EMA Cross: 52 variants, every one tested.

A backtest is only worth reading if the test itself is fair. Here is every choice we make before a single EMA Cross number is computed. So a good result cannot be an artifact of how we set the test up.

We match each signal to a timeframe that suits it

A fast crossover is a scalping signal, so it earns its pages on short intervals; a slow pair is a macro signal, tested on 4-hour and daily candles. We never run a scalping pair on a monthly chart. A mismatch would fail for the wrong reason and teach you nothing about the signal itself.

We test the parameters traders actually use

The canonical fast/slow pairs are the ones cited in real technical analysis, which is why each earns a page. We do not test arbitrary combinations tuned to look good in hindsight. Curve-fit soup like "37/142" is excluded on principle: it answers no question a real person asks, and it is exactly the overfitting this site exists to call out.

We run a clean, parameter-free baseline

The strategy exits only on the reverse signal: no take-profit, no stop-loss. That is deliberate. A take-profit caps the few big winners a trend-follower lives on while leaving losers uncapped, which flatters the raw signal. The clean baseline tells you what the signal is worth before you add knobs. A stop is your variant to run, not our baseline.

The short side is a falsification control

We test both directions on the same candles. The short side is the control. If the long side works and the short side fails across the board, that gap is the fingerprint of a rising market (most of 2020 to 2026 drifted up), not proof of a removed bias. A real edge should not depend entirely on which way the market happened to move.

One uniform leverage ladder, read at 2×

Every signal runs the same rungs (1×, 2×, 10×, 50×, 100×) on both sides. Higher rungs do not find an edge the lower ones missed; they find the liquidation cliff faster. 1× is the only rung that never force-closes a trade, so 2× is the fair-test compromise: leveraged enough to matter, low enough that the verdict is not an artifact of losses truncated by liquidation. We judge each strategy once, at one rung. The five rungs share one set of trades, so read them as one experiment at five zoom levels, not five independent proofs.

A fair test is not a perfect one. Every result here is in-sample, measured on the same history the strategies were selected from, so read it as a ceiling, not a forecast. And the engine does not yet model funding or slippage, so a fair comparison between strategies is still not a faithful copy of live trading.