▲ Work in progress. This site is built and researched in the open: pages, numbers and findings may change or be wrong, and the record has gaps. Not advice. More
The Lab
Ideas traders use, put to the record
← the Lab · signals tested · write-up
#64 Session momentum: do the first hours predict the last? No reliable edge
Past prices; no real orders were placed. Every figure is a share of the money, or a count.
Last checked: 18 October 2026.
Where the idea comes from
- The paper. Lei Gao, Yufeng Han, Sophia Zhengzi Li and Guofu Zhou, "Market intraday momentum" (Journal of Financial Economics, 2018). In US stocks, the first half-hour of the day (counted from the last close) pointed the way the last half-hour would go. A rule that took the last half-hour's side from the first did better than simply holding.
- Why it might work. Traders who hedge or rebalance late in the day, and late traders who react to the morning's news, can push the close the same way as the open.
- Patterns within the day were found before that in US stocks by Steven Heston, Robert Korajczyk and Ronnie Sadka, "Intraday patterns in the cross-section of stock returns" (Journal of Finance, 2010).
- The bar to beat is our page "Coin-flip traders: can luck beat the fees?". This is the first study in the series "Published edges against the coin": a published finding, tried on crypto, against traders who pick their side with a coin.
The question
Crypto trades all day and every day, but it still has busy sessions: Asia, London and New York. Do the first hours of a session point the way its last hours will go? And does that pay after the fees, on prices the rule hasn't seen, better than a coin?
How we tested
- Three coins: Bitcoin, Ether and Solana. Hourly futures prices (USDT perpetuals), 15 October 2022 to 15 October 2026, the same as the coin-flip traders.
- Four sessions, in UTC: Asia 00:00 to 08:00, London 08:00 to 16:00, New York 13:00 to 21:00, and the whole UTC day (the paper's own unit).
- A rule: the first 1, 2 or 4 hours of the session choose the side for its last 1, 2 or 4 hours:
- follow (the paper's momentum: up first, long at the end) or fade (the opposite);
- on every day, or only on days whose first hours moved more than usual (the middle, or the top quarter, of the first two years' moves). That is 54 rules for each coin and session, 648 in all. Every one was run.
- A trade: no leverage, in at the start of the last hours and out at their end. Fees: the exchange's market-order fee (0.055%) and its resting-order fee (0.02%), paid on the way in and on the way out.
- Picked on the first half, judged on the second. For each coin and session, the rule that did best from October 2022 to October 2024 at the market-order fee was then judged on October 2024 to October 2026, which it hadn't seen. The paper's own rule (the first hour, followed, for the last hour, every day) was judged on both halves, untouched.
- Two random twins, on the very days the rule traded:
- 10,000 coin-flip traders, each side a fair flip;
- 1,000 shuffles: the first hours' moves dealt out to other days, so the rule keeps its count of longs and shorts but loses the link.
- The bar, fixed before the run: a coin and session passes if its picked rule makes money on the second half at the market-order fee, and beats 95% of the coin-flip traders and 95% of the shuffles. An edge needs 3 or more of the 12 to pass. Luck alone gives about 0.6, and 3 or more about 2 times in 100.
The maths, step by step
1. The fee. In and out costs 0.055% twice, 0.11% of the trade. A rule that trades every day for two years makes about 730 trades: 0.9989 multiplied by itself 730 times is about 0.45, a loss of 55% from fees alone.
2. What the side must add. To break even, the side must earn more than 0.11% a trade on average, beyond what a coin-flip side earns.
3. The slope. The paper's own measure: draw a line through every day's (first-hours move, last-hours move). A slope above 0 is momentum, below 0 is a pull-back. Its t (the slope over its uncertainty) beyond about ±2 happens by luck about 1 time in 20.
4. Many tries. With 648 rules, about 32 would beat 95% of coin-flip traders on the second half by luck alone. So one rule beating the coin is not news. That is why the bar is 3 or more of the 12 picks, chosen before the test.
What we found
1. The paper's rule lost in every case. The first hour, followed, for the last hour, every day:
- On the second half it lost 47% to 64% at the market-order fee, and 12% to 40% at the resting-order fee, in all 12 coins and sessions.
- That is close to step 1's 55% from fees alone. The side added nothing: it beat the middle coin-flip trader in only 6 of the 12, and never beat 95% of them in either half.
- The paper's slope (first hour on last hour) was beyond ±2 in 5 of 24 cases (12 per half). 3 of the 5 pointed the wrong way, a pull-back, not momentum. The two that showed momentum didn't repeat: Solana in London (first half) and Solana over the UTC day (second half).
2. The best rules of the first half didn't last.
| coin and session | the rule picked | first half | second half |
|---|---|---|---|
| Bitcoin, Asia | first 4 h, fade, last 2 h | +5.1% | -10.2% |
| Bitcoin, London | first 1 h, fade, last 2 h | +1.3% | -35.4% |
| Bitcoin, New York | first 1 h, fade, last 1 h | -2.2% | -10.7% |
| Bitcoin, UTC day | first 1 h, follow, last 4 h | -2.5% | -9.6% |
| Ether, Asia | first 2 h, fade, last 4 h | +27.3% | -14.3% |
| Ether, London | first 1 h, fade, last 2 h | +3.5% | -32.1% |
| Ether, New York | first 4 h, follow, last 2 h | -6.7% | -34.4% |
| Ether, UTC day | first 4 h, follow, last 4 h | -8.8% | -32.6% |
| Solana, Asia | first 2 h, fade, last 2 h | +70.8% | +15.7% |
| Solana, London | first 2 h, fade, last 4 h | +51.7% | -26.2% |
| Solana, New York | first 4 h, follow, last 2 h | +14.6% | -19.1% |
| Solana, UTC day | first 2 h, follow, last 4 h | -2.5% | -2.4% |
At the market-order fee. Every pick traded only on days whose first hours moved more than three days in four (the top quarter).
Each cell's best rule on the first two years, then the same rule on the two years it had not seen.
- 1 of 12 passed: Solana in Asia, +15.7%, ahead of 98.6% of coin-flip traders and 99.7% of shuffles, on 119 trades. The bar needed 3. Luck alone gives about 0.6 a run, so 1 is what luck looks like.
- On the second half the picks averaged -17.6%, after +12.6% on the first.
- Across all 648 rules, 25 beat 95% of coin-flip traders on the second half. Luck alone would give about 32. Only 17 made money at the market-order fee.
- First-half winners didn't stay winners. Each rule's gain over coin-flip sides in the first half told nothing about its second half (rank correlation 0.02).
3. A small pull-back, which faded. Fading the start of a session was a little better than a coin from 2022 to 2024: +0.037% a trade above coin-flip sides, on average over all fade rules. That is a third of the 0.11% fee. From 2024 to 2026 it fell to +0.001%, next to nothing.
The 162 fade rules a coin, on average: a small pull-back in 2022-2024 (the market-order fee is 0.11% a trade), next to nothing in 2024-2026.
4. At the resting-order fee 6 of the 12 picks made money on the second half, but only Solana in Asia (+25.7%) beat 95% of the coin-flip traders. And a rule that must trade at the start of an hour can't count on a resting order filling.
What it means
- The paper's pattern isn't in these crypto sessions. The first hours did not point the way of the last, in either half, on any of the three coins.
- What was there was the opposite, and small: a slight pull-back in 2022 to 2024. It was a third of the fee, and it was gone by 2024 to 2026.
- Picking the best of 54 rules made the first half look good (+70.8% on Solana in Asia, +51.7% in London). The second half took most of it back. That is the "best of many tries" warning in action.
- The answer is no. Session momentum does not beat the coin-flip traders here.
Limits
- Hourly prices: the paper used half-hours. The first hour here is twice as long as theirs.
- Three large coins, four years, one exchange.
- Funding is left out: a trade of up to 4 hours pays it at most once.
- No slippage counted beyond the fee. Larger traders pay lower fees, but the side added next to nothing, so a lower fee only loses less.
Credits
- Lei Gao, Yufeng Han, Sophia Zhengzi Li and Guofu Zhou, "Market intraday momentum", Journal of Financial Economics, 2018: the first half-hour predicting the last, and the rule built on it.
- Steven L. Heston, Robert A. Korajczyk and Ronnie Sadka, "Intraday patterns in the cross-section of stock returns", Journal of Finance, 2010: patterns within the trading day.
- Our own: the crypto sessions, the follow-or-fade and size filters, the pick on the first half, and the two random twins (from our coin-flip traders page).
- Data: Bybit's public market data, 1-hour USDT perpetual prices; fees from Bybit's base fee schedule.
- PerpGremlin and Claude, Session momentum: do the first hours predict the last?, TerminalGremlin, 2026: the tests, the code and the words.
← Trend over monthsReal liquidation clusters →
Educational research from public market data; any trading discussed is paper, never real. Nothing here is advice. The full notice.