Polymarket Weather Trading: What the Leaderboard Data Actually Shows (2026)

What the Public Leaderboard Shows

The March 2026 monthly weather leaderboard on Polymarket (pulled March 27, 2026) tells a clear but incomplete story.

The biggest single win this month: Handsanitizer23’s position in “Highest temperature in Atlanta on March 17?” — a $18,734 entry that resolved for $64,600.

But three things the leaderboard does not show are at least as important as what it does.

Why Weather Markets Are Structurally Different

Three properties make Polymarket’s daily temperature markets more amenable to model-based analysis than politics, crypto, or culture markets. In short: weather may be the rare prediction market where the inputs are free, the settlement is objective, and the edge survives only because most participants still don’t bother to do the math. These are structural observations about the category, not claims about any specific trader’s edge.

Weather may be the rare prediction market where the inputs are free, the settlement is objective, and the edge survives only because most participants still don’t bother to do the math.

Objective resolution with minimal interpretation. Temperature markets resolve against a specific weather station reading — typically an airport station sourced from Weather Underground, with defined rounding rules specified in each market’s resolution criteria. The outcome is a number from a public sensor, not an editorial judgment, a vote count subject to dispute, or a price determined by another market. This makes temperature outcomes far less interpretive than political or cultural markets, though edge cases exist: station sensor failures, Weather Underground display delays, and ambiguity in rounding rules add operational risk that doesn’t appear in the leaderboard numbers.

Free, calibrated forecast models with measurable error. The Global Forecast System (GFS) updates every six hours. The European Centre for Medium-Range Weather Forecasts (ECMWF) runs twice daily. The National Weather Service issues point forecasts for specific stations across the United States. These models have measurable error characteristics that vary by city, season, and lead time — and those error characteristics are publicly documented in NWS verification reports.

This means a calibrated probability distribution across temperature brackets is constructible, not theoretical. For a given city, date, and forecast lead time, it is possible to estimate the probability of the high temperature landing in each 2°F bracket based on the forecast point and the historical error distribution for that location. Whether most Polymarket participants actually perform this calibration is a separate question — and one that public data cannot answer.

Thin liquidity amplifies potential mispricings. Individual daily temperature markets on Polymarket vary widely in volume — from under $40,000 for smaller international cities to over $400,000 for active U.S. cities on high-attention days. In thinner markets, a few uninformed trades can push prices meaningfully away from model-implied fair values. This is a structural observation about market microstructure, not a claim about participant sophistication. As volume grows and more systematic participants enter, this property weakens.

A Worked Example: Houston, March 26

On March 26, 2026, two weather bracket trades on Houston illustrate how the same city, the same day, and the same analytical approach can produce both a win and a loss. The market prices and outcomes below are from public, resolved markets. The probability estimates are illustrative — based on a hypothetical NWS forecast of 84°F for KIAH (Houston Intercontinental) with a city-specific historical forecast error σ of approximately 4.5°F for late-March Houston.

Bracket: “Will the high be 85.5°F or above?” The market priced YES at 28¢ — implying a 28% chance of reaching that threshold. The illustrative distribution places approximately 37% probability above 85.5°F. The model favored YES at 28¢, not NO at 72¢. The NO side ultimately won, but the model-implied edge was not clearly on that side.

Bracket: “Will the high be 83.5°F or above?” The market priced YES at 61¢ — implying a 61% chance of reaching 83.5°F. The illustrative distribution places approximately 54% probability above 83.5°F — meaning the market was overpricing YES by about 7 percentage points, making NO at 39¢ modestly +EV. But Houston’s actual high exceeded 83.5°F. The NO position lost.

One trade won. One trade lost. Same city, same day. The bracket closer to the forecast point (83.5°F, only half a degree below the 84°F forecast) carried far more uncertainty than the tail bracket (85.5°F). This is the reality that leaderboard screenshots don’t capture: even when the structural math favors a trade, individual outcomes are determined by where the temperature actually lands relative to a probabilistic distribution — not by whether the model was “right.”

Even when the structural math favors a trade, individual outcomes are determined by where the temperature actually lands — not by whether the model was “right.”

The broader pattern is worth noting. Tail brackets — those furthest from the forecast point — tend to offer the widest gaps between model-implied and market-implied probabilities. But they also resolve against you less often, creating a payoff shape where small, frequent gains are punctuated by occasional larger losses. This is consistent with the longshot bias documented extensively in sports betting and horse racing: low-probability outcomes are persistently overpriced, and the traders who profit from this pattern do so across hundreds of trades, not on any single bracket.

What Leaderboard Profiles Suggest (But Cannot Prove) About Strategy

Public Polymarket profiles display prediction counts, total volume, biggest wins, and category rankings. From these observables, we can identify patterns. We cannot determine causation, method, or intent.

The Structural Math: Why Category Properties Matter More Than Any Single Strategy

The structural math rarely discussed elsewhere is this: it applies to the category as a whole, not to any individual trader.

Forecast error is measurable and city-specific. NWS forecast error for daily high temperature is well-studied through official verification reports. City-specific standard deviations typically range from approximately 2.5°F for tropical and coastal cities (where marine influence stabilizes temperatures) to 7°F or more for continental cities during transitional seasons (where frontal passages create large forecast uncertainty). Summer forecasts are generally tighter than winter; 24-hour forecasts are tighter than 48-hour forecasts. These are not estimates — they are published statistics derived from decades of verification data.

However, these error distributions are not perfectly normal, and they exhibit known regional biases. NWS verification reports document warm biases in certain U.S. regions during frontal passages and seasonal transitions. A model that assumes symmetric, normally distributed errors without accounting for these biases will systematically misprice certain brackets — potentially in the wrong direction. Backtesting only winning periods, or only high-volume cities where outcomes happened to favor the model, overstates the edge.

Model-implied fair probability versus market price. When a temperature bracket is priced at 20¢ on Polymarket, the market implies a 20% probability that the temperature will land in that range. If a calibrated forecast distribution — built from the NWS point forecast and the city-specific error distribution for that lead time — estimates the model-implied fair probability at 12%, the bracket appears overpriced by 8 percentage points.

This gap is the same phenomenon documented in horse racing and sports betting as “longshot bias” — the persistent tendency for low-probability outcomes to be overpriced relative to their true frequency. In sports betting, the explanation involves psychological preference for large potential payoffs on small stakes. In prediction markets, the mechanism may be similar: cheap contracts (5¢–20¢) attract disproportionate buying interest regardless of whether the implied probability is calibrated.

Variance and compounding: the short-odds framework. A trader systematically selling NO on overpriced tail brackets at 85¢–95¢ faces a concave payoff structure — small gains on most trades, occasional large losses. But research from sports betting mathematician Joseph Buchdahl and others demonstrates that this payoff shape, despite looking fragile on any single trade, maximizes expected bankroll growth when the edge is genuine.

The key metric is Maximum Expected Growth (MEG), approximated as edge² / (2 × fractional odds). At short odds (near 1.05–1.15 in decimal), MEG per unit of risk is dramatically higher than at long odds (5.0–20.0) with the same percentage edge. Monte Carlo simulations published by the Smart Betting Club show that over 858 bets at short odds with genuine edge, zero out of 100 simulated bettors lost money — while 38% of simulated longshot bettors with identical edge went negative over the same period.

This framework explains why some weather traders maintain near-perfect win rates while others swing wildly: it’s not (only) about forecast quality. It’s about where on the probability curve you choose to trade. The weather edge, if it exists, is probably less about predicting the exact temperature and more about repeatedly finding tail brackets the market prices badly — then letting the short-odds math compound over hundreds of trades.

What Public Data Cannot Tell You

What Could Compress or Eliminate the Edge

For Sports Bettors Moving Into Prediction Markets

If you already evaluate player prop markets — “Will LeBron score over 27.5 points?” — you already have the framework for daily temperature brackets. The structure is identical: a continuous variable discretized into brackets with defined resolution sources. Estimate the probability distribution, compare it to the market-implied probability, calculate expected value after fees.

The key difference: weather forecast models are free, public, and update on fixed schedules. Sports injury and lineup information is gated and fast-moving. Weather markets move on a slower timescale — hours, not minutes.

Polymarket weather markets carry up to 1.25% peak taker fees after March 30. Kalshi offers daily temperature markets through its regulated U.S. exchange at its standard formula-based fee. For a detailed comparison, see the ChanceMetrics Prediction Market Fees Compared calculator.

Tools for Evaluating Weather Bracket Trades

These tools help estimate modeled expected value under your own assumptions — not predict outcomes.

Conclusion

Weather markets are among the most structurally analyzable categories on any prediction platform — objective data, free models, thin liquidity. The March 2026 leaderboard is consistent with systematic edge. But consistency is not proof, and the clock is ticking: volume is up fourfold since launch, fees are expanding March 30, and automated accounts are already trading at scale.

If the profits persist after fees expand and more systematic traders arrive, that will say more than any viral P&L screenshot ever could.

The leaderboard does not prove weather is easy. If that edge remains real after fees rise and systematic traders pile in, that will matter more than any viral P&L screenshot.

Frequently Asked Questions

Common questions about Polymarket weather trading, fees, and how temperature bracket markets work.