Reading a backtest report

A backtest replays your strategy over historical candles. But a great backtest doesn't mean it will make money live. This page shows you how to spot overfitting.

Baseline metrics

• Total return / annualized: overall profit, but misleading on its own.
• Max drawdown: how far equity fell from a peak — can you stomach it?
• Sharpe ratio: excess return per unit of volatility; more objective than raw returns.
• Win rate: combine it with the profit/loss ratio.

Watch for overfitting (PBO)

PBO (Probability of Backtest Overfitting). The engine runs CSCV cross-validation on your strategy — it splits history into blocks and repeatedly selects parameters on one part and validates on another:

• PBO near 0: robust and reliable.
• PBO > 0.5: high overfitting risk — good backtest, but live performance will likely disappoint.

When a strategy with high PBO is about to go live, the app shows a confirmation dialog on purpose — don't skip it.

Look-ahead bias and walk-forward (WF)

• Don't trade directly with parameters optimized on the whole history — that peeks into the future.
• Walk-forward splits the range into rolling windows; each window picks parameters only on the front part and validates on the out-of-sample (OOS) back part.
• Check whether the OOS curve stably beats buy-and-hold. Only-IS strategies should be heavily discounted.

Monte Carlo

Resample historical returns/drawdowns to see whether results cluster or spread: clustering means robust, spreading means luck-heavy.

How to use it

• After a run, read the credibility grade (A/B/C/D) plus the two numbers: PBO and consistency.
• If PBO is high, don't go live yet — tune and re-validate the same way.
• A backtest is a "mine-detector", not a profit promise.

---

Next: Paper → Live