Skip to forex research
Simple Forex StrategiesRules / Quote / Account

Forex Backtest Errors: Spread, News and Look-Ahead Bias

Use a price convention and causal timeline that match the decision you claim to test.

Original source review: 8 October 2026. Numerical scenarios are fictional; no broker account or paid signal feed was tested. Expanded edition: 14 October 2026.

Choose the Price Convention

A mid-price chart lies between bid and ask; it is not itself an executable quote. For a simple long, entry normally consumes the offer and exit sells at the bid. A research series may not contain those quotes. State whether the experiment uses actual bid and ask observations or a mid-price proxy with an added friction model.

An apparently tiny edge can vanish when the convention changes. In fictional arithmetic, a 4-pip gross benefit minus 1.5 pips of round-trip friction leaves 2.5 pips. Increasing that assumption to 3 pips leaves only 1 pip. These are sensitivity inputs, not measured spreads. Applying the cost twice is also wrong, so document exactly what the fill series already includes.

Do Not Trade a Bar Before It Exists

If a condition needs a completed hourly close, the decision occurs after that close becomes available. The entry cannot then be assigned to an earlier low in the same hour. Save observation time, calculation time, earliest order time and the research fill reference. A close-based signal and a hypothetical close-price fill need an explicit latency assumption.

A revised economic value is another source of future information. If a model uses announcement data, preserve the original release vintage where available. If only revised history exists, label the experiment accordingly. A well-formatted CSV does not establish that every field was observable at the time.

Keep Stop Sequencing Unresolved When Necessary

A bar that reaches both a stop and an objective cannot ordinarily establish their order. Use finer data, a predeclared conservative convention or an unresolved state. Do not select the favourable path individually after seeing the high and low. A level touch also does not prove a limit fill for the intended quantity.

Execution rules must cover gaps, missing quotes, partial fills and cancellations. If the experiment only measures movement between reference levels, call it a reference study. Do not turn that result into a broker return simply by adding an account-size column. Margin and liquidation would introduce additional assumptions.

Record the Selection Process

A trend, range and breakout rule can all be plausible hypotheses. Trying many thresholds on the same period and reporting only the best one is not independent validation. Save the variants, chronological split and selection objective. Let the untouched period evaluate the frozen choice instead of deciding which variation should have been chosen.

Cost sensitivity and neighbouring parameters help reveal fragility but do not become new independent samples. If the rule fails once costs are included, retain the failure. Do not lower the cost assumption until the preferred strategy passes without new execution evidence supporting that change.

Publish Reproducibility, Not Just a Percentage

A useful evidence package includes product identity, feed convention, session clock, entry and exit rules, costs, parameter history and a complete outcome ledger. State cash-flow and capital assumptions before presenting an account-level return. Include excluded and unresolved observations so the denominator can be reconstructed.

No empirical forex backtest or provider ledger was independently reproduced for this publication. The checklist is educational and the worked figures are fictional. Its purpose is to help a reader distinguish a testable claim from a chart that requires guessing the most important inputs.

Primary Sources

Forex Base-Unit Sizing Worksheet

Continue the Research

A flat midpoint need not produce zero side-based change

With invented entry and exit quotes both 1.0998 bid and 1.1000 ask, a long 10,000-unit comparison uses 1.1000 entry and 1.0998 exit. Price-only change is negative 2 quote units despite an unchanged midpoint. Adding 2 commission and a 1 financing debit gives negative 5. This is conditional arithmetic, not a measured spread or actual fill. The entered side gap is not deducted again. A final economic release value is another separate source and cannot settle what a model knew before a decision.

The next record to acquire

Preserve the original product, source version, availability, eligible population and exact cost/quantity basis for the chosen question. An official mechanism reference does not authenticate a paid account, complete channel archive or current instrument entitlement. Failed variants and unresolved observations remain part of the study. The linked workbook contains authored examples, not extra independent evaluation data. A genuine instrument task must still complete its authorised factory and independently verified master; this educational addition neither promotes a model nor changes native exit ownership or the established research-fill convention.