Does taker flow when a grid starts say how that grid will end?
FRAME
- Subject: the neutral BTC grid of grid-neutral (BUY and SELL grids side by side, 500 USDT each,
lib/grid.py) on Binance Vision BTCUSDT USDT-M perpetual 1-minute bars, 2025-04-23 to 2026-02-01 (284 days); taker flow from the same klines. - Why it matters: a grid loses when price runs away from it (a stop). This maps whether one-sided taker flow at the moment a grid starts lines up with how it ends, before any bet on waiting for balanced flow.
- Scope: before 2026-02-01 only; holdout untouched here, but already used once by grid-strike-lb6, so a later grid bet on it would not be blind. Descriptive, no claim. Method locked in
docs/bets/2026-11-18-grid-flow-baseline.md, commitf03b57e, written before computing. 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: executor = one grid from start to end (take-profit exit or stop; the one still running at the window end per side is left out). Taker-buy share at start = taker-buy quote volume / quote volume over the 60 minutes before start (one grid step, 0.5%, is BTC's median 1-hour range). Against flow = BUY grid started while share < 0.5, SELL grid while share > 0.5. Result = realized net USDT of the executor on 500 USDT, fees and market close included. CI = 2,000 bootstrap resamples, seed 0.
MEASURES
By side:
| side | executors | stop_pct | mean_pnl | total_pnl | median_hours |
|---|---|---|---|---|---|
| BUY | 103 | 47.6 | -0.484 | -49.89 | 32.5 |
| SELL | 103 | 45.6 | +0.255 | +26.22 | 45.4 |
With / against flow at start:
| side | flow | executors | stop_pct | mean_pnl | total_pnl | median_hours |
|---|---|---|---|---|---|---|
| BUY | with | 49 | 44.9 | +0.080 | +3.93 | 47.9 |
| BUY | against | 54 | 50.0 | -0.997 | -53.81 | 25.5 |
| SELL | with | 61 | 44.3 | +0.432 | +26.38 | 41.3 |
| SELL | against | 42 | 47.6 | -0.004 | -0.15 | 48.1 |
- BUY: stop share against minus with +5.1 pp (95% CI -14.6 to +24.9); mean result against minus with -1.077 USDT (95% CI -3.896 to +1.488).
- SELL: stop share against minus with +3.4 pp (95% CI -15.3 to +23.6); mean result against minus with -0.436 USDT (95% CI -3.136 to +2.353).
By quintile of taker-buy share at start:
| side | quintile | share_range | executors | stop_pct | mean_pnl |
|---|---|---|---|---|---|
| BUY | 1 | 0.327-0.444 | 21 | 47.6 | -0.311 |
| BUY | 2 | 0.447-0.470 | 20 | 50.0 | -2.418 |
| BUY | 3 | 0.470-0.533 | 21 | 47.6 | +1.345 |
| BUY | 4 | 0.534-0.565 | 20 | 45.0 | -1.114 |
| BUY | 5 | 0.566-0.668 | 21 | 47.6 | -0.045 |
| SELL | 1 | 0.327-0.426 | 21 | 33.3 | +1.291 |
| SELL | 2 | 0.428-0.462 | 20 | 55.0 | -0.645 |
| SELL | 3 | 0.462-0.501 | 21 | 42.9 | +0.622 |
| SELL | 4 | 0.507-0.560 | 20 | 45.0 | +0.459 |
| SELL | 5 | 0.565-0.611 | 21 | 52.4 | -0.487 |
UPDATE
- Status: MEASURED.
- What it means: grids started against flow ended in a stop +5.1 pp (BUY) and +3.4 pp (SELL) more often than with flow; every CI above includes zero, and stop share has no trend across flow quintiles. Flow at start does not separate grids that end well from grids that end badly on this data.
- Rejected:
grid-wait-balanced-flow(before 2026-02-01 grids started against flow ended in a stop only 3 to 5 pp more often, with 95% CIs about 20 pp wide on each side, and stop share showed no trend across flow quintiles; the only holdout left for grids was already used by grid-strike-lb6, so a test would not be blind either (decided 2026-11-18))