OI-weighted funding & Market Pulse: lessons from 36 failed event tests
After BH-FDR correction, 36 single extreme-event tests showed no edge. How we rebuilt around normalized composite scoring and volatility filters.
01 / Why 36 single-event tests failed multiple-testing correction
We tested 36 single-indicator extreme events (256 statistical tests total) across orderbook skew, funding extremes, and limit-wall pulls. After Benjamini-Hochberg FDR correction, the lowest q-value was 0.128 — no single extreme trigger beat random timing after fees.
02 / Moving to OI-weighted funding and a normalized Market Pulse
Instead of reacting to noisy single-metric spikes, we weight perpetual funding by exchange Open Interest and normalize orthogonal inputs onto [0, 1] before averaging into MARKET_PULSE (0–100).
OI_WEIGHTED_FR = sum(fr_i * oi_usd_i) / sum(oi_usd_i) // Bybit + OKX + Binance
score_i = clamp((val_i - ref_lo) / (ref_hi - ref_lo), 0, 1)
MARKET_PULSE = 100 * mean(scores) 03 / Extending the pipeline
Because we own the D1 time-series schema, adding custom factors like session/weekday seasonality and candle-momentum features requires zero third-party dependencies.