A backtest is a model of the past
Backtesting asks how a defined strategy would have behaved on historical data. It can expose obvious weaknesses, compare rule variations, and show how results changed across different market conditions. What it cannot do is guarantee that the future will behave the same way.
The more parameters you tune after seeing the data, the easier it becomes to fit noise rather than a durable pattern.
Define rules before looking at results
A test should be reproducible. Write the entry, exit, stop, sizing, market universe, timeframe, and fee assumptions before you run it. If the strategy depends on phrases like “enter when it looks strong,” you do not yet have a rule set that can be tested cleanly.
Clear rules also make it easier to compare the backtest with live or paper execution later.
Include the frictions real traders pay
Fees, spread, and slippage can turn a small theoretical edge into a loss. This matters most for high-turnover strategies. Use assumptions that are conservative enough to reflect the markets you actually intend to trade.
If the backtest only works with zero fees and perfect fills, the result is not robust.
Use out-of-sample data
A common approach is to develop the strategy on one period and evaluate it on a later period that was not used for tuning. This does not eliminate overfitting, but it creates a harder test.
Also compare performance across bull, bear, and sideways conditions. A strategy that succeeds only during one favorable regime should be understood as regime-dependent, not universally reliable.
Look beyond total return
Total return can hide uncomfortable details. Review drawdown, losing streaks, trade count, average gain and loss, dependence on a few outliers, and the stability of results when small parameters change.
A useful backtest should help you understand failure modes. If it only gives you a large green number, it is not telling you enough.