What copying whales returned
The test is the z-score against the price the whale paid, not a return percentage. A strategy passes only if its one-sided z is at least 2.64. On Oct 23, 2026, follow-all lost and proven wallets were not significant.
Following whales, pre-registered Oct 23, 2026
Frozen result. Judged on 4,218 unseen picks from 10,543 graded flags. Edge is wins minus the win rate the entry price implied. z = edge ÷ standard error. This table is not updated when new bets resolve, and it is not a return.
| Strategy | Edge | z | Result |
|---|---|---|---|
| Follow every whale | -2.69 pts | -3.95 bar 2.64 | Fail. Lost to the price. |
| 24h net flow | -4.44 pts | -2.34 bar 2.64 | Fail. Negative, short of the bar. |
| Proven wallets, earlier data only | +2.97 pts | 1.58 bar 2.64 | Fail. Not significant. |
Same rules, on the graded flags we have now. Dollars use a $100 stake on every bet after a 2% taker fee (rate × min(price, 1−price) per share). The fee changes the dollars. It does not change the z-score, and it cannot turn a fail into a pass. 24-hour net flow stays in the frozen table only. This recompute is follow-every-whale, proven wallets, and every whale in that later window so the proven row has a baseline.
| Strategy | Bets | Edge | z | Result | After 2% fee |
|---|---|---|---|---|---|
| Follow every graded whale Every scored $10k+ whale bet. | 145 | +2.1 pts ±3.6 | 0.60 | Fail. Not significant. | +$221 fee drag −$177 |
| Every whale after the cutoff Same later window as the next row, every wallet. | 15 | -10.9 pts ±12.2 | -0.89 | Too few bets. | +$143 fee drag −$21 |
Proven wallets: 0 wallets, cutoff Oct 28, 20:00 EDT. The dollar column is not the edge. A win at 20¢ pays more dollars than a win at 80¢, so the dollars can disagree with the points. The z-score is the test. Hypothetical fills at the whale's price plus the fee and extra cost above. Ignores the spread, partial fills, and selling before resolution.
One wallet
Put a fixed stake on every graded flag one wallet made. A wallet picked off the leaderboard was chosen after the fact, so its z-score is not the pre-registered test above.
Why this overstates what you'd get
- You'd be late. Flags show up after the trade, often after the price has moved. Use "Copy delay" to price each bet 1 to 24 hours later from market history, or add an extra cost per fill.
- Picked after the fact. Wallets that look good here were chosen because they did well. Tick the out-of-sample box to test only bets made after the cutoff.
- Small samples are noisy. Look at the z-score. The follow strategies above need 2.64 to pass. A few lucky long shots can make the P&L.
- Fees: many Polymarket markets charge no trading fee. Some charge a taker fee of rate × min(price, 1−price) per share. The follow table defaults to 2%. The one-wallet fee box starts at 0 until you set it.
- Ignores spreads, liquidity, partial fills, and sells before resolution. It counts only graded flags (placed ≥1h before close, 3¢–97¢).
Hypothetical results from public data. Not financial advice.