How to Backtest a Trading Strategy

Backtesting is applying a strategy's exact rules to historical data to estimate how it would have performed. Done honestly, it answers cheaply what live trading answers expensively: does this rule set have positive expectancy, and what does its pain look like? Done dishonestly — and most casual backtests are — it manufactures confidence in a curve that never existed.

The procedure

1. Freeze the rules first. Entry trigger, invalidation, management, written before touching the data. Rules edited while looking at outcomes are being fitted, not tested.
2. Walk the data forward. Bar by bar or replayed tick by tick, deciding only on information available at that moment. Scrolling a finished chart and marking the trades you "would" have taken is the look-ahead bias machine.
3. Record every signal — including the losers you would rather skip — with entry, stop, exit and R. The skipped-loser habit is how a 0.1R system becomes a backtested 0.8R fantasy.
4. Charge costs. Spread on every entry, slippage on stops, swap on holds. On short-timeframe systems costs routinely decide the sign of the result.
5. Read the sample honestly: expectancy, win rate, average win/loss, profit factor, maximum losing streak, maximum drawdown — and how many trades stand behind them.

Sample size and regimes

Thirty trades is an anecdote. A few hundred trades, spanning trending and ranging periods and at least one genuinely nasty stretch, begins to be evidence. A strategy tested only on the regime it likes will meet the other one live, at full size.

Overfitting: the quiet killer

Every parameter you tune to the data (this moving average, that session filter, this stop multiple) makes the backtest better and the future worse. Defences: fewer parameters, round numbers instead of optimised ones, testing on data the tuning never saw, and suspicion of any result that looks too smooth. A robust edge survives being slightly wrong about its own settings.

From backtest to live

The backtest's most practical output is the cost of doing business: the losing streak and drawdown you must sit through for the expectancy to show up. Write those numbers down. Live, at small size, your journal becomes the forward test — and the first thing it usually reveals is the gap between the tested system and the executed one.