Research · test #6

Does extreme funding predict reversals? No edge

The idea: when funding is very high, too many traders are long, so price should fall - and a short even gets paid funding while it waits. When funding is very negative, the reverse. We wrote the rules down first, then checked every funding payment on 522 Binance coins since January 2024. Neither side has an edge: shorting high funding wins often and small, then a squeeze takes it all back; buying negative funding mostly loses.

In short: 1.8 million funding payments, 2024-01-01 to 2026-10-20. Short when funding is +0.10% per 8 hours or more: 8,560 trades, 55% winners, median +0.70%, average -0.33% after costs and funding; on the second half of the data -0.51%. Long when funding is -0.10% or lower: 16,439 trades, 43% winners, average +0.07%; second half 0.00%.

The rules (fixed before the run)

• Coins: Binance USDT-M perpetuals on crypto (tokenized stocks and commodities excluded) that were trading on 21 October 2026, hourly candles and the full funding history.
• Signal: each funding rate is converted to 8 hours (coins pay every 1 to 8 hours). Short if the rate is +0.10% per 8 hours or more, long if it is -0.10% or lower.
• Entry at the close of the hour in which the payment happened - the moment the rate is known. Exit after 24 hours (main) and 72 hours (secondary). One trade per coin and side per 24 hours.
• Result = price move in the trade's direction + every funding payment during the hold (a short receives positive funding, a long receives negative) - 0.15% for fees and slippage.
• Control: the same position on the same coins at every funding payment in the period, 10,000 random samples of the same size.
• Edge requires: average above zero, t above 2 and p below 0.05 against the control - on the first part (to 30 June 2025) and the second part (from 1 July 2025) separately.

Result: no edge on either side

The short looked promising on the first part (59% winners, p 0.017), but its t of 1.00 is far from 2 and on the second part it lost money. The long beat the control on the first part only because long positions in general lost money then; its own average was not distinguishable from zero (t 1.83), and on the second part it was exactly zero.

Shorting high funding: small wins, rare squeezes

Every 24-hour short after funding of +0.10% or more, after costs and funding:

Most shorts land within 5% of zero, and slightly more of them on the winning side. The left tail is where the money goes: 96 trades lost more than half the position in a single day, 21 lost more than all of it, the worst -318%. The average win was +7.2%, the average loss -9.5%.

Drop the worst 1% of trades and the short turns positive; drop the worst 5% and it looks like a strategy. But nobody knows in advance which trade will be the squeeze. With leverage, each of those days is a liquidation (see how fast you would get liquidated).

A more extreme rate does not help

The most crowded longs - funding of +0.30% per 8 hours and above - were the worst coins to short. One possible reason, not tested here: funding that high comes with a strong trend, and the trend outran the funding the short collected.

What else we looked at

• Bitcoin direction. A short after Bitcoin had fallen over the previous 24 hours averaged -0.76% (3,516 trades); after Bitcoin had risen, -0.02% (5,044). A long after Bitcoin had fallen averaged +0.22%, after it had risen -0.09%. None of these pass.
• The 72-hour exit. Shorts were worse (second part -1.77%, worst trade -1,258%). Longs on the second part averaged +0.72% (t 3.12), but on the first part only +0.33% (t 1.38), and 72 hours was not the main exit. It is a candidate for a separate test with its own rules written in advance, not a result.

What this test does not say

• Coins delisted before 21 October 2026 are missing. Those coins may have had extreme funding more often, so the result for them could differ. We cannot say by how much.
• Entries are at the hourly close. In a fast move, slippage can be larger than the 0.15% we deducted.
• Trades on the same coin follow each other and are not independent; that is why the number of coins is shown next to every line.
• Funding combined with something else - open interest, a breakout, liquidations - is a different strategy. This test is the funding rate on its own.

Data and method

Rules were committed before any data was loaded; the code and every trade are kept with the project. Data: Binance public futures funding history and hourly candles. The same approach as in our other tests: a random control from the same coins and period, two time periods checked separately, costs included. Live rates: funding rates on four exchanges, the funding heatmap, Bitcoin funding rate history; and the funding cost calculator.

FAQ