Analysis & metrics

Edge

By Jarosław Wasiński • Published: Apr 21, 2026 • Updated: Oct 11, 2026

A repeatable advantage expected to produce a positive result after relevant costs under specified conditions. Historical profit is evidence to examine, not proof of an edge. Evaluation needs uncertainty estimates, realistic execution and genuinely unseen data; no fixed number of trades establishes an edge for every strategy.

Example. A hypothetical rule has a 45% winning probability, a +2R gain and a −1R loss before costs. Its model expectation is 0.45 × 2 − 0.55 × 1 = +0.35R. A 0.15R cost on every trade reduces expected value to +0.20R. These are stipulated payoffs and probabilities, not proof that an actual strategy achieves them.

A positive-outcome hypothesis

An edge concerns specified rules, instrument, trading hours, execution and position scale. It is not a property of an indicator’s name or an attractive chart. In the example, cost affects every outcome, reducing expectation by 0.15R rather than only the winning component. If charges are already reflected in fills, do not subtract them again. Winning probability and payoffs must be estimated together.

From idea to evidence

Record complete entry, exit and invalidation criteria before testing. Include unfilled signals, spread, commission, slippage and financing. Separate observations used to choose the rules from final evaluation. Searching many variants and keeping the best increases the chance of accidental success; a positive historical mean alone is insufficient. Report rejected variants in the research record, rather than only the winner.

Limitations and common mistake

The relationship can weaken when liquidity, costs or market participants change. Positive expectation still permits substantial drawdowns. A larger position can worsen execution, so a small account’s performance does not scale automatically. Do not treat confidence, analytical knowledge or discipline as proof of an edge. The process should support an honest assessment, rather than guarantee a profit.

Separate a hypothesis from accessible execution

A price relationship may be inaccessible at your volume, latency or fee schedule. Record unfilled signals and prices available after the decision. An attractive quote seen before calculations finish is not automatically an achievable fill. Net-performance analysis should represent the actual process, rather than a theoretical choice of the best instants. Document each change in filters or the execution path so results remain interpretable.

Apply it to decisions

Compare the estimated effect with uncertainty, costs and sensitivity to assumptions. Define criteria for further testing and review, retaining strategy versions. Distinguish a promising idea, a positive estimate and a sufficiently supported model for the chosen scale. In the course: edge and expected value and the systems graveyard.

FAQ

Is there a fixed trade count that establishes an edge?

No. Required information depends on effect size, dispersion, dependence and the selection procedure.

Does an edge guarantee a profit in every series?

No. Outcome distribution, sizing and extreme losses matter even with positive expectation.

← Back to the full glossary