A backtest can be mathematically correct and still misleading. The key question is whether the test uses information and trading conditions that could reasonably have existed at the time. Three common problems are look-ahead bias, survivorship bias and overfitting.

Freeze what was knowable

Look-ahead bias occurs when a rule uses information before it became available. A company’s year-end financial period does not mean the final accounts were public on that date. Use the actual publication date when the strategy depends on a filing.

Similarly, if a rule needs a completed closing price, consider when an order based on that price could execute. Entering at the same close without a justified execution mechanism can overstate results.

Include the missing companies

Survivorship bias appears when a historical universe includes only companies that remain successful or listed today. Delisted, merged or failed securities can matter to the result.

Document how the investment universe was formed on each date. If point-in-time data are unavailable, clearly state the limitation rather than describing the test as a faithful historical reconstruction.

Limit repeated optimisation

Trying hundreds of parameters and reporting only the best version can fit noise. Suppose you test 200 indicator combinations on one historical period. The winning combination may owe much of its apparent strength to the selection process itself.

Separate development data from evaluation data. Avoid repeatedly changing the rule after viewing the evaluation period, because that period then becomes part of development. Keep an experiment log including unsuccessful variations.

Model practical friction

Include spreads, slippage, relevant charges and liquidity constraints. Adjust data appropriately for corporate actions and confirm how missing values are handled. A candle touching a limit price does not establish that the intended quantity could be filled.

Report drawdowns, trade count, exposure and sensitivity to costs as well as return. A short sample or a result dominated by one trade deserves particular caution.

Does a positive backtest prove future profitability?

No. Market conditions change and historical assumptions may be incomplete.

Is an out-of-sample test always independent?

Only if its results have not been repeatedly used to redesign the rule.

Build your testing foundations with FinDraft. Keep backtesting exercises distinct from any claim about features available in a specific platform.

Official learning references

Educational content only; not personalised investment advice or a recommendation. Investments and trading involve risk. All numerical examples are hypothetical. FinDraft provides independent preparation; official NISM examinations and certifications are administered by NISM.