The project started with a classic idea: on perpetual futures, the funding rate is a live read of how crowded one side of the market is. When everyone is long and paying to stay long, a squeeze should follow.
Funding crowding
- On 2024–25 majors it looked real: +0.39% per trade, t = 2.43 — but it was one hit in 24 variants, exactly what chance produces.
- Funding then collapsed: since late 2025 it sits pinned at its baseline on the majors, and 61% of all perps sit exactly at baseline. The signal stopped firing.
- On smaller perps where funding still moves, the edge was not significant and turned negative with realistic position limits.
- The highest-funding coins had no spot market to hedge against — the hedgeable coins were precisely the low-funding ones.
Momentum, with a pre-registered holdout
Cross-sectional momentum across 232 perps produced drawdowns of −60% to −87% and no significant variant. Trend following on BTC — long while price is above a slow moving average, flat otherwise — was selected in discovery and run once on the holdout, a bear market:
Trend ruleBTC buy-and-hold
Total return−3.6%
−31.9%
−31.9%
Worst drawdown−21.4%
−53.0%
−53.0%
| Holdout | Return | Worst drawdown |
|---|---|---|
| BTC trend rule | −3.6% | −21.4% |
| BTC buy-and-hold | −31.9% | −53.0% |
The trend rule did what trend following is supposed to do in a falling market: it lost far less. Every moving-average length from 10 to 200 days beat buy-and-hold in the holdout, so the result was not tied to one parameter.
It became half of the final strategy.