We Tested the EMA Crossover on Three Years of Bitcoin Data. It Lost to Random Entries in 40 of 40 Settings

The short version: we measured the classic exponential moving average (EMA) crossover on Bitcoin across 40 settings. In all 40, its hit rate was below that of random entries. Flipping the signal did not rescue it either.

What we tested

The idea is simple and widely taught: when a fast EMA crosses above a slow one, buy; when it crosses below, sell. We tested the 8-over-10 and 8-over-20 pairs, in both directions, on two data sets:

• 129,600 one-minute bars of Bitcoin futures, and
• 114,336 fifteen-minute bars of Bitcoin spot — a little over three years.

Every cross was compared with random entries taken on the same bars, and with realistic trading costs. A signal only counts if it beats chance and its edge is larger than what the trade costs.

What we found

• 40 of 40 settings lost to random entries on hit rate.
• On one-minute charts, 98.9% of crossovers were "false" — the lines crossed back within 60 bars.
• The best edge we found was about one-ninth of the trading cost. In plain terms: even the best setting paid the exchange more than it earned.
• Reversing the signal does not work either. The confidence interval of the reversed version includes zero.
• The size of the gap between the two lines tells you nothing about what happens next (rank correlation +0.0001).

The one thing that survived — and why you still can't trade it

The moment of the cross carried no information, but the position of the lines did, slightly — and in the opposite direction to what most traders expect. On 15-minute bars, when the fast EMA was below the slow one, Bitcoin was higher four hours later 54.0% of the time, versus 48.2% when it was above.

That 5.8-point gap is statistically solid on 15-minute data (week-blocked 95% interval 4.4–7.2 points). But it disappears on 4-hour bars, and the average return difference is about 0.004% — roughly one twenty-fifth of the cost of a trade. It is a fingerprint of short-term mean reversion, not a strategy.

Why we publish this

Most indicator content shows the trades that worked. We publish the whole distribution, including the ideas that failed, so you can see what a signal is worth after chance and costs. This result is part of our Evidence Graveyard: every idea we tested, with the rule written down before the test.

Method in one line: written rule before testing · random-entry baseline on the same bars · costs deducted · results kept whether they help us or not.