Crypto perpetual futures trade without a closing bell. The tape never stops. Yet a new paper suggests the market still behaves as if it has one.
According to research cited by TechFlowPost, perpetual contracts on Binance show a recurring surge in activity at fixed times each hour. At 0, 15, 30 and 45 minutes past the hour, trading intensifies, turnover rises and short-term price swings widen, even when no news hits the market. Smaller versions of the same pattern appear at 5-minute marks and at the start of each minute, while the top of the hour produces the strongest burst.
The paper argues that crypto has developed its own version of an opening bell, not because trading pauses and resumes, but because software, chart defaults and automated execution systems carve a continuous market into repeated micro sessions.
A four-year sample across six Binance perpetuals
The study was conducted by Korean policy researcher Chan Kim and Peter Reinhard Hansen of the University of North Carolina. It uses tick-level trade data for six Binance perpetual futures contracts from Jan. 1, 2021 through Oct. 31, 2024, covering 1,400 full days of uninterrupted trading.
The contracts in the sample are Bitcoin, Ethereum, XRP, Solana, Dogecoin and Cardano.
The paper focuses on perpetual futures, which allow traders to bet on price moves and use leverage to expand position size. Unlike traditional futures, perpetuals have no fixed expiry date. Positions can stay open as long as margin is sufficient, while periodic funding payments between longs and shorts keep contract prices close to the spot index.
When a perpetual trades above the spot index, long traders pay funding to shorts. When the contract trades below spot, shorts pay longs.
Because perpetuals account for a large share of global crypto trading, the authors say these short bursts matter beyond the futures book itself. Prices in perpetual markets guide cross-exchange arbitrage, hedging and market making, so turbulence in futures can spill over into spot trading for Bitcoin and other assets.
The jump is concentrated in the first 10 seconds
When the researchers mapped one hour into a circular chart, the pattern stood out clearly. Distinct peaks appeared at minute 0, 15, 30 and 45, with most of the surge packed into the first 10 seconds after each boundary.
Across the six contracts, that 10-second window showed:
- a 26% increase in trade count compared with normal periods
- a 32% increase in dollar-denominated trading volume
- a 26% increase in absolute price movement
The absolute return measure captures two-way price movement rather than direction. In other words, the market becomes more volatile in those opening seconds of a 15-minute interval whether prices move up or down.
The paper’s polar charts also show how absolute returns and trading volume behave at different minute marks within the hour for BTC, ETH, XRP, SOL, DOGE and ADA perpetuals.
The pattern holds across assets with very different scale. During the sample period, Bitcoin averaged 1.54 million trades per day and $14.58 billion in contract volume, while Cardano averaged about 290,000 trades and $544 million. Even so, the timing of their activity was highly aligned.
That cross-asset consistency is one of the paper’s central findings. The researchers argue it points to a market-wide trading mechanism rather than a feature unique to any single token.
How candlestick intervals shape a continuous market
The proposed explanation is straightforward. Most trading platforms package a nonstop stream of prices into standardized candlesticks such as 1-minute, 5-minute and 15-minute bars.
A 15-minute candle summarizes the opening price, closing price, high and low over that interval. That format helps human traders read the market. It also gives algorithmic systems a standardized unit of data.
At the end of each candle, technical indicators are recalculated and automated strategies refresh their orders using the latest completed bar. Algorithms that split large orders may send the remaining slices at the boundary. Market makers may adjust quotes based on expected flows. Faster quantitative systems may position ahead of the shift.
Once enough programs rely on the same time grid, what began as a charting convention becomes part of price formation itself. A quiet 15-minute stretch can end with a burst of trading that resembles the opening auction in traditional markets.
In equities or futures with fixed sessions, the opening bell gathers orders that accumulated during the closure. Crypto has no such pause. The paper’s point is that a common bar structure and default software settings recreate similar moments of crowding every 15 minutes, without the market ever stopping.
Order-size patterns hint at machine-driven activity
Binance trade data shows the instrument, the amount traded and the execution price, but it does not identify whether an order came from a human trader, a market-making firm, a liquidation engine or another automated process. Kim and Hansen therefore looked for indirect evidence in order size.
Their reasoning is that human traders tend to prefer round numbers, such as 0.1 BTC or an order value close to $10,000. Algorithms, by contrast, often size trades from volatility, available capital, current exposure or large-order execution targets, producing quantities that look irregular from a human perspective.
The researchers measured how often order values ended in zeros. During the few seconds after a pulse begins, the share of round-number orders fell noticeably.
To avoid distortion from exchange minimum order sizes, they limited the sample to sufficiently large trades so that small tickets would not be misclassified as non-human activity.
The effect strengthened as the time boundary became more important. The share of round-number orders slipped slightly at the start of an ordinary minute, fell more at 5-minute marks, dropped further at 15-minute marks and showed the biggest divergence at the top of the hour.
For Bitcoin orders meeting the study’s double-zero criterion, the round-order share deviated from its normal level by 0.04 standard deviations at ordinary minute openings. At the top of the hour, the deviation reached 0.20, five times larger.
The paper is careful on interpretation. A standard deviation does not measure the exact share of machine orders. It only shows that when activity jumps, the market becomes less dominated by habitually round submissions, which is consistent with heavier automated participation.
Order size alone cannot identify the source of each trade. Institutional slicing, forced liquidations and funding-rate arbitrage can all generate irregular values too. The authors present this evidence only as an indirect confirmation that quantitative trading becomes more active in pulse windows.
The effect remains after funding-settlement windows are removed
The authors ran several checks to rule out other periodic events as the source of the pattern.
During the sample period, Binance settled funding at 00:00, 08:00 and 16:00 UTC. After excluding those windows, the 15-minute pulse remained statistically significant. Even after removing all observations at the top of the hour, the features at 15, 30 and 45 minutes were still present.
An independent analysis of Bybit data produced a highly similar result.
The paper says this supports the view that the pattern reflects a broad form of electronically synchronized trading. Traders could in principle choose any timing convention, but exchange data structures, chart settings and common technical indicators steer many programs toward the same temporal boundaries. The most watched boundaries, the top of the hour and the quarter-hour marks, attract the greatest concentration of flow.
A detectable signal, but not enough to beat fees
After establishing periodicity, the research team tested whether pre-window market data could predict the direction of returns in the 10 seconds after each 15-minute boundary.
The rolling forecast model used prior 15-minute returns together with classic price-and-volume indicators, relying only on information available at the time to generate out-of-sample forecasts.
Across the six contracts, the backtest found:
- 56.6% directional accuracy
- an average out-of-sample R-squared of 3.4%
- an area-under-the-curve score of 0.60, where 0.5 is random guessing and 1.0 is perfect prediction
In a noisy 10-second market, those numbers indicate that the pattern carries repeatable signal content. But the economic value is limited.
If a trader followed the model and traded at every 15-minute boundary, the average gross return before fees was only 0.51 basis points, or 0.0051%, per trade. On a $10,000 position, that comes to roughly $0.51 in gross profit.
During the sample period, Binance’s base taker fee was 5 basis points and its maker fee was 2 basis points. A $10,000 taker order would cost about $5 to open, and closing the position would incur another fee. The model’s average gross edge was less than one-tenth of the entry fee alone.
That gap, the paper argues, is one of the most important takeaways in the dataset. Statistical predictability does not translate neatly into profits that ordinary traders can actually capture. A market can exhibit a repeatable short-term pattern and still remain difficult to arbitrage once basic trading costs are included.
Execution and market-making desks may still benefit
The authors say the findings can still help market makers and large institutions improve execution.
Liquidity providers quoting both sides of the book may widen spreads during the 10-second pulse window. If they expect one-way flow, they may also reduce quoted size. Institutions working large orders may choose to avoid crowded time boundaries to reduce slippage caused by their own execution.
The paper also says the first 10 seconds of a 15-minute window may carry information about longer-horizon moves. If aggressive buy volume exceeds sell volume during a quarter-hour boundary, that order imbalance often coincides with upward price pressure over the next 4 to 12 hours. When selling dominates, medium-term price action tends to stay under pressure.
Order imbalance here means the difference between aggressive buy and aggressive sell volume relative to total turnover in that window. It is used to measure which side is pushing harder.
At the 4-hour horizon, the paper says medium-term returns often inherit the flow signal released at the earlier 15-minute boundary. At 8-hour and 12-hour horizons, traditional price-and-volume indicators carry more explanatory power. The authors say this is consistent with the idea that quantitative systems use the 15-minute interval as a common signal checkpoint for information absorbed across the market.
They also warn that the longer-horizon result should be treated carefully. The 4-hour, 8-hour and 12-hour return windows overlap, so a single large move can appear repeatedly in several statistical groups. Although the paper uses block bootstrap methods suitable for non-independent data, aggregated trade records still cannot tell whether orders carried private information, whether they responded to the same public news, or whether a move simply reflected market makers absorbing a large one-way trade.
A market with no close still creates its own open
Strip away the statistical machinery and the result is simple.
Crypto removed the closing bell and made trading continuous. But APIs, candlestick intervals and automated strategies have built countless mini opening moments back into the day. Every 15 minutes, thousands of independently running programs reach the same temporal checkpoint. In a matter of seconds, a market designed never to stop begins to look as if everyone is trying to pass through the same door at once.
