Core insight
Blind copytrading of "smart money" often fails—not because these addresses don't make money, but because you're copying all their behavior: exploratory positions, hedges, low-conviction bets, and the rare high-conviction positions.
Key understanding: Effective copytrading is not replication, it's translation. You need to judge how serious each trade is to the original trader.
Why direct copytrading often fails
- Exploratory positions: Small positions testing market temperature, not strong convictions.
- Hedge positions: Serve risk management, don't necessarily represent directional views.
- Noise trades: Exploratory, low-conviction bets.
- High-conviction positions: The minority truly worth following closely.
What Conviction Score measures
Conviction Score attempts to infer a trader's confidence level from behavioral data. Common reference dimensions include:
- Single bet size relative to account size
- Whether continuously adding to the same market
- Performance of similar historical signals
- Trade timing and position concentration
Backtest summary
| Threshold | Trade count | Coverage | Avg return | Win rate |
|---|---|---|---|---|
| All (0.0) | 1147 | 100% | $20.35 | 75.2% |
| ≥ 0.2 | 715 | 62.3% | $22.18 | 77.9% |
| ≥ 0.4 | 448 | 39.1% | $25.93 | 81.0% |
| ≥ 0.6 | 358 | 31.2% | $26.99 | 80.7% |
Conclusion is direct: stricter filtering usually means higher per-trade quality, but fewer opportunities.
How to use in practice
- 0–0.2: More opportunities, but more noise. Suitable for portfolio replication.
- 0.2–0.4: Balanced quality and frequency. Suitable for most users.
- 0.4–0.6: Fewer signals, but higher marginal efficiency. Suitable for small capital with high selectivity.
pred101 implementation recommendations
An actionable copytrading framework should first find people, then filter trades. Wallet discovery is just the first step. The real edge often comes from filtering per-trade quality, not indiscriminately copying all on-chain buys.