Funding as a signal: does it work? A backtest on 20 coins

Setup

• Data: the 20 most-traded Binance USDT-M perpetuals as of 2021-01-01, the same set as in the golden cross study, 4h bars, 2020-01-01 → 2026-10-01. The funding rate is the last settled 8-hour rate.
• Two events: hot funding, above 0.05% per 8 hours — longs paying a lot — and negative funding, below 0 — shorts paying. 15,842 hot bars and 56,529 negative bars in total.
• Statistics: 40 Research Lab scans (20 coins × 2 events): the mean return 1, 3 and 7 days after the event against the coin's usual return over the same horizon, with a multiple-testing correction (q-value), a rotation test that respects overlapping windows, and a split into two halves of the period. Then one job on BTC that held back the last 30% of the period, repeated the test on 12 other coins and over 8 rolling windows.
• Trading: two fixed rules, not tuned — short when funding is hot, long when it is negative, exit after 42 bars (7 days) — at 1x, isolated margin, 0.05% fee and 0.05% slippage per side, funding charged and paid.

After hot funding, prices kept rising

The fade idea says a crowd paying high funding is about to be wrong. On 17 of 20 coins the 3 days after hot funding were better than usual, not worse — and none of the 20 was significant on the hot side except EGLD, whose two halves of the period disagree. Hot funding is a feature of bull markets: 79–95% of each coin's hot bars fall in the first half of its history.

Negative funding was followed by a better-than-usual 3 days on 14 of 20 coins and a better 7 days on only 11 of 20. It passed the correction on two coins: BTC — +1.22% over 3 days against the usual +0.44%, q 0.02, both halves positive — and THETA. Three of the 40 scans produced a significant finding — about the four that chance alone gives at this threshold.

Does the funding level predict the next move?

Read as a number rather than an event, funding passed the correction in 0 of 60 tests (20 coins × 3 horizons). On BTC the relation was slightly negative in all 8 rolling windows — higher funding, slightly weaker next days — but it was not significant in-sample, the platform found no agreement on the held-out 2024-09 → 2026-10 tail, and only 1 to 3 of 12 other coins showed the same sign.

Trading it

Shorting hot funding made money on 2 of 20 coins and finished below -90% on 12. It collects funding while it waits, but hot funding comes with manias: a 1x short is liquidated when the price doubles, and that happened on 10 coins. Buying negative funding did better — 9 of 20 in profit, BTC +333.12% — but on the 13 coins that rose it beat simply holding only once (BCH).

Backtest

Funding fade: short when funding > 0.05%, hold 7 days

BTCUSDT · 4h · Futures · 2020-01-01 → 2026-10-01

Return

+6.89%

Max drawdown

63.97%

Sharpe

0.18

Win rate

50.0%

Trades

42

Open the report →

Backtest

Funding fade: long when funding < 0, hold 7 days

BTCUSDT · 4h · Futures · 2020-01-01 → 2026-10-01

Return

+333.12%

Max drawdown

40.84%

Sharpe

0.77

Win rate

51.6%

Trades

155

Open the report →

Backtest

Funding fade: short when funding > 0.05%, hold 7 days

XRPUSDT · 4h · Futures · 2020-01-01 → 2026-10-01

Return

-100.00%

Max drawdown

100.00%

Sharpe

-0.86

Win rate

57.9%

Trades

57

Open the report →

Conclusion

• An extreme funding rate on its own did not predict a reversal on Binance perpetuals in 2020–2026 — after hot funding, prices more often kept rising.
• Fading the crowd paying hot funding was a way to get squeezed: losses on 18 of 20 coins, liquidations on 10.
• Negative funding on BTC was the one statistically solid finding, and it did not replicate across coins or survive the held-out tail — treat it as a hypothesis, not a signal.

What happens in the last hour before a settlement — do traders sell to avoid paying? That is a separate test: the hour before funding.

More research: DCA or lump sum and the golden cross on the top-20 coins.