Graveyard / Order flow / Binance perps
Cumulative volume delta (CVD) strategies
Does cumulative volume delta, buying volume minus selling volume, predict Bitcoin?
The idea
CVD adds up the difference between market buys and market sells. Traders read bursts as aggression and divergences with price as exhaustion.
How we tested it
| Data | BTCUSDT perpetual futures. Construction 1: a delta-burst z-score on 1-hour bars with fixed take profit and stop. Construction 2: a CVD divergence from price on 1-minute bars, to catch exhaustion. Construction 3: CVD z-score momentum. |
|---|---|
| Period | Not stated. |
| Costs | Not stated. |
| Holdout | Full gate; a no-lookahead correlation diagnostic before trading. |
| Control | None stated. |
Result
| Delta burst (1 hour) | real edge, but fails profit factor and Sharpe with fixed exits |
|---|---|
| Wider-stop retest | also failed |
| Exhaustion divergence | in a bull market high CVD was continuation, so shorts fired in uptrends |
| CVD momentum | only noise-level correlation; fires too often |
| Distinct constructions tried | 3 of 3 failed |
Why it failed
The first version found a real signal but lost it to fixed take-profit and stop levels. The second read strong buying as exhaustion when it was continuation, so its shorts were sold into rallies. The third was noise. The notes close the family: three different constructions, three different reasons.
What would make it worth retesting
Only with exits driven by the signal itself instead of fixed take-profit and stop levels. The notes state it is not viable as it is.
Past results of a backtest say nothing certain about the future.