The Lab

How we tested

• Every Binance futures coin, including those later removed (892), so the past isn't judged only by the survivors. January 2021 to September 2026, weekly.
• The coins each Monday: the 20 or 50 most traded over the last 30 days, listed for at least 60 days.
• The book:
  • rank them by their return over the last 1, 2, 4 or 8 weeks;
  • buy the top fifth and sell short the bottom fifth;
  • half the money each side, equal weights, held a week;
  • also the reverse (buy the losers), which some research (Jegadeesh, 1990; Lehmann, 1990) finds over short spans.
• Costs: the market-order fee (0.055%) on every change, and the funding that futures pay or charge.
• Fair judging:
  • the setting was picked on January 2021 to September 2025 only;
  • then judged once on October 2025 to September 2026;
  • compared with 1,000 books of coins picked at random from the same list.
• To pass:
  • the chosen setting had to beat the best of 16 random books (one for each setting we tried) in 95% of draws;
  • make money in the last year;
  • and beat 90% of the random books there.

The maths, step by step

1. The book. With 20 coins, the top fifth is 4 coins and the bottom fifth 4. Each gets half the money divided by 4: 12.5%.

2. A week. If the 4 bought rise 3% on average and the 4 sold fall 2%, the book makes 50% × 3% + 50% × 2% = +2.5%, whatever the market did.

3. Funding can be huge. A coin whose short-sellers pay 40% a week to its buyers pays a 12.5% position 12.5% × 40% = 5% of the whole book in a week.

4. So can the risk. The same 12.5% position in a coin that falls 97.5% in a week costs 12.5% × 97.5% = 12.2% of the book.

5. Luck among many tries. We tried 16 settings and kept the best. To see what luck alone gives, each random draw also builds 16 random books and keeps the best. The real choice must beat that best in 95% of draws.

What we found

1. By the test set in advance: no.

• The chosen setting (the top 20, ranked by 4 weeks) had a risk-adjusted score (Sharpe ratio) of +0.76 in the first years.
• It beat the best of 16 random books in only 67% of draws: with coins this wild, a random book often scores that well by luck.
• It did well in the last year (+2.8% a week on average, beating 96% of random books). But the first part failed, so the whole test fails.

2. Yet the direction held, which is rare.

• All 8 momentum settings made money in the first years, and all 8 in the last year, each beating 86% to 100% of random books there.
• Every reverse setting lost in both. It is the same book turned around, so this is one finding, not sixteen.
• The books barely moved with Bitcoin.

3. About half of the last year's profit was funding, from squeezed small coins.

• In 2026, small coins such as LAB, RIVER, RAVE and PIPPIN joined the 20 most traded while their short-sellers paid 25% to 55% a week to stay in. The momentum book was buying them, so it was paid.
• The chosen setting received 1.45% of the book a week in funding in the last year, against 0.30% in the first years. Price moves alone made about 1.4% a week.
• The same coins could move 90% or more in a week: LAB fell 97.5% in one week while held.

4. Fees were small: 0.04% to 0.09% a week. The worst fall of the momentum books ranged from 15% to 61%.

Update, 18 October 2026: after the papers

Part of our series "Published edges against the coin". The findings above are unchanged.

The question. Liu, Tsyvinski and Wu's crypto momentum used coins from the start of 2014 to the end of 2018. Their paper went public in May 2019 and was printed in 2022. Did their momentum last after that?

What they found. Each week they bought the top fifth of coins by their last 1, 2, 3 or 4 weeks and sold the bottom fifth short. That made 2.7%, 3.3%, 4.1% and 2.5% a week before costs, and 4.2% among the larger half of coins.

Why it might not last. McLean and Pontiff (2016) followed 97 published patterns in US stocks. They earned 26% less after the papers' data ended, and 58% less once the papers were published.

How we checked. Our momentum books already cover every week after the paper's data: Binance spot from August 2018 to 2020 (our page "Momentum across coins, tested again on 2018–2020"), and Binance futures from 2021 to September 2026 (this page). We read them again:

• before fees and funding, as the paper measured;
• year by year, and in two periods: 2019–2021, while the paper circulated, and 2022–2026, after it was printed;
• against 200 random books each time;
• the bar, set before reading: the setting this page chose (the top 20, ranked by 4 weeks) "lasted" if from 2022 it kept at least half the paper's 2.5% a week and beat 95% of random books.

What we found: it shrank.

• About 30% of the paper's momentum was left, in both periods. The bar needed half. So the answer is no: it didn't last at full size.
• Something was left. It still beat 96% of random books, and every one of the 8 settings (20 or 50 coins, 1 to 8 weeks) was above 0 in both periods. The 1- to 4-week ones kept 13% to 39% of the paper's figure.
• It came in bursts. 2022 and 2023 lost (−0.39% and −0.27% a week). 2025 alone made +2.78%, the only year to reach the paper's 2.5%.
• The loss, about 70%, is in line with McLean and Pontiff's 58% for stocks. That is what a published pattern usually looks like a few years on.

Limits. The paper ranked every coin above $1 million on CoinMarketCap; we ranked the 20 or 50 most traded on Binance, so the numbers are close relatives, not copies. Spot to 2020, futures after. 2026 is January to September.

Credits

• Narasimhan Jegadeesh and Sheridan Titman, "Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency", Journal of Finance, 1993: momentum across stocks.
• Robert A. Levy, "Relative Strength as a Criterion for Investment Selection", Journal of Finance, 1967.
• Yukun Liu and Aleh Tsyvinski, "Risks and Returns of Cryptocurrency", Review of Financial Studies, 2021; Yukun Liu, Aleh Tsyvinski and Xi Wu, "Common Risk Factors in Cryptocurrency", Journal of Finance, 2022: momentum in crypto.
• Narasimhan Jegadeesh, "Evidence of Predictable Behavior of Security Returns", Journal of Finance, 1990; Bruce N. Lehmann, "Fads, Martingales, and Market Efficiency", Quarterly Journal of Economics, 1990: short-term reversal.
• Halbert White, "A Reality Check for Data Snooping", Econometrica, 2000: the idea behind judging the pick against the best of many random books.
• William F. Sharpe, "Mutual Fund Performance", Journal of Business, 1966: the Sharpe ratio.
• R. David McLean and Jeffrey Pontiff, "Does Academic Research Destroy Stock Return Predictability?", Journal of Finance, 2016: published patterns earn less after publication.
• The universe, the random books and the pass bar are our own design.
• Data: Binance's public futures prices and funding, delisted coins included.
• PerpGremlin and Claude, Momentum across coins: buy the strongest, sell the weakest, TerminalGremlin, 2026: the tests, the code and the words.