ColdMath Polymarket: Inside the Most Famous Weather Trading Strategy

Who Is ColdMath?

ColdMath is a Polymarket wallet address whose trading activity is concentrated almost entirely in daily temperature markets. The handle appears on Polymarket's public weather leaderboard, which ranks traders by cumulative net profit from weather-category markets.

The on-chain data is publicly verifiable: every trade placed, every resolution, every USDC flow in and out of the wallet is readable from the Polygon blockchain and indexed by Polymarket's public interfaces. What is not publicly known is who controls the wallet, what software runs the strategy, or whether “ColdMath” is one person or a team.

As of early 2026, the ColdMath wallet shows approximately $120,000+ in cumulative net profit from weather markets on Polymarket's public leaderboard. This figure is real on-chain data. The frequently-cited claim that ColdMath turned “$300 into $219,000 over three months” originates from a Medium post by an analyst who studied the wallet; the $120K leaderboard figure represents the verifiable net-profit snapshot, while the $219K figure may reflect a different calculation methodology or a specific portfolio peak.

The more conservative interpretation: ColdMath is a consistently profitable, high-activity algorithmic weather trader with a confirmed six-figure P&L from public on-chain data.

What the On-Chain Data Shows

The @armouredme thread performed a detailed wallet autopsy — examining each transaction, city, position size, and outcome over a multi-month period. Several patterns were consistent across the analysis.

City Concentration

ColdMath's trading was concentrated in secondary markets — not the highest-volume global cities like Shanghai or Tokyo, but:

• Buenos Aires (Argentina)
• Cape Town (South Africa)
• Dallas (Texas, when active)
• Atlanta (Georgia, when active)

These cities share characteristics: lower bot saturation, wider spreads, and less algorithmic competition than Shanghai or London. This is not a coincidence. Choosing secondary markets reduces direct competition from well-resourced algorithmic competitors while still finding sufficient liquidity to execute.

The flip side: secondary markets have lower volume, which caps maximum position size. ColdMath's per-trade sizing appears to have been constrained partly by this liquidity ceiling.

Position Sizing

ColdMath's individual positions were small — typically $50–$150 per bucket, rarely exceeding $200 on a single outcome. This is consistent with Kelly-fraction sizing applied to a modest-to-mid bankroll, and with the constraint that thin secondary markets can't absorb large orders without moving the price against the trader.

The total position count over the analyzed period was high — hundreds of individual trades. The law of large numbers is the mechanism: many small edges compound into a large total gain.

Cross-Category Performance

The analysis found that ColdMath's performance in non-weather categories was negative or flat: political markets, crypto markets, Venezuela markets, and other Polymarket categories all showed poor or zero returns. Weather was the sole source of sustained profit.

This is exactly what you'd expect from a specialist algorithmic strategy. The model is calibrated for daily temperature markets. It has no particular advantage in markets that require different signal types — news parsing, political modeling, on-chain crypto data.

Strategy Type: The Barbell

The @armouredme analysis described ColdMath's approach as a barbell strategy: a combination of:

1. Many small positions on underpriced tail buckets (5–15¢ YES contracts in buckets with genuine non-trivial probability).
2. Occasional higher-conviction positions on central buckets when the model shows strong consensus.

The barbell dynamic makes intuitive sense as a weather strategy. Tail buckets on thin secondary markets are often severely underpriced — retail concentrates on the modal bucket, ignoring lower-probability but non-trivial adjacent outcomes. A systematic buyer of underpriced tails wins a smaller fraction of the time but gets paid 6–20× when they win, and the math works in their favor.

The Inferred Strategy: What It Likely Is

Based on the on-chain analysis and what's publicly known about Polymarket weather mechanics, the ColdMath strategy most likely operates as follows:

Forecast layer: Pulls ensemble temperature forecasts (probably GFS/ECMWF or a commercial aggregator) at the resolution-station coordinates for the target cities. Computes a probability distribution over the market's buckets.

Edge calculation: Compares model-implied bucket probabilities to current YES prices. Identifies buckets where the model probability exceeds the market price by a minimum threshold.

Entry filter: Appears to focus on buckets below $0.15 where model probability is materially higher — consistent with tail-buying. May also filter for minimum market volume to ensure fills are possible.

Sizing: Small fixed or fractional-Kelly sizing per trade (~$50–$150), with no aggressive concentration into any single outcome.

Execution: Automated (the volume and consistency of trading across multiple cities is incompatible with manual execution at the observed frequency).

Cities: Secondary markets with wider spreads and lower competition.

This is not a complex strategy. It is a well-disciplined execution of a basic probabilistic framework: find systematic mispricing, bet proportionally, repeat many times. The sophistication is in the discipline and consistency, not in exotic modeling.

The $300 to $219,000 Claim: What to Believe

The claim that ColdMath turned $300 into $219,000 in three months spread widely and is almost certainly why the wallet became the most-discussed in the Polymarket weather community. It needs careful parsing.

What's likely accurate:

• ColdMath has generated substantial cumulative profit in weather markets, confirmed by the public leaderboard.
• The strategy appears to have had a particularly strong performance period in 2025 when weather markets were less competitive and secondary cities like Buenos Aires had very wide spreads.
• The percentage return over the early period of the strategy was genuinely high — a small starting bankroll compounded at high Kelly fractions in less competitive conditions can produce large percentage returns.

What requires caution:

• The specific “$300 starting capital” figure is self-reported or reconstructed — wallets can be funded at any time, and initial capital is not verifiable from transaction history alone.
• “Three months” as the timeframe for the compounding is not independently verifiable from public data.
• Current market conditions are substantially more competitive than 2024–2025. Secondary cities like Buenos Aires and Cape Town now attract more algorithmic traders following the ColdMath viral moment. Replicating those percentage returns starting today, against a more competitive player pool, is significantly harder.

The on-chain P&L is real. The specific origin-story framing should be treated as approximate and illustrative, not as a verified trading record.

What Traders Tried: Copy-Trading ColdMath

Following the viral X thread, numerous traders attempted to copy ColdMath by:

1. Following the wallet on Stand.Trade / PolyCop — tools that mirror another wallet's trades with a slight delay.
2. Manually monitoring the wallet and replicating trades.

Results were mixed, and this is instructive.

Why copy-trading ColdMath is harder than it looks:

Slippage. ColdMath likely enters before the market has reacted to the forecast signal. Copy-traders enter after. On thin secondary markets, the 5–10¢ spread means a delayed copy-trade fills at a significantly worse price than the original trade.

Feedback loops. A large community of copy-traders following a single wallet on small markets creates a self-defeating dynamic: as soon as ColdMath buys into a thin Buenos Aires bucket, copy-trade bots push the price up, narrowing the spread and eliminating the edge for late followers.

Lag exposure. ColdMath's strategy profit is earned in the gap between market price and true probability. By the time the trade appears on-chain and is copied, that gap may have closed. The copy-trade captures a fraction of the edge at best, zero at worst.

Directional concentration. When hundreds of traders copy the same position in a thin market, they all lose together when the trade doesn't work. The strategy was designed for a single well-sized wallet, not for crowd replication.

Several copy-traders on Polymarket Discord reported profitable periods in the first weeks of copying, followed by degrading returns as the market adapted. This pattern is consistent with edge decay driven by the copy-trade crowding.

What Actually Makes ColdMath Profitable (The Principles to Extract)

Forget the narrative. Here are the transferable principles from what the on-chain data actually shows:

1. Specialize in a category. ColdMath made all its money in one category: daily temperature markets. Zero demonstrated edge in politics, crypto, or other markets. Specialization allows deep calibration of a specific model to a specific market type. Generalism usually produces diluted edge across the board.

2. Compete where competition is weakest. Secondary markets (Buenos Aires, Cape Town, Atlanta) were chosen specifically because the competition was less fierce. This is a classic market-selection principle: better to have a 15% edge in an imperfect market than a 2% edge in a highly efficient one.

3. Use small, consistent position sizes. No trade was sized aggressively enough to break the strategy on a single bad outcome. The consistent $50–$150 per trade means even a 10-trade losing streak (which happens in binary markets even with edge) doesn't materially damage the bankroll.

4. Focus on tail mispricing. The barbell focus on underpriced tails is a response to a real market inefficiency: retail concentrates on modal outcomes and ignores adjacent lower-probability buckets. A systematic tail-buyer exploits this without needing speed or exotic models.

5. Run it algorithmically at volume. The frequency and consistency of ColdMath's trading is humanly impossible to replicate manually. The strategy is automated. This is the prerequisite for the statistical law-of-large-numbers to manifest in actual P&L within a reasonable timeframe.

Should You Build a ColdMath-Style Strategy Today?

The honest answer: the pure ColdMath approach is harder to replicate now than it was 18 months ago.

Secondary markets have become more competitive. Buenos Aires and Cape Town now attract algorithmic participation that didn't exist before the viral moment. Spreads have tightened. Edge windows on tail buckets have narrowed.

This doesn't mean the category is dead — it means the minimum viable edge threshold has risen, the forecast calibration required has improved, and market selection now requires identifying the next underexplored market rather than copying the last one.

What remains permanently valid from the ColdMath playbook:

• Specialize and go deep rather than covering everything superficially.
• Choose markets where you have a calibration advantage relative to the competition.
• Size consistently and repeat at volume.
• Run it algorithmically.

The specific cities may need to change. The principles won't.

The Broader Lesson

ColdMath is interesting not because the strategy is exceptionally sophisticated — it isn't — but because it demonstrates that methodical, disciplined application of basic probabilistic principles can generate real, sustained profit in a prediction market category that most people overlook.

Daily temperature markets are not glamorous. They don't trend on X. They don't get analysts making guest appearances on CNBC. The price action is quiet, the subjects are mundane, and the winning trades resolve in hours.

That's exactly the point. The most persistently exploitable edges in prediction markets exist in the categories that attract the least sophisticated attention. For now — and probably for a while — daily temperature markets on Polymarket are still that category.

Related articles

• The Polymarket Weather Leaderboard: Who's Winning and How
• Polymarket Weather Strategy: How to Find and Keep an Edge
• Polymarket Weather Bot: How Automated Trading Works