← Research

Why an edge should exist / hypothesis

Idea: aggressive accumulation leaves a footprint in the aggregate tape you can ride. Testable hypothesis: an anomalous flow footprint predicts short-horizon direction beyond cost.

Construction

Causal test of 4 aggregate-flow features (signed-vol, run-length, clip-repeat, round-share), both directions (with-flow and counter), horizons 5–300s, gated by taker costs.

Test data

Binance USDT-M daily aggTrades, 30 days × 5 coins (DOGE/WIF/1000BONK/SOL/SUI) = 164,902 events.

Results

Gross across all gates/directions/horizons 0.000–0.002%, MFE/MAE 0.95–1.06, net ≈ −0.10% everywhere (just the fee). A with-flow>counter micro-asymmetry sits ~50× below cost.

Validation

N=165k, per-coin consistently negative. 4th independent confirmation after informed-flow / imbalance-bars / iceberg.

Root cause

Public data has no participant ID → only an aggregate footprint is observable, and it carries no direction. A skilled executor also randomizes accumulation across venues/time by design.

Verdict

Dead on tape. The open door is L2 order-flow imbalance from the HF book (being collected forward).

Source

scripts/research/aggflow_anomaly.py · PROGRESS 2026-08-29