THE PROOF / WHY A BACKTEST LIESREAL BITCOIN · 2019–2026
Two backtests. One is curve-fit. Can you tell which?
Both of these are real strategies on Bitcoin, same fees, same rules. One is a boring setting nobody tuned. The other was chosen by trying thousands of settings and keeping the one that scored best on this exact stretch of history. Up to the line, they both look like winners.
Strategy B was the curve-fit one. Here's the same two strategies, split into the years B was tuned on, and the years neither of them had ever seen.
In-sample · 2019–Jun 2023 · the years B was tuned on
Strategy B (curve-fit)+581%
Strategy A (untuned)+326%
On the history it was tuned on, B looks nearly twice as good. This is the part of the chart that sells you the strategy.
Out-of-sample · Jun 2023 on · data neither had seen
Strategy A (untuned)+190%
Strategy B (curve-fit)+105%
$1,000 from the split: the boring A grew to $2,898, the tuned B to just $2,049. The boring one won on the data that actually counts.
Strategy B was never a better strategy. It was the same kind of strategy, tuned until it fit the past. The moment it met data it had never seen, that huge in-sample lead shrank to less than the boring setting nobody optimised. A single beautiful equity curve tells you almost nothing, because you can always find one that fits a chart you already know the ending to.
Honest detail — both are EMA crossovers on daily Bitcoin, long/flat, 0.06% fee each side, next-close fills (no look-ahead). "Curve-fit" = we scanned roughly 4,300 fast/slow settings and kept the single one with the best return over 2019–Jun 2023, then ran both untouched on the years after. Neither collapsed to zero, because trend-following has some genuine signal on Bitcoin. That's the point: the tuned one's dominance was mostly an illusion of the period it was fit to, and only the out-of-sample half revealed it.
This is the whole reason the site exists
Every verdict here is decided on the out-of-sample half a strategy never got to see, not the pretty part. See exactly how we grade →
Soon you'll be able to run your own strategy through the same split and find out if its backtest is telling you the truth. Meanwhile, walk the graveyard →