ETF-Flow Residual Check — Momentum vs. Genuine Edge

2026-05-19 · status: dropped · 1.8s

Hypothesis: If the ETF-flow signal from Exp #18 (IC +0.37) is mainly BTC daily momentum, little remains after residualising on the pre-release return (22:00 UTC marker) (IC < 0.05). If residual_IC > 0.20, ETF flow is a genuine directional edge.

Verdict: MOMENTUM ARTEFACT — residualised IC = +0.0754 collapses vs. original +0.0445 (drop -69.6%). The supposed ETF edge was almost entirely BTC daily momentum. The ETF-flow feature provides no independent information beyond ret_pre.

Key metrics

Approach

Marker = 22:00 UTC daily close (NY-Close, ETF-Release-Moment). Construct three returns:

• ret_pre(t) = log(close_t_22:00) − log(close_{t−1}_22:00) — everything before release
• ret_fwd(t) = log(close_{t+1}_22:00) − log(close_t_22:00) — everything after release
• flow_t = Farside aggregated net flow daily

Pooled ICs of all pairwise relationships plus one OLS residualisation of flow_t on ret_pre. If the daily BTC momentum (ret_pre) explains most of Cov(flow, ret_fwd), IC(flow_residual, ret_fwd) collapses.

Pooled OOS Pairwise ICs

Decomposition statistics

• OLS β (flow_t per unit ret_pre): +4995.28 US$m per log-return unit

• OLS α (intercept): +91.82 US$m

• R² of flow explained by ret_pre: 16.40%

• Residualisation drops IC by -69.6% (+0.0445 → +0.0754)

Event-conditional spreads (fwd 1d return, bps)

Walk-forward stability (11 windows)