Track 17Closed
HL MM-behavior prediction — forced unwind after a liq flush
predict a market maker's forced unwind after it absorbs a liquidation flush — MM inventory on Hyperliquid is public on-chain
Key metric
causally-clean NULL across 40 coins / 63 days; partial-corr ≈ 0, realizable fade 0/40 coins
- Why an edge should exist / hypothesis
- HL is the one venue with public inventory. Hypothesis: an MM that built inventory absorbing liquidations must unwind it, giving a predictable forward move against its position.
- Construction
- Aggregate MM inventory by address, partial-corr with forward return (controlling for the recent move), realizable fade of inventory extremes.
- Test data
- HL fills (SQD Portal) with addresses, 40 coins × 63 days.
- Results
- Partial-corr of aggregate MM inventory with forward return ≈ 0 (mean +0.006 on thin and mid coins). Where weakly significant, the sign is POSITIVE — against the unwind hypothesis. Realizable fade of inventory extremes is positive on 0/40 coins (mean −7.3 bps).
- Validation
- Causally clean (controls for the recent move); 40 coins consistently null or negative.
- Root cause
- The MM hedges and unwinds inventory off HL (on a CEX) — there is no observable on-chain footprint of intent. We see the leg, not the intent: the same disease as follow-the-whale.
- Verdict
- Dead. Closes the entire HL arc (follow / whale / MM / liq-map).
- Source
data/hl_mm/* · scripts/research/hl_sqd_fetch.py · hl_mm_predict_stream.py · PROGRESS 2026-09-08
Want the same check on your strategy?
Validation →