Every hour, our scanner evaluates 75 crypto perpetual futures. Each instrument must pass 5 independent filters before generating a signal. Over 95% of setups are rejected.
Most trading alerts fire on a single indicator — price crossed a moving average, RSI hit 30, a candle pattern appeared. The problem? Single-indicator signals are noisy. You get dozens per day, most of them wrong.
We take the opposite approach. Before any signal reaches you, it passes through 5 independent filters. Each gate tests a different dimension of the setup — statistical relationship, extremity of move, market regime, liquidity, and timing. If any single gate fails, the signal is rejected.
75 instruments scanned per hour → 1–3 pass all 5 gates → >95% rejection rate
Does this instrument actually move with Bitcoin? We run a rolling regression over a 180-bar window. If the R² is below 0.30, the relationship is too weak — the residual is noise, not signal. Only instruments with a statistically significant BTC relationship proceed.
Is the price unusually far from where it should be? After removing BTC correlation, we compute a z-score on the residual. A z-score above ±2.0 means the instrument is in the top 5% of historical deviations — rare enough to be meaningful, likely to revert.
Is the overall market in a state where mean-reversion works? During breakouts, prices keep running instead of snapping back. Our regime classifier (GateKeeper) reads BTC momentum, volatility, and trend structure to determine whether conditions favour reversion trades.
Is there enough trading activity to enter and exit cleanly? We require a minimum of $1M hourly volume. Low-volume instruments have wide spreads and slippage that destroy edge. This gate ensures every signal is on a liquid, tradeable instrument.
Is this a new signal, or the same stale setup from last hour? We track z-scores across cycles and only alert on fresh threshold crossings or moves that are still deepening. This prevents re-posting the same signal repeatedly and ensures you see opportunities as they develop.
The core insight behind Engine 1 is BTC-neutral analysis. In crypto, almost everything correlates with Bitcoin. If BTC drops 5%, most altcoins drop too — that’s not a signal, that’s just correlation.
We strip out the BTC component using linear regression. What remains is the residual — the part of an altcoin’s movement that can’t be explained by Bitcoin. When this residual hits an extreme, it tends to revert. That’s the trade.
Signal format: When you see ENA SHORT z=+2.19, it means ENA is trading 2.19 standard deviations above its expected value relative to BTC — a statistically rare divergence likely to compress.
The z-score measures how extreme a move is, expressed in standard deviations from the mean. Think of it like a thermometer — the higher the reading, the more stretched the price is from normal.
z = ±1.0 — Normal variation, happens 68% of the time. Not interesting.
z = ±2.0 — Significant, happens only 5% of the time. Signal territory.
z = ±3.0 — Extreme, happens 0.3% of the time. High-conviction setup.
A positive z-score means the instrument is expensive relative to BTC (short opportunity). A negative z-score means it’s cheap relative to BTC (long opportunity).
The statistical framework draws on decades of mean-reversion research. Poterba & Summers (1988) demonstrated mean reversion in stock prices. Blitz, Hanauer, Vidojevic & Vliet (2017) documented residual momentum effects. Our regime filter is supported by Hurst, Ooi & Pedersen (2017) at AQR, who studied time-series momentum across 67 markets spanning 137 years.
The full research document with all citations is available in our Regime-Based Trading documentation.