Backtesting applies a predefined trading strategy to historical data to estimate how its rules would have behaved under past market conditions. It is a research tool, not a guarantee of future profitability.
A useful backtest asks whether the strategy's logic produces a coherent distribution of outcomes after realistic costs and constraints. A weak backtest asks only whether an equity curve can be made to look attractive.
Freeze the strategy definition before judging the result
The rules being tested should be specified clearly enough that the researcher knows what counts as a valid setup, entry, exit and risk decision. If the rules change every time the historical result disappoints, the researcher is no longer testing one strategy.
This is why a stable trading plan or research specification should exist before the main performance judgment.
Historical prices are not the same as executable prices
Simulated trading does not reproduce every condition of live execution. Bid-ask spreads, commissions, slippage, gaps, order priority, liquidity and market impact can change realised results. The CFTC explicitly warns that hypothetical results can overestimate or underestimate performance because the trades were not actually executed under market conditions.
The more sensitive a strategy is to exact entry and exit prices, the more important realistic execution assumptions become.
Data quality can create false performance
A backtest can be wrong even when the code runs correctly. Missing observations, survivorship bias, incorrect corporate-action treatment, stale prices, look-ahead information or a data universe that was not actually available at the time can all contaminate the result.
The research process should document the data source, time period, instrument universe and any transformations that affect what the strategy was allowed to know.
Use more than one headline metric
Total return alone does not explain the path taken to earn it. A useful review may include drawdown, trade count, average win and loss, expectancy, turnover, exposure, holding time and sensitivity to costs.
The relevant metrics depend on the strategy. A high-turnover intraday system and a low-frequency position strategy should not be judged by identical execution assumptions.
Separate development data from genuinely unseen evidence
If the same historical observations are used to invent the idea, select parameters and report performance, the final result is partly an in-sample result. Holding back a genuinely unused segment can reduce that problem, although one out-of-sample test is not immune to repeated experimentation.
The next stage, forward testing, observes the frozen process on information that arrives after the strategy was designed.
Repeated testing creates selection bias
Trying many indicators, thresholds, assets or parameter combinations increases the chance that one configuration will look exceptional by luck. Research on backtest overfitting shows why high simulated performance can emerge after enough alternatives are searched.
The overfitting page owns that problem in depth. The practical implication here is to record how many alternatives were tried, preserve rejected results and be suspicious of fragile parameter choices.
Stress the assumptions, not just the strategy
A robust review asks what happens when costs rise, entries slip, a few large winners are removed, parameters move slightly or the market regime changes. If modest assumption changes destroy the result, the historical edge may be too fragile for confident deployment.
Backtesting is one part of validation
Strategy validation combines several forms of evidence: historical behaviour, unseen observations, execution realism, sensitivity analysis and logical consistency with the proposed edge. A backtest can reject a weak idea, but a strong historical result does not prove the idea will persist.
Common backtesting mistakes
- using future information in a historical decision;
- ignoring costs and execution;
- optimising many variants and reporting only the winner;
- using too few independent observations;
- treating a smooth equity curve as proof of a durable edge;
- changing the strategy after viewing out-of-sample results and still calling them out of sample.
Backtesting inside the MFXG framework
The Trading Systems pillar treats backtesting as disciplined falsification. The objective is to expose weaknesses, estimate behaviour under stated assumptions and decide whether the strategy deserves further testing—not to manufacture a perfect historical story.