Research Findings
Dec 2025 – Jun 2026 · 3,645 × 1h bars · 60/20/20 time-ordered split
VALIDATED PHENOMENA (Research → Validation → OOS)
1
Volatility Clustering (ARCH)CONFIRMEDR²≈0.35–0.42
Today's range predicts tomorrow's range. Active in LoVol/MidVol. Disappears in HiVol.
2
Vol Shock → Range (GARCH proxy)EXPLAINEDR²≈40–80% lift
Volume predicts next range but has zero effect after controlling for current range. It's just ARCH.
3
Range Contraction → CompressionPROBABLER²≈OOS ✓
After 3+ contracting bars, compression persists (opposite of squeeze narrative).
4
Lag-5 Mean ReversionREJECTEDR²≈Research only
Significant in research set (p=0.0006), vanishes completely in validation and OOS. False positive.
5
Direction Prediction (all features)REJECTEDR²≈<0.001
No feature tested — candle structure, volume, wick, close position — predicts direction.
Strategy Design
A · Trend+Vol
Basis: ARCH active in LoVol/MidVol. Follow EMA20 trend only when ATR ratio < 1.5.
Edge: None confirmed. Direction is random. Tests whether trend-following works in predictable-vol regime.
TP 1.5× ATRSL 0.8× ATRMax hold 8hSize $500
B · Spike Fade
Basis: HiVol spikes are isolated (ARCH disappears). Low follow-through = exhaustion.
Edge: Structural — fade extreme spike + quiet next bar. The Jun 1 22:00 setup.
TP 1.0× ATRSL 0.8× ATRMax hold 4hSize $500
C · Vol Breakout
Basis: After N≥3 compression bars, first expansion bar enters in breakout direction.
Edge: OOS confirmed: compression persists, then eventually a real break occurs.
TP 2.0× ATRSL 1.0× ATRMax hold 6hSize $500
D · Random Baseline
Basis: Coin flip every 4 candles. Symmetric TP/SL = 50% EV before fees.
Edge: None by design. Any strategy failing to beat this is statistically noise.
TP 1.0× ATRSL 1.0× ATRMax hold 4hSize $500
Important: This system is a scientific instrument, not an alpha engine. The primary finding is that SILVER/USD direction is statistically random at 1h resolution over the observed sample. The paper trades here will accumulate evidence over time about whether any structure-based entry achieves edge above the D (random) baseline. No conclusions should be drawn until each strategy has ≥100 closed trades.