AI-generated research, checked but not peer-reviewed. It may contain errors.

E886　 KILL　 Not replicatedTested · October 2026

A patient maker resting 2% off the index loses once fills are verified

Resting orders 2% from the index price on thin crypto perpetuals earned +107.5 bp per fill on candle data, but with tick-verified, size-capped fills the dollar-weighted result is -14.2 bp (interval -39.7 to +14.0 bp).

market-making liquidity-provision adverse-selection fill-model thin-markets perpetuals

Readable summary Full research record JSON recordFeedback on E886

Thin crypto perpetuals get hit by panicked and liquidated market orders that print far from the global price, so a patient quote resting 2% away should be paid for catching them. Every minute we posted a buy 2% below and a sell 2% above each coin's index price, and exited 15 minutes after a fill. On one-minute candles the selection sample earned +107.5 bp per fill. On validation data, 55.5% of the candle-based candidates (15-minute exit) had no book print through the level. With fills checked against individual trades and capped at $500, the dollar-weighted result was -14.2 bp, interval -39.7 to +14.0 bp.

Dollar-weighted net per fill-14.2 bp15-minute exit, $500 cap, n 1,960; 95% interval -39.7 to +14.0 bp

Candidates with no book print through the level55.5%2,440 of 4,400 candle-based candidates, 15-minute exit

Selection-sample candle result+107.5 bp186 candle fills in 11 coins; did not survive tick verification

Sample P&L at $500 per fill-$135/monthpre-registered economic gate: at least +$50/month

01 · The idea being tested

A patient quote resting far from the market on a thin crypto venue gets paid when forced sellers or buyers sweep the book, because the price snaps back.

Why people believe it

Leveraged retail traders on a small venue are liquidated or panic out with market orders. In a thin book those orders print well away from the global price. A resting bid far below the index, or an offer far above it, is filled at the extreme, and the perpetual then tends to converge back to its index.

Why it would persist The argument, as stated before testing

Depth on thin venues is supplied by a few market makers who do not quote deep, because a deep resting order can be run over by a crash that keeps going. The reward is meant to compensate for that risk.

02 · How it was tested

Every minute and for every coin in a random sample, we placed an imaginary buy 2% below the index price of the previous minute and an imaginary sell 2% above it. A one-minute candle that traded through the level suggested a fill. We then checked each suggested fill against the individual trades of that minute: only trades printed strictly through the level count, and the fill is capped at $500 or at the dollar volume that actually traded through the level, whichever is smaller. After a fill we closed the position 15 minutes later at the taker price, and in a secondary run 60 minutes later.

Pass required

Validation, 15-minute exit, $500 cap: dollar-weighted net at least +30 bp, coin-cluster bootstrap 95% interval above 0, equal-weight median above 0, at least 100 fills, at least 15 coins, top 5 coins under 60% of dollar P&L, and sample P&L at least $50 per month.

Killed if

Dollar-weighted net at or below 0, or the bootstrap interval's lower bound at or below 0.

Costs

Maker 2 bp on entry, taker 5 bp on exit, plus half the quoted spread at exit (floor 2 bp, taken from one tickers snapshot, so the same for the whole sample).

Controls

Comparison of candle-based candidates with tick-verified fills; decomposition of the selection-sample P&L into convergence to the index versus index movement. Coin-cluster bootstrap for intervals.

03 · What happened

Why it was rejected

The strategy failed both pre-registered tests: the dollar-weighted net is -14.2 bp (at or below 0) and the lower bound of its 95% interval is -39.7 bp.

In the 40-coin validation sample there were 8,677 candidate-exit pairs (4,400 with the 15-minute exit, 4,277 with the 60-minute exit; the same minute can appear in both). For the 15-minute exit, 2,440 of 4,400 (55.5%) had no book print through the level and would not have filled a resting order; 1,960 had one.
Primary cell (15-minute exit, $500 cap): 1,960 fills in 36 coins, dollar-weighted net -14.2 bp, interval -39.7 to +14.0 bp. Sample P&L at $500 per fill was -$135 per month, against a gate of +$50.
Equal-weighted, the same 1,960 fills averaged +4.3 bp with a median of +38.8 bp, but the worst fill lost 5,206 bp. Weighting by fill size turns the small average gain negative. Before the 7 bp of maker and taker fees, and after the exit half-spread, the dollar-weighted result is -7.2 bp.
The only size bucket with a positive result holds fills averaging $31, too small to matter; the whole strategy lost $135 per month at $500 per fill.
60-minute exit (secondary, n 1,911): dollar-weighted net +4.4 bp, interval -22.5 to +26.5 bp, far below the +30 bp pass line; the top 5 coins carried 390% of the P&L.
Selection sample (11 coins, 2026-03..2026-04, full one-minute data, candle-based fills): 186 fills, net +107.5 bp, median +151 bp, worst -1,026 bp, gross +114.5 bp, index move -32 bp. The gain came from the perpetual converging to its index. A tick check of 30 of these fills found a median of $99 traded through the level.

More than half the candle fills never happened

Candidates in the validation sample with the 15-minute exit where the one-minute candle traded through the level, and how many had a book print through the level in the trade records. The rest would not have filled a resting order.

Chart data (candidates)

Stage	Count
Candle traded through the level	4,400
Book print through the level	1,960

Dollar-weighted results never clear the pass line

Dollar-weighted net per fill with coin-cluster bootstrap 95% intervals, $500 cap. The 15-minute exit is the primary cell and is negative; the 60-minute exit is near zero. Both intervals include zero and sit far below the +30 bp pass line.

Chart data (bp)

	Mean	95% low	95% high
15-minute exit, primary (n 1,960)	-14.2	-39.7	14
60-minute exit, secondary (n 1,911)	4.4	-22.5	26.5

04 · What it means

Why it failed Interpretation · not established

The selection-sample result came from two things: fills read off one-minute candles, and equal weighting. A candle that dipped through the level does not mean a resting order was reached, and the prints that did reach it were often tiny. This is an interpretation; the test did not identify who traded in each print.

Lesson

A strategy that sells liquidity has to be tested on fills a resting order could have received, at the size it would trade. Candle highs and lows overstate fills, and an average that treats a $1 fill like a $500 fill measures something that cannot be scaled.

Scope of this result

Covers resting orders at 2% from the index price, a 15-minute and a 60-minute taker exit, a $500 and a $100 size cap, 40 randomly drawn crypto perpetuals on one retail-accessible venue, validation 2026-05..2026-10. It does not cover other distances, venues or instrument types.

A rejected experiment applies to the tested conditions and does not prove an idea can never work. bp = basis point, 0.01%.

1. Test design

Pre-registered

Yes: rule and criteria fixed before the test data was examined

Data

One-minute trade-price candles and index (spot) candles for linear USD-margined perpetuals — a retail-accessible crypto venue, public chart history; 1 minute; 2025-10..2026-10; public, anonymous | Individual trades (tick-level fills) with price, dollar value and order type, for each candidate minute — a retail-accessible crypto venue, public market history; tick; 2026-05..2026-10; public, anonymous

Universe

Tradeable crypto perpetuals on a retail-accessible venue, about 200 contracts; tokenised-stock and other traditional-finance perpetuals excluded. The decisive run used 40 coins drawn at random (fixed seed); 38 of them produced at least one candidate minute and 36 produced at least one fill.

Signal

Every minute m, resting buy at index close(m-1) x 0.98 and resting sell at index close(m-1) x 1.02. Candidate if the perpetual traded with volume and its 1-minute low (high) went strictly through the level; minutes where both levels were crossed are skipped. A candidate becomes a fill only if the trade records show a book print strictly through the level; trades the venue assigns off the book are excluded because a resting order could not receive them.

Execution

Fill price = the resting level; fill size = min(Q, dollar volume traded strictly through the level), Q = $500 primary and $100 secondary. Minutes with zero volume through the level are not fills. One position per coin, 60-minute cooldown counted from real fills. Exit as a taker at the close of the first traded minute at or after 15 minutes (60 minutes as secondary), minus half the quoted spread.

Costs

Maker 2 bp on entry, taker 5 bp on exit, plus half the quoted spread at exit (floor 2 bp, taken from one tickers snapshot, so the same for the whole sample).

Controls

Comparison of candle-based candidates with tick-verified fills; decomposition of the selection-sample P&L into convergence to the index versus index movement. Coin-cluster bootstrap for intervals.

Sample split

Selection 2025-10..2026-04, used only to design the tick-level run; validation 2026-05..2026-10, which decides. The design was fixed on selection data before the validation run.

Pass criterion

Kill criterion

Dollar-weighted net at or below 0, or the bootstrap interval's lower bound at or below 0.

Power

Not computed in advance. The primary interval, -39.7 to +14.0 bp, excludes true effects above about +14 bp; it does not exclude losses of up to about 40 bp.

2. Result

Observations

1,960

Instruments

Net mean

-14.2 bp

95% interval

-39.7 bp to +14 bp

Median

+38.8 bp

Worst

-5,206 bp

Finding 1

In the 40-coin validation sample there were 8,677 candidate-exit pairs (4,400 with the 15-minute exit, 4,277 with the 60-minute exit; the same minute can appear in both). For the 15-minute exit, 2,440 of 4,400 (55.5%) had no book print through the level and would not have filled a resting order; 1,960 had one.

Finding 2

Primary cell (15-minute exit, $500 cap): 1,960 fills in 36 coins, dollar-weighted net -14.2 bp, interval -39.7 to +14.0 bp. Sample P&L at $500 per fill was -$135 per month, against a gate of +$50.

Finding 3

Equal-weighted, the same 1,960 fills averaged +4.3 bp with a median of +38.8 bp, but the worst fill lost 5,206 bp. Weighting by fill size turns the small average gain negative. Before the 7 bp of maker and taker fees, and after the exit half-spread, the dollar-weighted result is -7.2 bp.

Finding 4

The only size bucket with a positive result holds fills averaging $31, too small to matter; the whole strategy lost $135 per month at $500 per fill.

Finding 5

60-minute exit (secondary, n 1,911): dollar-weighted net +4.4 bp, interval -22.5 to +26.5 bp, far below the +30 bp pass line; the top 5 coins carried 390% of the P&L.

Finding 6

Selection sample (11 coins, 2026-03..2026-04, full one-minute data, candle-based fills): 186 fills, net +107.5 bp, median +151 bp, worst -1,026 bp, gross +114.5 bp, index move -32 bp. The gain came from the perpetual converging to its index. A tick check of 30 of these fills found a median of $99 traded through the level.

3. Interpretation

Why it failed

Lesson

Scope

4. Reproducibility

Steps

1) Download one-minute trade and index candles for each coin from the venue's public chart history. 2) For every minute build the resting levels from the previous index close and flag minutes where the trade low (high) is strictly through a level with volume. 3) For flagged minutes pull the individual trades, keep book prints only, and sum the dollar volume strictly through the level; fill size = min($500, that volume). 4) Exit as a taker at the first traded minute at least 15 minutes later, minus half the quoted spread, and subtract 2 bp maker and 5 bp taker costs. 5) Weight by fill size, bootstrap over coins, and compare with the pre-registered criteria.

Grade

COMPLETE