generated 19 Nov 2026 11:26 WIB
Is a thin order book followed by bigger BTC moves, and by more grid stops?
FRAME
- Subject: BTCUSDT USDT-M perpetual: order book depth (Binance Vision daily
bookDepth) and 1-minute klines; the neutral grid of grid-neutral (lib/grid.py).
- Why it matters: a grid loses at a stop, when price runs away. A thin book lets price move further on the same orders; if thin books come before bigger moves, a grid could pause or widen while the book is thin.
- Scope: 2025-04-23 to 2026-02-01; the depth feed is frozen 2025-04-16..2025-05-18 (one value all day; minutes identical to the previous one are dropped), so depth starts 2025-05-19. 2026-02-01..2026-08-31 unused by this line; 2026-09-01..2026-10-31 reserved for its decisive bet. Descriptive, no claim. Method locked in
docs/bets/2026-11-18-depth-baseline.md, commit 1b347ff, written before any depth data was loaded. Grid parameters: step 0.5% (median BTC 1h range before 2026-02-01: 0.46%); half-range 3%, 13 levels (median 24h range 2.63%; LB6 median hold 24h); take-profit one step; stop one step beyond the far end; at most 2 resting open orders; maker fee 0.02%, taker fee 0.05%; 1,000 USDT capital, leverage 1; re-anchors at the current price when an executor ends.
- Terms: depth = bids + asks resting within 1% of the price, million USDT, last snapshot before the hour (or executor start), at most 5 minutes old. Next-hour range = (high - low) / open of the hour, %. Recent range = mean hourly range of the previous 24 hours. Rank corr = Pearson correlation of ranks. Executor, stop as in grid-flow-baseline. CI = 2,000 bootstrap resamples, seed 0.
MEASURES
By fifth of depth (all hours):
Within thirds of recent range (fifths of depth recomputed inside each third):
Grid executors (BUY and SELL together) by fifth of depth at start:
Fifths of depth for executors are cut on executor start times, so their bounds differ from the hour fifths above. By side:
Post-hoc (after the measures above were seen): partial rank correlation of depth with next-hour range, holding recent range fixed by a straight-line fit on ranks: -0.05. Grid executors by fifth of recent range at start (the last full hour's 24-hour mean range):
UPDATE
- Status: MEASURED.
- What it means: depth vs next-hour range -0.28 overall, -0.13 to -0.04 once recent range is held roughly fixed; recent range alone +0.51. Grid stop share, thinnest fifth minus the rest: +7.2 pp (-9.2 to +25.0).
- Next bets:
grid-step-by-recent-range (Is a grid whose step follows the last 24 hours' range more efficient than a fixed-step grid?)
- Rejected:
grid-pause-thin-book (before 2026-02-01 grids started in the thinnest fifth of depth were stopped 7 pp more often than the rest, 95% CI -9 to +25, with no order across the fifths; depth mostly mirrors the last 24 hours' range, which a grid can read from price alone (decided 2026-11-18))