Who this is for — Readers who want to distinguish favourable results from a measurable advantage without treating a backtest or a short winning run as a permanent property of a system.
In trading, an edge is the hypothesis that a specified set of rules produces positive net expectancy in the population of trades to which it is applied. It is something to estimate, not a label attached to a setup: it depends on signal selection, sizing, exits, costs, data quality, instruments and market regime.
In plain terms — Edge does not predict the next trade or guarantee that an observed advantage will persist. It means that, under stated conditions, estimated net average outcome is above zero; the estimate always retains uncertainty.
Prerequisites — Read Expectancy, sample size, transaction costs and backtest first.
Where it may come from
| Component | Observable effect | Required control |
|---|---|---|
| Selection | Changes the distribution of opportunities considered | Rules written before evaluation |
| Management | Changes the frequency and size of gains and losses | The same rules in simulation and application |
| Execution | Spread, commissions, slippage and impact reduce gross results | Net results with documented assumptions |
| Context | The same process may behave differently across regimes | Segmented results and ongoing stability checks |
Two operators may start with the same signal and obtain different distributions because they apply different sizing, exits or costs. This does not prove that either has an edge: it creates two distinct processes that require separate evaluation.
How to test it without a magic threshold
No universal trade count certifies an edge. Estimation precision depends on variability, dependence between observations, event frequency, regime changes and the number of alternatives tested. A credible assessment:
- defines rules, universe, horizon and metric before testing;
- uses coherent data and excludes information unavailable at the simulated time;
- separates development from out-of-sample or walk-forward evaluation;
- accounts for how many variants were tried, because multiple testing raises false positives;
- includes realistic costs and execution constraints;
- compares different periods and regimes and keeps monitoring new data.
Limit — Backtested results are hypothetical. They may overestimate or underestimate live outcomes and cannot fully reproduce liquidity, execution, the pressure of losses or future conditions. Even statistically favourable evidence can decay.
Reading card
- Better question: which net expectancy is compatible with these data, and with what uncertainty?
- Document: rules, costs, samples, tested variants and application conditions.
- Typical mistake: calling the best result found after many trials on the same history an edge.
Sources
- Campbell R. Harvey and Yan Liu, Backtesting, multiple-testing control, sample size and strategy significance.
- CFTC, Commodity Trading Systems Sold on the Internet, limitations of hypothetical results and differences from live conditions.
- SEC Office of Investor Education and Advocacy, Investor Bulletin: Performance Claims, costs, methodology, backtests and the non-predictive nature of past performance.