Why Single Signals Lose and Multi-Signal Consensus Wins
Every retail trader has their favorite indicator. Every pro knows that one indicator is noise. Real edge comes from consensus across uncorrelated signals — and from inverting the bad ones.
The data behind the claim
We track 264 market signals — technical, on-chain, institutional, microstructure, macro. We score every prediction at 5min, 15min, 1h, and 4h horizons against actual price movement. The goal: identify which signals have edge, and which are coin-flips.
The result is uncomfortable. 202 of 264 signals (76%) are pure noise — accuracy between 45% and 55%. Statistically indistinguishable from random. These get ignored.
The 27 that actually work
Of the rest, 27 signals consistently score above 65% accuracy at 4-hour horizon (n > 50 predictions). Top performers right now:
| Signal | Accuracy 4H | Direction |
|---|---|---|
| auto_coinm_extreme_high | 80.0% | BUY |
| auto_trend_score | 79.2% | BUY |
| auto_oi_aave | 75.1% | BUY |
| auto_cmc_btc_dominance_change_7d | 67.7% | BUY |
| auto_coinalyze_funding_agg | 67.1% | BUY |
| auto_velocity_boost | 64.7% | BUY |
The hidden gold: anti-signals
Equally valuable: signals that are consistently wrong. Below 35% accuracy means inverting them gives 65%+ accuracy on the opposite side. We call them anti-signals:
| Anti-signal | Raw accuracy | Used as |
|---|---|---|
| auto_reversal_bottom_score | 22.1% | INVERTED → SELL |
| auto_hl_funding_btc | 31.8% | INVERTED → SELL |
| auto_hl_premium_btc | 32.9% | INVERTED → SELL |
| auto_eurusd_chg_1d | 34.0% | INVERTED → SELL |
Most traders would discard these as broken. They're not broken — they're contrarian indicators. When a signal is consistently wrong, that's information.
Consensus is alpha
The real edge isn't any single signal. It's voting. When 10+ buy-side signals (>65% acc) align AND fewer than 3 sell-side signals fire, probability of upward movement exceeds 70%. That's tradeable.
Example: in the past 30 days, our consensus buy threshold (10+ buy signals, ratio 3:1 over sell) has triggered 23 times. Outcome at 4h: 17 wins, 4 losses, 2 break-even. Win rate 78%. Avg PnL +0.42%/trade.
Why this beats single-indicator strategies
Indicators are correlated. RSI, Stochastic, Williams %R — they all measure momentum from price. They confirm each other in trending markets, lie together in choppy markets. The illusion of consensus.
True consensus requires uncorrelated sources: technical (price action), on-chain (network activity), institutional (ETF flows), microstructure (order flow), macro (DXY, VIX). When 3 of 5 categories align, the signal is real.
One indicator is opinion. Ten correlated indicators is louder opinion. Five uncorrelated indicators is information.