▲ 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 · strategies tested · write-up
#39 Momentum across coins, tested again on 2018–2020 Not confirmed; the lead holds
Past prices; no real orders were placed. Every figure is a share of the money, or a count.
Last checked: 16 October 2026.
Where the idea comes from
- Momentum across assets was documented by Narasimhan Jegadeesh and Sheridan Titman (1993). Its history, maths and catch are on our theory page "Momentum: do recent winners keep winning?".
- Testing a fixed setting again on data it never saw is the standard guard against luck among many tries. This page does that for the setting picked in our study "Momentum across coins: buy the strongest, sell the weakest".
The question
Our momentum study found that buying the coins that rose most over recent weeks, and selling short those that fell most, made money in every period on every setting. But its chosen setting failed the test we had set in advance. So we tested that one setting again, unchanged, on older prices none of it had seen: Binance's spot market, 2018 to 2020. Every rule was written down before those prices were read.
How we tested
- Every Binance spot coin priced in USDT, including those later removed (199), from 30 July 2018 (the first week with enough coins) to the end of 2020: 127 weeks.
- The setting, fixed:
- the 20 most traded coins each Monday;
- ranked by their last 4 weeks;
- buy the top fifth, sell short the bottom fifth, held a week;
- no coin above 12.5% of the money.
- The other 7 settings (20 or 50 coins, 1 to 8 weeks) were run beside it, but not chosen from.
- Costs: the spot market-order fee, 0.1%, on every change.
- What this measures: only the price part. Spot pays no funding, and in those years most of these coins couldn't be sold short at all.
- To pass: a profit after fees, beating 95% of 1,000 random books, and making money in at least 2 of the 3 years.
The maths, step by step
1. A week of fees. If 2 of the 8 coins are swapped, the book trades 2 coins out and 2 in, at 12.5% each: 50% of the money × 0.1% = 0.05%.
2. The result. The book averaged +0.33% a week after fees. Steady weeks of +0.33% would compound to +52% over 127 weeks. The real weeks swung, and a fall costs more than the same rise gives back (−10% then +10% leaves 99%), so they came to +33%.
3. Against luck. Random books of the same size averaged a risk-adjusted score (Sharpe ratio) of −0.33, because they pay fees for nothing. One in 20 of them reached +0.81. The real book scored +0.51: better than 89% of them, short of the 95% bar.
What we found
1. The fixed setting made money, but not enough to rule out luck.
- +0.33% a week, 2 of 3 years up: 2018 −16% (five months), 2019 +32%, 2020 +25%.
- It beat 89% of random books. The bar was 95%, so it is not confirmed.
2. The direction held again.
- All 8 momentum settings made money on average, from +0.05% to +0.73% a week.
- The 50-coin ones were strongest: +0.45% to +0.73% a week, beating 95% to 99.7% of random books.
- They don't count toward the bar, because they weren't the setting fixed in advance. Until late 2019 there were only 12 to 20 coins to choose from, so the 20-coin and 50-coin books often held the same coins.
3. Fees mattered little: at the futures fee (0.055%) the fixed setting made +0.37% a week.
What it means
- From 2018 to 2026, on spot and then on futures, momentum made money in every period on every setting. Nothing else we tested this round did.
- But the one setting fixed in advance fell just short on fresh data. It stays a lead, not a proven edge.
- The honest next step would be a live paper book, scored against random books from day one.
Limits
- 127 weeks, and few coins until late 2019.
- Prices only: no funding, and no cost or availability of short selling.
Update, 16 October 2026: two corrections
Added after this page was published. The text above is left as it was.
- The yearly figures were added up, not compounded. The weeks of each year were summed. Compounded, as the +33% total is, the years were: 2018 −16% (five months), 2019 +30%, 2020 +21%. Together they make the +33%. Still 2 of 3 years up, so the result is unchanged.
- "Nothing else we tested did" no longer holds for momentum itself. Our later page "Momentum among blue chips only: Bitcoin and the top 10" ran the same idea on the biggest coins only, and on futures from 2021 it lost clearly.
Update, 18 October 2026: a chart
The same chart as on our page "Momentum across coins: buy the strongest, sell the weakest": the fixed setting (the top 20, ranked by 4 weeks) at the 0.1% fee, grown week by week, against the middle 95% of 200 random books with the same universe, sizes, cap and costs. These random books are drawn again for the chart; the test's own 1,000 were not kept.
The book ended a third up, but inside the random books' band, having dipped just below it in late 2018, so luck could not be ruled out.
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.
- William F. Sharpe, "Mutual Fund Performance", Journal of Business, 1966: the Sharpe ratio.
- The setting comes unchanged from our study "Momentum across coins: buy the strongest, sell the weakest". The cap and the re-test design are our own.
- Data: Binance's public spot daily prices, delisted coins included.
- PerpGremlin and Claude, Momentum across coins, tested again on 2018–2020, TerminalGremlin, 2026: the tests, the code and the words.
← Momentum across coinsMomentum among blue chips only →
Educational research from public market data; any trading discussed is paper, never real. Nothing here is advice. The full notice.