Track 16Closed

Thin-coin range-MM — “buy the low of every candle”

passive market-making on thin coins: buy the low and sell the high of the range on almost every candle, harvest by frequency, gated by regime

Key metric

honest MM sim net −9…−15 bps even at maker=0; 6 independent angles converge

Illustrated

The setup

Price sweeps a range; a long lower wick pierces support — the temptation to “buy the low, sell the high” on almost every candle.

How it concretely fails

Gross forward return after the dip is noise (±1–5 bps, pooled after a long wick −0.4 bps), while the taker round-trip is ≈ 14 bps. The edge sits far below cost; WR 0.6–0.77 with negative expectancy. The reversion is a Roll bid-ask bounce (VR_mid ≈ 1.26), not a mid move.

Why an edge should exist / hypothesis: Thin coins have a genuinely wide range (ZORA/BOME range 6–10× vs ROBO 1.0) and real regimes — the eye sees price sweeping back and forth. Testable hypothesis: a passive bid at the bottom of the range captures the bounce beyond cost.
Construction: Six angles: honest passive-MM sim (bid at the low, ask at the high, regime gate); taker entry on the dip; a precise wick-rejection rule (lower wick ≥2× body, revisit low, target high); coin separation (wick% and variance-ratio); a 15-second timeframe slice.
Test data: Real MEXC L2 book and print candles, 26 coins. Pooled forward return after a long lower wick is −0.4 bps.
Results: Honest MM net −9…−15 bps across all configs even at maker=0: the passive bid only fills on a down-tick (adverse selection). Taker: gross forward return of the mid after a dip is ±1–5 bps (noise) against a ~14 bps taker round-trip. The wick filter clears cost on 0/26 coins. VR_mid ≈ 1.26 for all — the reversion is pure Roll bid-ask bounce, not the mid.
Validation: Six angles converge. High WR 0.6–0.77 with negative expectancy = the “scalp illusion”. The 15-second timeframe does surface a faint real mid micro-reversion (+2…+5.6 bps gross) — which is why the eye sees an edge — but it sits ≪ ~14 bps cost and is N-tiny.
Root cause: A long wick describes the PAST bar, it does not predict the next one. The mid is a martingale; the “wicks” are a bid-ask-bounce artifact, not a capturable mid move. The range is real, but excursion ≠ PnL.
Verdict: Dead. A showcase case of “excursion ≠ PnL”. It does not kill the spike trade (taker, 20×-ATR extremes, rare).
Source: scripts/research/thin_mm_sim.py · thin_taker_sim.py · thin_wick_pattern.py · PROGRESS 2026-09-13

Want the same check on your strategy?

Validation →