There's a number that shows up in nearly every trading post, and it's almost always an illusion: "This strategy did +180%." Or +900%. Or a 94% win rate.
Numbers like that aren't lies. They're worse — they're true but misleading. They come from someone picking a lucky entry date, looking at too few trades, measuring against the wrong benchmark, and optimizing the strategy on the exact data they then test it on.
We build a backtesting tool. It would be easy to hand you numbers like that — they sell well. We deliberately don't. This post explains the four principles we hold ourselves to, even though they produce smaller, less spectacular results.
1. The average of all entries — not the best one
Most backtests report the outcome of a single entry date. And a single date can always be chosen so the result shines. Move the start by a week and the same strategy can turn negative.
Backtesting Arena doesn't report one entry. The tool runs the same strategy from every possible start date across the period, then averages the outcome. One lucky day can't flatter it; one unlucky day can't sink it.
The reason is simple: a backtest is a claim about repeatability, not about one perfect trade. If the return only exists on a single start date, it isn't an edge. It's chance dressed up as skill.
2. Fewer than 30 trades is an anecdote, not a proof
A 90% win rate sounds elite — until you notice the strategy made six trades. Six coin flips can land 90% heads too.
Under roughly 30 trades, a backtest isn't evidence. It's an anecdote. Win rate, CAGR, drawdown — all of them swing wildly when the sample is tiny. You're reading noise and mistaking it for signal.
Two runs can both show a 60% win rate: one with 8 trades, one with 120. Same headline — but the first's confidence interval is enormous, the second's is tight. Only one is trustworthy.
So a quality note sits right on the chart. A low trade count gets flagged, not hidden. You see "small sample" before you fall in love with a pretty equity curve. And it cuts both ways: a gorgeous-looking strategy with nine trades gets the same caution as a bad one. We'd rather under-claim than sell you a number that can't hold up.
3. The honest benchmark: Average Buy & Hold
Most backtests compare against "buy on day one and hold." That benchmark is rigged by its start date. Pick a date before a moonrun and buy-and-hold looks unbeatable. Pick one before a crash and anything beats it.
We benchmark against Average Buy & Hold instead — the same logic as our strategy averaging, applied to plain holding: hold, but starting from every possible entry, then average. No single lucky start can flatter passive holding.
That changes the picture. On a coin that mooned once then bled for years, day-one buy-and-hold looks amazing. The averaged-across-all-entries figure is often deeply negative — and that's the number real holders actually faced.
The uncomfortable part: against Average Buy & Hold, some strategies that "beat the market" stop beating it. Good. That's exactly what the benchmark is for.
4. Out-of-sample as the publishing gate
Any strategy can be tuned to look perfect on past data. Enough parameters, enough attempts, and you'll fit the noise exactly. That's not an edge. That's curve-fitting.
The trap: optimize RSI period, SMA period, every filter — all on the same history — until the equity curve is beautiful. It will be beautiful. And it will fall apart the moment real, unseen data arrives.
Our rule before any performance claim goes out: out-of-sample. Tune on one slice of history. Test on a slice the tuning never saw. If the edge survives, it might be real. If it evaporates, it was fitting.
A concrete example from our own work: we once built a Bitcoin cycle indicator we had publicly argued for. Honest testing showed it badly underperformed buy-and-hold. So we shipped it as a context tool, not a trading signal — with the caveat printed directly above the chart. That's exactly why you won't get a "94% win rate, +900% CAGR" headline from us. Those numbers exist. They're almost always in-sample fantasy.
What this means for you
These four principles make our results smaller and less spectacular than they could be. That's on purpose. A smaller, real edge you can stand on is worth more than a big number that collapses on the first real trade.
Test it yourself. Crypto backtests are free, no signup wall — and every result is the average of every entry, with the trade count visible and Average Buy & Hold sitting right next to it.
Study the Past — Improve your Future. 🥋