Backtesting definition: what it is
Backtesting is the process of evaluating a trading approach using historical market data, by applying the same decision rules you would use in live trading. A “backtesting definition” matters because it sets the exact meaning of key parts of the process, such as:
- What data is used (price series, time stamps, and how missing data is handled).
- What is simulated (entries, exits, position sizing, and order timing).
- What assumptions you make (execution model, slippage, fees, and whether you use bid/ask or mid prices).
In forex, these choices are not small details. Forex trading involves bid/ask spreads, continuous pricing, and cross-currency pricing conventions, so the way you define backtesting can change whether your results reflect a plausible process or an optimistic reconstruction.
How the definition affects decisions you can trust
A precise backtesting definition determines what kinds of conclusions you can reasonably defend.
First, it affects whether your test is time-consistent. If your definition implies that a signal computed at time t can still use information that was only known later, the test can become unrealistically accurate. This can happen through data leakage (using future candles, revised rates, or misaligned timestamps).
Second, it affects whether the test is execution-consistent. A definition that assumes perfect fills or ignores spread and transaction costs can overstate performance. In forex, ignoring costs or using the wrong price basis (for example, treating mid as if it were executable bid or ask) changes expected returns and risk metrics.
Third, it affects what you mean by “outcomes.” If your definition does not specify the assumptions behind every calculation (including compounding method, rounding, and whether you rebalance), two people using the same general idea may produce incompatible results.
A realistic example of how assumptions change results
Imagine two backtests of the same rule set, both using the same historical currency prices.
- Backtest A defines execution as “fill at the decision candle’s closing price, with zero fees, zero slippage.”
- Backtest B defines execution as “enter at the next available time step, using bid/ask prices, and subtract a fixed spread and fee model.”
Even if the rule logic is identical, the results can differ because the definition changes the effective entry/exit prices and the cost burden. That means the backtesting definition is a boundary for interpretation: you are not just testing a strategy idea; you are testing a specific, explicitly simulated process.
Material limitations and common failure modes
Historical backtesting has limits that follow from its definition and from market behavior.
One major limitation is that historical relationships do not guarantee future outcomes. Markets adapt: liquidity can change, volatility regimes shift, and microstructure effects may differ from the historical period.
A second limitation is overfitting. If your definition implicitly allows repeated tweaking until the test looks good, the test may be measuring adaptation to the past rather than a stable behavior. This is not fixed by “more data” alone; it depends on how you separate calibration and evaluation.
A third limitation is model mismatch. Execution, fills, and costs are often simplified. If your definition uses an unrealistically smooth execution model, the backtest can fail when real orders face uneven liquidity or different fill timing.
How to verify meaningfully, without assuming certainty
To verify independently, focus on the parts that a reviewer can reproduce from your description.
Start by checking that the definition states:
- Exact timing: when decisions are computed and when orders are assumed to fill.
- Price basis: which prices are used and how bid/ask and spreads are handled.
- Costs and frictions: fees, spreads, and slippage assumptions.
- Data integrity: whether there is any time misalignment or leakage.
Then check robustness: compare performance across separate time periods and different volatility conditions. If results only appear under narrow historical windows, that is evidence that the definition may be capturing noise rather than durable behavior.
Next question to ask
If you want to use backtesting results as education rather than prediction, ask: “Which assumptions in the backtesting definition would most likely fail in live forex execution?”