Graveyard / Order flow / Binance perps
VPIN and order flow toxicity
Can VPIN, a measure of informed trading, tell you when to trade Bitcoin?
The idea
VPIN estimates the share of trading volume coming from informed traders. The idea is to use it as a signal, or to trade only when it is low (calm, uninformed flow).
How we tested it
| Data | BTCUSDT perpetual futures. Test 1: VPIN built on $50M volume buckets as an entry signal. Test 2: fade large moves only when VPIN is low (gate values of at most 0.05 or 0.10, and others), 18 configurations that produced trades. |
|---|---|
| Period | Not stated. |
| Costs | Not stated. |
| Holdout | Full gate. |
| Control | Filtered against unfiltered in 9 comparisons. |
Result
| VPIN as a direction signal | noise |
|---|---|
| Large-move fade with VPIN filter: configurations passing | 0 of 18 that traded |
| Tightest two VPIN gates (0.05, 0.10) | zero trades |
| Filtered versus unfiltered | worse in 9 of 9 comparisons |
Why it failed
VPIN measures how strong informed flow is, not which direction it points. A large move mechanically raises VPIN while it is happening, so 'a big move with low VPIN' is a nearly empty and adversely selected set. A filter cannot make an unprofitable base signal profitable.
What would make it worth retesting
A leading, not concurrent, measure of informed flow, and only once the base fade clears a profit factor of 1.0 on its own.
Past results of a backtest say nothing certain about the future.