Why open interest failed our test.
Open interest is one of the most intuitive crypto signals — more leverage piling on should mean something. We tested it properly, and dropped it. Here is exactly what we found, with the numbers.
Why open interest looked promising
Open interest (OI) is the total value of futures positions open on a coin. The intuition is seductive: when leverage stacks up fast, the coin is fragile — crowded, prone to liquidation cascades. If funding tells you which side the crowd is on, OI tells you how big the pile is. A natural second factor.
And unlike most candidates, it cleared the first bar: its cross-sectional correlation with funding crowding was only about 0.04. So OI was a genuinely different axis — not just funding wearing a different hat. That’s rare and worth chasing.
How we tested it
We backfilled daily open interest from the public Binance metrics archive (2023 onward) and ran it through the same gauntlet every candidate faces:
- Non-overlapping 14-day windows — overlapping windows inflate significance; we refuse that trap.
- Marginal information over funding + momentum — the real question isn’t “does OI predict returns?” but “does it add anything our existing signal doesn’t already know?”
- Multiple-testing control and a fragility check — re-run on a fresh data pull and see if the result holds.
The numbers that killed it
Standalone, OI showed only a faint tilt. The decisive test — its marginal information after controlling for funding and momentum — came back statistically insignificant, around t ≈ −1.6 to −1.8. Not nothing, but nowhere near a bar you’d build a score on.
Worse, it was fragile: the raw significance flipped between p = 0.04 and p = 0.16 on a routine data refresh. A signal whose verdict changes when you re-pull the data is not a signal — it’s noise that happened to line up once.
Independent of funding, yes. Reliably informative beyond it, no. So open interest does not enter the IOX score.
Dropped from the score — kept as context
Open interest moved to the display layer: useful market-structure context to show, never a driver of the calibrated read. That’s our three-layer rule — a metric can be worth showing without being worth scoring.
What would make us reconsider
- A richer OI construction — rate-of-change, or OI-versus-price divergence — that we’re accumulating live data for now.
- A clean result that survives the same non-overlapping, marginal-over-funding, fragility-tested gauntlet — published here either way.
This is the same fate that met momentum, low volatility, on-chain activity, DeFi TVL and illiquidity. One factor passed: funding crowding. The pile of rejections is the point — it’s why the survivor is worth trusting.
Common questions
Did open interest predict crypto returns at all?
On its own it showed a faint tilt, but once we controlled for funding crowding and momentum, the marginal information was statistically insignificant (t ≈ −1.6 to −1.8 on non-overlapping 14-day windows) and fragile — its raw significance flipped between p = 0.04 and p = 0.16 on a routine data refresh. That is not an edge we would stake the score on.
Is open interest just the same thing as funding?
No — and that was the interesting part. Their cross-sectional correlation was about 0.04, so open interest is genuinely a different axis (size of positioning) than funding (the skew of positioning). It failed not because it duplicated funding, but because as a standalone signal it was simply weak.
So is open interest useless?
Not useless — just not score-worthy. It is useful market-structure context, so its honest home is a display layer, not the IOX score. We may revisit a richer open-interest signal (rate of change, OI-vs-price divergence) once enough live data has accumulated.
Why publish a factor that failed?
Because the discipline that kills weak candidates is the product. Most crypto tools hide their failures and ship a 30-indicator dashboard. We show the gauntlet so you can trust the one signal that passed it.
Ioxer is research, not investment advice. IOX is a crowding read — not a price prediction, not a buy/sell signal.