Graveyard / Order flow / Binance perps
Order book imbalance trading
Does a lopsided order book predict the next price move?
The idea
If much more size is waiting to buy than to sell, price should tick up. The idea extends to large resting orders ('walls') that may attract or block price.
How we tested it
| Data | Test 1: order book imbalance of at least ±0.35 across the top 20 depth levels on BTCUSDT perpetuals. Test 2: the same on 7 altcoin perpetuals (SOL, XRP, ZEC, DOGE, BNB, ADA, 1000PEPE) by 20-minute polling. Test 3: detect large resting orders on BTCUSDT and trade the next 2 hours. Binance does not keep historical depth, so these ran on data collected as it happened. |
|---|---|
| Period | Not stated. |
| Costs | Not stated. |
| Holdout | Full gates for tests 1 and 3. |
| Control | None stated. |
Result
| BTC imbalance | noise, no persistence |
|---|---|
| Altcoins: decay with horizon | no consistent pattern |
| Thinnest altcoin (1000PEPE) | weakest signal |
| Large-order wall: time-exit rate | over 2.5x the weekly cap |
Why it failed
In a tight, ranging market the tilt of the book is momentary and does not reflect lasting pressure. The thinnest coin, where a thin book should matter most, showed the weakest signal, which contradicts the idea that thin books carry strong signals. A large resting order does not predict where the price goes two hours later.
What would make it worth retesting
Only with a regime filter, a tighter profit target matching how long the signal persists, and a clean account; for walls, an exit tied to the wall being eaten rather than a fixed clock, and independent evidence first. For altcoins: a multi-day, multi-regime dataset and a revised theory of thin books.
Past results of a backtest say nothing certain about the future.