Selling bitcoin's three-day high when it falls back under

The idea: price poked through the high of the last three days and came back, so we sell. We tested four versions of it, and not one of them beat a random entry in the same window.

Why it sounded right

This is one of the oldest ideas in trading, and its logic is strong. A big coin spends most of its time chopping, not trending. In that kind of market, a push above a local high is usually not the start of a move but a sweep: somebody took out the protective buys above the level and sold into them. The sign of it is immediate — price went past the level and could not hold, closing back under it. So whoever was pushing up is already out.

The trade comes out very neat. The entry location is known in advance — the boundary of the last three days. The entry moment is unambiguous — the return back under it. The risk is unambiguous — above the poke. And the horizon is short: if the poke was false, price comes back into the range in hours, not weeks. All of which makes this an ideal candidate for mechanizing: it needs no view on the market, no news, no choosing a coin.

One more thing supports it, and it is genuinely true: price oscillates around range boundaries constantly, pokes that come back happen often, and the share of cases where price then goes back deep into the range is high. From which the natural conclusion: if the return happens often, selling it should pay.

How we tested it

We tested not one version but four, so that nobody could say "the idea is right, the detail was unlucky." Version one: strictly a poke that closes back under the boundary. Version two: the same thing in both directions, buying pokes of the low as well. Version three: define the boundary not just as the extreme but by the number of touches — a level price has banged against several times counts as real. Version four: work off the range as a whole rather than off a line — sell at the upper boundary, buy at the lower.

One bar for all of them: a random entry in the same window. Same exit rules, same average holding time, but entering at random moments instead of at signals. If the move catches anything, it has to beat that bar. The window was about two years, long enough to include both a rise and a chop.

How it ended

Zero. Not one of the four versions beat the random entry. The observation itself is correct: pokes that come back really are frequent, and price really does often go back into the range afterwards. But it goes back into the range not because there was a poke — it does that without a poke too, simply because the market chops. A random entry harvests exactly the same chop. Selecting by the poke adds nothing to it, and it does add costs: the spread and the fee on every entry.

Defining the level by the number of touches did not save it either. Intuitively, a level price has banged against many times should be stronger than a random boundary, but it made no difference to the result. The idea is closed both to the upside and the downside — so it is not a case of us selling into a rising market.

Where this could be wrong. We measured one big coin and one window of about two years, which is not enough to speak about different market regimes separately. The protective order was left at one fixed size rather than swept widely, and you cannot tune it on this same data: that way you find a lucky setting that means nothing.

What this teaches

• The bar "a random entry in the same window" is mandatory for any move that bets on a return to the middle. A chopping market produces a positive result by itself, and that is easy to mistake for the rule working.
• How often an event happens is not proof. "Price returns to the range most of the time" is true with no poke involved.
• Test a family of versions, not one. If all four versions of one idea come out at zero, the problem is the idea, not its implementation.
• "A strong level" is a nice phrase but so far an untested one. The number of touches on the boundary did not improve the result.