Copy-trading on prediction markets — riding whale flows

Definition

Copy-trading in prediction markets means watching a curated list of historically profitable wallets and placing the same trade within seconds of detecting their activity. Unlike equity copy-trading, prediction markets offer a structural advantage: the CLOB is public and order history is on-chain, so any wallet's full track record is auditable before you decide to follow them.

The strategy sits between information arbitrage and social proof. You are not modeling the market — you are modeling whether a specific person models it better than you do.

When it works

The conditions that make copy-trading profitable are precise:

The target has genuine edge, not luck. A wallet with 200+ resolved trades, positive net PnL after fees, and consistent sizing across market types is likely skilled. A wallet with 12 trades and a 90% win rate is probably noise. Verify via the Polymarket Data API (/data-api/v2/activity?user=0x...), not just the public leaderboard.

You can enter before the price moves. The copy-trade is only valuable if you catch the order before adverse selection eats the edge. Liquid markets (top-20 crypto binaries, major election markets) absorb large whale orders with a 1-3 cent slippage window. Illiquid niche markets can move 10+ cents on a single large order, meaning the copied trade is already stale by the time you fill.

The whale takes taker orders, not GTC limits. A wallet that posts GTC maker orders is queue-dependent; you cannot reliably front-run their fill. Wallets that sweep takers are far easier to mirror because the detection-to-fill window is deterministic.

Market has runway. Copying a trade with 10 minutes to resolution is usually gambling, not copying. The best candidates have 30 minutes to 24 hours of remaining life so the thesis can play out.

When it fails

Latency arms race. The Polymarket Data API polling interval is your ceiling. At 500ms polling you can detect most whale trades within 1-2 seconds. But dedicated infrastructure (co-located polling, websocket taps on the CLOB, mempool watching for on-chain settlement) can detect in under 100ms. If faster bots are already running, you are copying at a worse price than the whale entered — effectively subsidizing their exit.

Exit timing divergence. The whale bought YES at $0.40 and you mirrored at $0.42. If the market resolves YES, you both win. But if the whale exits early at $0.60 and you hold through a reversion to $0.30, your outcome is completely different despite "copying" the trade. polycopy implements independent exit discipline: it does not mirror exits, it runs its own stop-loss logic.

Adverse selection on edge cases. Whales with documented edge tend to concentrate in markets where their information advantage is structural (geopolitical events, sports insider knowledge, aggregated public data that retail traders ignore). When they trade a market that falls outside their demonstrated category, their accuracy drops to near-random. Blindly copying all activity from a wallet that happens to be good at soccer props will lose money on every other market they touch.

Permissioned copy-trade bans. Polymarket does not currently prevent copy-trading, but large whales are aware of followers and sometimes probe with decoy positions.

Fee math

Copy-trading faces the standard crypto-binary taker fee on both legs:

fee = shares * 0.25 * price * (price * (1 - price))^2

For a mirrored trade at p=0.45 for 100 shares:

fee_entry = 100 * 0.25 * 0.45 * (0.45 * 0.55)^2 = 25 * 0.45 * 0.06075 = 25 * 0.027338 ≈ $0.68

If the market resolves YES, each share pays $1.00. Net per share: $1.00 - $0.45 entry - $0.0068 fee = $0.5432.

Now consider copying at $0.48 due to slippage (whale moved the book by 3 cents):

fee_entry = 100 * 0.25 * 0.48 * (0.48 * 0.52)^2 = 25 * 0.48 * 0.06222 ≈ $0.75

Net per share: $1.00 - $0.48 - $0.0075 = $0.5125. You paid 3 cents more entry plus marginally higher fees — a 6% reduction in gross profit. Across many trades, this slippage delta determines whether copy-trading is profitable or just an expensive way to track the whale's PnL.

For maker-side exits (posting GTC limit orders to exit), fee is 0% plus rebates, partially recovering the taker cost.

Real-world examples

The most studied public wallet in the Polymarket copy-trading community is 0x7c3db7... (known as "Car" on prediction market Discord communities). This wallet accumulated a documented net PnL above $22K by posting GTC maker bids on outcomes priced 20-50 cents — primarily underdogs in sports and political markets. The strategy is less copy-trading fodder than it appears: Car's edge is in the GTC queue, and copiers who enter as takers at the already-filled maker price pay extra fees Car never paid.

A more copyable pattern is large taker sweeps on crypto binaries. A wallet placing a $500 FOK order on BTC YES at $0.52 with 4 minutes to the next 5-minute window is making a directional bet. Detection within 2 seconds and entry at $0.53-$0.54 leaves roughly 50% of the original edge intact, assuming the underlying thesis (BTC price at window end) is correct.

In practice, polycopy users report fill prices averaging 1.5-4 cents above the detected whale entry price across markets, with slippage higher on events (election day, FOMC announcements) and lower on routine daily crypto binaries.

Common variants

Mirror trading (1:1 sizing): Buy exactly the same number of shares as the whale. Simple but dangerous — a $10K wallet and a $200 wallet should not have the same position size. Risk of ruin is very different.

Scaled copy (% of bankroll): The safer default. If the whale commits 5% of their observed balance, you commit 5% of yours. polycopy implements this via a configurable maxPositionPct parameter.

Filtered copy: Only mirror trades in markets where the whale has a documented edge category. If a wallet has 80% win rate on NFL markets and 50% on crypto, only follow NFL. Requires maintaining a per-wallet, per-category performance breakdown.

Delayed copy (signal extraction): Watch 10+ whale wallets. When 3+ independently enter the same market within 5 minutes, treat that as a consensus signal and enter as if it were your own thesis — not as a copy-trade. This removes single-whale dependency and is less vulnerable to decoy probes.

Exit-only copy: Watch the whale for exits, not entries. If a wallet you trust starts unwinding a position you already hold, that is a strong exit signal even if the oracle hasn't moved.

Bots implementing this strategy

polycopy is the production implementation. It polls the Polymarket Data API for whale wallet activity, filters by configurable wallet addresses and position thresholds, and executes taker orders on the CLOB within seconds of detection. Position tracking and independent exit logic (stop-loss, time-based expiry) are handled in lib/state.js. Alerts go directly via Telegram without depending on bot-monitor.