Why backtests aren't guarantees

A simulation that walks the strategy forward through every 30-minute candle since 2020, using only data the strategy would have had at that moment (no look-ahead). Fees and rough funding are baked in. The score is year-weighted to avoid one anomalous year dominating.

This is more rigorous than the usual 'one-shot in-sample fit' you see in trading content. Walk-forward is the industry-standard way to estimate out-of-sample performance.

A promise the strategy will deliver the same numbers live. Markets evolve: new participants, new market structure, new macro regimes. The 2020-2024 conditions are not a complete sample of what 2026-2030 will look like.

A guarantee against drawdowns. Backtests show periods of -30% to -40% drawdown that lasted months. Live, you may hit one of these from day one — there's no rule that says good performance has to come first.

Check the worst year, not the best. Can you live with that scenario emotionally and financially? If the worst backtest year is -25% and you'd panic-sell after -15%, this strategy isn't for you regardless of average returns.

Check the OOS portion specifically. In-sample numbers are unreliable (the strategy was tuned on them). OOS is the part that's never been seen by the optimisation.

Check the trade count. A backtest with 1000+ trades has more statistical weight than one with 50.

Start small. Allocate an amount you'd be okay losing 30% of. If after 3-6 months the live curve looks sane (similar shape to the backtest's hardest periods), consider scaling up. If it diverges dramatically, stop and investigate.

Don't extrapolate backtest CAGR onto your capital. The strategy might do +50% annual in some periods and -10% in others. Plan for the rough years and the smooth years equally.