Levandor

Copy-Trade Analysis — What the Top Polymarket Weather Traders Actually Do

7 min read

Superseded/extended by the v2 decypher

A larger, gated, critic-corrected redo of this analysis off the official Polymarket weather leaderboard lives at weather-top50-decypher-2026-06-19 (87 candidates → 65-wallet gate → top 50). It sharpens the thesis to “capacity, not return”, concedes the ≤0.95 fillable edge is plausible but not proven, and corrects two wallets named here (BeefSlayer → mid-range directional, not a 0.99 whale; ColdMath → reconstruction-failure, not a negative control).

Reverse-engineered the on-chain trade history of the top Polymarket weather traders to learn what actually makes money. None of the winners run the naive “buy cheap, sell the pop” play. The real money buys the already-winning bucket near-certain at the close (whales), or buys a forecast edge at open and holds. The longshot-pop scalp is what the smallest winner does. This independently confirms the intraday-drift-study-naive-no-go arc: the surviving edge is nowcasting the winning bucket and trading the market’s repricing lag — pure information-timing.

For Agents

This is a research finding (copy-trade reconnaissance), not code documentation. Bottom line: the edge is nowcasting the winning bucket and trading the market’s repricing lag, not buying-cheap-and-selling-the-pop. Three archetypes exist (see Two-and-a-half winning archetypes); the late near-certain settlement play is where the real money is, the early forecast value play is the other winner, and the round-trip / sell-the-pop play (the user’s original idea) is run by only two traders — one of whom is LAST in PnL. The killer example (Killer example — the nowcast edge in action) shows a 0.33→1.00 intraday reprice captured by the top trader. This independently confirms the intraday-drift-study-naive-no-go conclusion (naive drift NO-GO −28%; forecast-conditioning → priced-in/breakeven; live nowcast = the only surviving edge). Data clients used to fetch trades: wx-market-trades-client (same data-api /trades endpoint, queried by user=<addr> here). Market-structure ground truth: polymarket-weather-market-structure.

Source & method

The leaderboard ranks GENERAL PnL, not weather-specific

weatherbot.fi’s /api/leaderboard lists 13 traders, but it ranks general Polymarket PnL, not weather-specific performance. Two of the 13 (Poligarch, russell110320) trade ~0 weather — they are on the board for unrelated markets. Filtering to genuine weather activity leaves ~10 real weather traders, and they trade dozens of cities — far beyond weather-bet’s current 4 (London / Paris / Seoul / Beijing).

  • Leaderboard: weatherbot.fi /api/leaderboard → 13 candidate addresses (general PnL ranking, not weather-filtered).
  • Strategy reconstruction: pulled each trader’s on-chain trades from data-api.polymarket.com/trades?user=<addr> — the ~3,500 most-recent trades per trader (the API caps offset at 3000, so this is the recent-history ceiling per address).
  • Same endpoint as the live trader’s wx-market-trades-client, queried by user=<addr> (vs asset=<token> for per-bucket history).

Script + data on disk

  • Analysis script: research/copytrade/analyze.py
  • Derived profiles: research/copytrade/profiles.json
  • Raw per-trader trade dumps: research/copytrade/raw/

Two-and-a-half winning archetypes

Reverse-engineering each trader’s trades surfaces three distinct playbooks. Critically, none of the winners run the naive “buy cheap, sell the pop” that motivated the original project idea — and the one trader who does is last.

1. Late near-certain settlement — the whales, the real money

The biggest weather PnL comes from buying the already-winning bucket once the day’s high is effectively observed, in size, at near-certain prices, and collecting the last few cents to settlement.

  • HondaCivic — the standout: 1.00.
  • Also BeefSlayer, aboss, ColdMath run the same play.
  • Mechanism: buy the bucket that is already winning at 0.90–0.999 once the day’s high is effectively known (live observation), in size, and collect the residual cents to settlement.
  • Edge: capacity + fast execution at the close, not prediction. Very low variance — you are buying a near-certainty and clipping the last spread.

This is the real-money play

The largest weather PnL is not a forecasting edge — it is a capacity + execution-at-the-close edge. You need (a) a live read that the bucket has effectively won, and (b) the size + speed to buy it at 0.90–0.999 before it finishes converging to 1.00. Low variance, high capital requirement.

2. Early forecast value — buy the open on a forecast edge, hold

The other genuine winner: take a position at market open on a forecast edge and hold to resolution.

  • Railbird — buys favorites (~0.85) at open.
  • dpnd — enters at a median 33h before resolution (i.e., at/near open), in the cheap/mid buckets.
  • Poligarch — same early-forecast posture (though near-zero weather volume overall).
  • Mechanism: the edge is in the forecast read at open; hold through the day rather than scalping the intraday move.

3. Round-trip / sell-the-pop — the user’s original idea, and it is the WORST

The longshot-pop play is what the smallest winner does

Only two traders run the buy-cheap / sell-the-pop round-trip:

  • huskyvs — 47% round-trip, cheap-longshot scalper — and LAST in PnL ($18K, rank 12).
  • ShyGuy1 — 34% round-trip.

The naive drift/round-trip strategy that motivated weather-bet is precisely the play of the least profitable weather trader on the board. This is the copy-trade confirmation of the intraday-drift-study-naive-no-go NO-GO.

Killer example — the nowcast edge in action

The single trade that crystallizes the whole thesis:

HondaCivic — London 18°C, Apr 30

  • BUY 7,523 sh @ 0.328
  • SELL 4,001 sh @ 0.999 — six minutes later

The bucket repriced 0.33 → 1.00 intraday because HondaCivic knew (from live observation) it was about to win before the market repriced. The ~3× return in six minutes is not drift and not luck — it is trading the market’s repricing lag on a bucket you can already see has won.

This is the nowcast edge made concrete: observe the realized high coming in, buy the winning bucket before the order book catches up, and let the reprice come to you.

Convergent conclusion

The edge is nowcasting the winning bucket and trading the repricing lag

Every winning archetype — whales clipping the close, forecast-value traders holding from open, and even the killer six-minute reprice — is a variant of knowing which bucket wins before the market fully prices it. The losing archetype (huskyvs, round-trip scalp) is the one that doesn’t — it trades the drift blind. This is exactly the information-timing thesis the price-drift research arc converged on.

This independently confirms the intraday-drift-study-naive-no-go research arc — reached from a completely different angle (on-chain behaviour of real winners vs. a backtest):

Research stepVerdictSource
Naive intraday drift (buy-open / sell-pop, blind)NO-GO, −28% (survivorship bias)intraday-drift-study-naive-no-go
Forecast-conditioning on public infoPriced-in / breakeven (~+1.2%)intraday-drift-study-naive-no-go
Live nowcast (track realized temp, trade reprice lag)The only surviving edgethis note + drift study

Two independent lines of evidence now agree

The backtest predicted the edge is winner-selection-via-nowcast, not drift. The copy-trade recon shows the top weather trader makes their money precisely there — buying the near-certain winning bucket and clipping the reprice. The naive round-trip is empirically the weakest play (last-place trader) on both fronts.

Next step

  • Scope a nowcast strategy off the cached price curves:
    • Entry rule — when do we declare a bucket “effectively won” / sufficiently favoured to buy?
    • Observation source + latency — what realized-temperature feed, how fresh, and how far ahead of the order book can we act?
    • Repricing-lag backtest — replay the cached intraday price curves (research/data/) to measure how long the book lags an observed high, i.e. the window the HondaCivic six-minute trade exploited.

Graph View

Backlinks