond the quoted ladder. That reflects how the strikes happened to be listed rather than anything about forecast quality, so it should count neither for nor against ForecastEx.

Further Reading

Prediction market basics

• The Value of Climate Prediction Markets — The case for why prediction markets can represent the best collective knowledge about our energy and climate future.
• Daily High Temperature Markets at ForecastEx — how the daily temperature contracts were built and settled when they were first listed (not all information is up to date and an update is forthcoming).

Temperature markets and energy

• Energy expenditure implications of next-day temperature forecasts — what day-ahead temperature uncertainty costs utilities, energy traders and risk managers.

Hurricane and climate contracts

• Hurricane Forecast Contracts — how the count and landfall contracts work.
• Disaster Insurance Applications of Forecast Contracts — using landfall contracts as parametric cover.
• Hurricane Forecast Contracts as a Diversifying Asset in an Investment Portfolio — correlation properties against conventional assets.
• Bundling “NO” Hurricane Landfall Forecast Contracts to Reduce Downside Risk — constructing a bundle across locations.

Climate Contracts

• Global Crop Yield Forecast Contracts (view here).
• How Quickly Will The Globe Warm? Paris Agreement Forecast Contracts (view here).
• Will the Atlantic Overturning Circulation (AMOC) Collapse? Prediction Markets Can Quantify Sentiment (view here).

About the author

Patrick T. Brown is the Head of Climate Analytics at Interactive Brokers, where his work focuses on the information discovery and risk-transfer applications of prediction markets in weather, climate, and natural disasters.

He holds a PhD in Earth and Climate Science from Duke University, a master’s degree in Meteorology and Climate Science from San Jose State University, and a bachelor’s degree in atmospheric and oceanic sciences from the University of Wisconsin, Madison. He is an adjunct faculty member (lecturer) in the Energy Policy and Climate Program at Johns Hopkins University and has conducted research at the Carnegie Institution at Stanford University, NASA JPL at Caltech, NASA Langley in Virginia, NASA Goddard in Washington, D.C., and NOAA’s GFDL at Princeton University. He has published scientific papers in Nature, PNAS, and Nature Climate Change, as well as many disciplinary journals, and his research and commentary have appeared in The New York Times, The Wall Street Journal, The Economist, CNBC, CNN, The BBC, The Washington Post, NPR, Newsweek, The Guardian, The Atlantic, Foreign Policy, and The Los Angeles Times, among other venues.