The Lab

Where the idea comes from

• Trend-following is old trading practice: buy what is rising, sell what is falling, on each market by itself.
• The paper. Tobias Moskowitz, Yao Hua Ooi and Lasse Pedersen, "Time series momentum" (Journal of Financial Economics, 2012). In 58 futures markets, each market's own last 12 months set its side for the next month: long if it had risen, short if it had fallen. Each was sized to the same swings (40% a year). The book made money in most markets and most decades. Brian Hurst, Ooi and Pedersen (2017) carried it back a century.
• In crypto, Yukun Liu and Aleh Tsyvinski, "Risks and Returns of Cryptocurrency" (Review of Financial Studies, 2021), found it at a short horizon. On Bitcoin from 2011 to May 2018, the weeks after its best fifth of weeks averaged 11.2%; the weeks after its worst fifth, 2.6%.
• How this differs from our other momentum pages. "Momentum across coins" ranks coins against each other. This page asks each coin about its own past only. Our grid page with a trend sleeve used trend rules only beside a grid.
• The bar to beat is our page "Coin-flip traders: can luck beat the fees?". This is part of our series "Published edges against the coin".

How we tested

• The coins: the 10 largest at the start of each year, as listed then (CoinMarketCap's first weekly snapshot, stablecoins left out), the same lists as our page "Momentum among blue chips only". And Bitcoin alone.
• Two windows, judged apart:
  • spot: Binance, July 2018 to December 2020, prices only, a 0.1% fee;
  • futures: Binance, January 2021 to September 2026, a 0.055% fee, and the funding paid or received. Both start after the papers' data ended.
• The rule, every Monday, held a week: for each coin, long if it is up over the last 1, 4, 12, 26 or 52 weeks, short if down. Each coin is sized to 40% a year of swings, as in the paper, and never above the money.
• The two rules we judge, fixed before the run:
  • the paper's: the 10 coins, the last 52 weeks;
  • Bitcoin, a week: Bitcoin alone, the last week. The other lookbacks are shown beside, not judged.
• Two random twins, 1,000 each, on the same coins, weeks and sizes:
  • random timing: each coin's longs and shorts dealt out to random weeks. In a rising market a trend rule is long more often, and this twin is long just as often. Only the timing is lost;
  • coin flips: each side a fair flip, as on the coin-flip page. We also show always long, for comparison only.
• The bar, fixed before the run: a rule holds if, in both windows, it makes money after costs and its Sharpe ratio beats 95% of both twins.

The maths, step by step

1. The side. The price was 100 a year ago and is 130 now: up, so long for the week. At 80 it would be short.

2. The size. A coin that swings 80% a year gets 40 ÷ 80 = half its share of the money. One that swings 20% would get double, but is held at 1x. With 10 coins, each share is a tenth.

3. The swings. Each day's move is squared, and recent days count more (a weighted average with a 60-day centre, as in the paper). The square root of that, times the square root of 365, is the yearly swing.

4. The Sharpe ratio. The average week divided by how much weeks vary, times the square root of 52. It rewards a steady gain over a wild one.

5. Why random timing matters. In 2021 to 2026 the 52-week rule was long 62% of the time while the coins rose. A random book that is long 62% of the time also gains. The rule earns credit only for beating that.

What we found

1. The paper's rule (10 coins, 52 weeks) did not hold.

• Spot, 2018 to 2020: it lost 17%. It beat only 34% of random timing.
• Futures, 2021 to 2026: it made +22%. It beat 49% of random timing, about what a coin does.
• Always long did better in both: +76% and +114%.

2. Bitcoin by its last week did not hold either.

• Spot: +52%, beating 78% of random timing.
• Futures: −25%, beating 61%.
• Always long made +197% and +69%.

3. The other lookbacks.

Gain or loss after costs, and in brackets the share of random-timing books it beat.

• 3 of these 20 beat 95% of random timing: 4 weeks on futures (both the 10 coins and Bitcoin), and 12 weeks on Bitcoin's spot. By luck about 1 would.
• 4 weeks stands out on futures. But it was not chosen before the run, and on the 10 coins' spot window it did no better than random (53%).
• Picked fairly, the best lookback on spot (1 week, for the 10 coins) beat 82% of random timing on futures, short of 95%.

Update, 21 October 2026: the 4-week lead, tested forward

Added after this page was published. The text above is left as it was.

The page called 4 weeks on futures "a lead, not a finding", needing its own test on prices not yet seen. That test is now set, with its rules and bar written down before the first price it judges:

• The rule: exactly as above, with a 4-week lookback, on futures:
  • the first 10 coins of the 2026 list of the largest (no later list is used);
  • Bitcoin alone.
• Judged: 52 weeks, from Monday 26 October 2026 to the week ending 25 October 2027.
• A first look after 13 weeks (to 25 January 2027), reported in February. It decides nothing.
• The bar, in November 2027, for each of the two:
  • a gain after costs;
  • and beating 97.5% of random-timing books and of coin-flip books.
  • That is stricter than the 95% above, because two rules are tested.
• How much a year can show: at the Sharpe found above (0.74 and 0.92), one year would pass that bar only about 1 time in 9 and 1 time in 7, even if the edge is real. So a "no" is the likely answer either way, and a pass would mean a lot.

• Tobias J. Moskowitz, Yao Hua Ooi and Lasse Heje Pedersen, "Time Series Momentum", Journal of Financial Economics, 2012: the 12-month rule, the sizing to 40% a year, the 60-day weighting.
• Brian Hurst, Yao Hua Ooi and Lasse Heje Pedersen, "A Century of Evidence on Trend-Following Investing", Journal of Portfolio Management, 2017.
• Yukun Liu and Aleh Tsyvinski, "Risks and Returns of Cryptocurrency", Review of Financial Studies, 2021: Bitcoin's time-series momentum at a week.
• William F. Sharpe, "Mutual Fund Performance", Journal of Business, 1966: the Sharpe ratio.
• Our own: the yearly top-10 lists (from our blue-chip momentum page), the random-timing and coin-flip twins, and the bar.
• Data: Binance's public spot and futures prices and funding; CoinMarketCap's public weekly snapshots for the lists.