15-Minute Spike in Binance Bitcoin Perps Hits $14.6B Daily

Researchers Chan Kim and Peter Reinhard Hansen analyzed completed-trade records for perpetual futures on Binance and found recurring bursts of activity at 00:00, 15:00, 30:00 and 45:00 minutes of each hour. The pattern concentrated in the first ten seconds of those quarter-hour boundaries and appeared across Bitcoin, Ethereum, XRP, Solana, Dogecoin and Cardano during Jan. 1, 2021, to Oct. 31, 2024, a sample of 1,400 trading days.

Across the six contracts, the ten-second quarter-hour windows contained 26% more trades and 32% more dollar volume than comparable ten-second periods. Absolute returns, a measure of price movement regardless of direction, were 26% larger in those windows. During the sample, Bitcoin perpetual futures averaged about $14.58 billion in daily contract volume and roughly 1.54 million trades per day.

Perpetual futures, or perps, are contracts without a fixed expiration date that use recurring funding payments between long and short positions to keep perp prices close to spot indexes. The researchers note that perps account for a large share of global crypto trading and can affect arbitrage, hedging and market making.

To test whether automated systems were driving the bursts, the authors measured trade-size roundness. Human traders often choose round quantities, while algorithmic orders tend to produce irregular sizes tied to formulas. The study found round trade sizes became less common in the opening seconds of the quarter-hour bursts. For Bitcoin trades eligible to end in at least two zeros, the round-size share fell by 0.04 standard deviations at ordinary minute openings and by 0.20 standard deviations at the top of the hour.

The authors ran robustness checks. Removing windows that coincided with Binance funding payments at 00:00, 08:00 and 16:00 UTC did not eliminate the quarter-hour pattern. A separate analysis of Bybit trade data revealed a similar minute-of-hour structure on that exchange. The paper links the pattern to common exchange data formats, chart defaults and automated strategies that use standard candle intervals such as one-, five- and 15-minute bars.

The researchers tested predictability using a rolling model that used prior quarter-hour returns and standard price and volume indicators. The model correctly forecasted the direction of the ten-second return 56.6% of the time, had an out-of-sample R-squared of 3.4% and an area-under-the-curve score of 0.60. Trading on the model’s signals at every quarter-hour produced an average gross return of 0.51 basis points per trade, about $0.51 on a $10,000 trade.

Trading costs exceeded the model’s average gross return. During the sample period, Binance’s base taker fee was 5 basis points and maker fee 2 basis points, making a $10,000 taker trade cost about $5 to open. The authors note that the statistical pattern is smaller than typical fees, and they identify potential responses by large traders and market makers such as widening spreads or timing executions away from crowded boundaries.

The paper also reports that buyer-initiated volume exceeding seller-initiated volume at quarter-hour boundaries was associated with returns over the next four to 12 hours. The authors used block-bootstrap methods to account for dependent data and stated that trade records alone cannot reveal whether initiating orders contained private information or were synchronized responses to shared public inputs.