8 strategies, run side by side — Crypto

Read the difference column before anything else. Every one of these strategies is being asked a single question — did it beat simply owning the same things over the same years? — and a return that looks handsome on its own can still be a way of losing to doing nothing. Worst fall is the other half of it: most of these exist to sit out the bad part, and giving up return to do that can be a fair trade.

Per unit of pain is return divided by the worst fall — how much you got paid for the worst moment you had to sit through. It is the column where most of these stop looking like failures: losing to buy-and-hold on return while halving the drawdown is not the same thing as losing. It is also the column that flatters a strategy which has barely traded, so read it next to the trade count, not instead of it.

Live calls only. Everything in the scoreboard above is backfilled — the rules replayed over history they had already seen — and turning that into dollars would be money invented out of hindsight. These start moving when a live trade closes, and each one resets independently in your own browser.

The overnight strategy cannot work here, and the page shows it failing rather than hiding it. It exists to capture the jump between a close and the next open. Crypto never closes, so there is no jump: across these coins the gap between one day’s close and the next day’s open is about 0.01%, which is not a gap, it is the same price a minute later. A strategy whose entire premise is a market being shut, run on a market that never shuts, is the cleanest way to show what these three are really made of.

The other two transfer more honestly. RSI(2) and momentum are claims about how prices behave, not about opening hours — so asking them about an asset class that did not exist when they were published is a real out-of-sample test, and a rarer thing than it sounds. RSI(2) kept almost exactly its published win rate here, about 69%, which is the closest thing to a genuine result anywhere on this site.

Momentum is the one to look at. On US funds it loses to buy-and-hold by a mile (+630% against +1,140%) and earns its place on drawdown alone. On crypto it returns +4,508% against a market that returned +5,654%, while holding its worst fall to 70% where simply owning the coins fell 96%. Giving up some of the climb to take about a quarter off the crash is the shape the strategy was always supposed to have, and it shows up far more clearly in an asset that crashes properly.

That drawdown figure used to read much better here, and it was wrong. This page once said momentum turned a 92% crash into roughly a quarter of that. It did not. The equity curve was only ever sampled at trade EXITS, and these trades ran for years at a time — so a fall that happened in the middle of a holding period was invisible to it, however deep. The strategy now books a trade every month, the curve is sampled every month, and the same positions over the same days report a 70% fall instead of 26%. Nothing about the strategy changed. The measurement stopped flattering it.

The trade count moved for the same reason and deserves the same suspicion. Crypto momentum shows 232 trades and a t-statistic of +3.36 where it used to show 22 and +1.75 — but that is the identical profit and loss cut into more rows, not new evidence arriving. A t-statistic is not comparable across a change in how often you sample, so do not read this one against the older number, and do not read either as proof of anything.

The volatility-scaled version is the one that answers the question properly. Sizing the position by how wild the market is — the piece the momentum paper specifies and my first implementation left out — does not make crypto more profitable. It makes it far less painful: the return drops from +4,508% to +2,396% while the worst fall goes from 70% to 41%, because a market this violent gets held well below full size. Per unit of pain it is the best thing on this site, and it got there by implementing the source more faithfully rather than by hunting for a number that fit. On US funds the same change does the opposite — it levers up, and finally beats buy-and-hold on return, at the cost of a deeper drawdown.

Two caveats worth holding on to. Crypto trades every day of the year, so a window here covers about 40% more sessions than the same window on shares. And these ten coins move together even harder than the twelve funds do — when Bitcoin falls, all of it falls, so ten symbols is nowhere near ten independent tests. The basket was widened from four coins and three funds on 6 August 2026, and the reason was frequency rather than evidence: RSI(2) fired about eight times a year per symbol, so on three funds the live record would have taken years to say anything. The alternative was to loosen the rule until it fired more often, which is picking the number that gives the answer you wanted. The rule is untouched; there are simply more places to look for the same signal. More opportunities, not proportionally more proof.

Every coin on these pages is now priced from Binance spot (the venue the research verdicts were measured on) instead of Yahoo’s multi-exchange composite, after Yahoo dropped or served stale Bitcoin daily bars three times in two weeks. Two visible consequences, neither of them silent: the backfilled record was rebuilt at venue prices on the switch date — backfill is a reconstruction and is resynced to whatever the rules produce on the current feed, so its trades and dates can shift by a few basis points — and history now starts at each coin’s Binance listing (August 2017 for Bitcoin) rather than Yahoo’s longer composite series, because a price from a venue where no order could rest is not a price this record should stand on. Live rows are evidence, not reconstruction: they keep the prices that were written down when the calls were made, on the feed that was traded at the time.

60 ideas tested once each against a fixed bar; 1 survived. Concluded 11 August 2026. Why the zero is credible, the closest miss, the nastiest data trap found, and what the one survivor actually proves — read the findings.