Trading edge

A trading edge is a positive expected outcome per trade after costs that persists on data the rule was not designed on; it is a statistical property of a repeated procedure, not a single good trade, and it can only be claimed with an out-of-sample test and enough trades to distinguish it from chance.

Senzoukria · Glossary · Updated September 2026


Expectancy, the number behind the word

Expectancy is the average result per trade: the share of winners times the average win minus the share of losers times the average loss. The beginner's guide works a hypothetical example, 40% winners averaging 3 ticks and 60% losers averaging 1 tick, to show that a high win rate is not required for positive gross expectancy. It also shows the other half: positive gross expectancy can disappear after costs. An edge is the net figure, after fees and slippage converted with the instrument's actual tick value.

Expectancy in dollars mixes strategies with different stops. Expressed in R, each trade divided by what it risked at entry, it compares across strategies: a $40 gain on a trade that risked $8 at its stop is 5R; the same $40 gain on a trade that risked $40 is 1R.

What separates an edge from a lucky sample

  • Out-of-sample survival: the rule keeps its sign on a period chosen before any parameter was tuned.
  • Enough trades: a handful of large winners can carry a whole run; a result that is too good for its sample size is a warning, not a confirmation.
  • Distinguishable from chance: a permutation test asks whether shuffled outcomes would produce a similar curve.
  • Survives the number of trials: the more configurations tried, the higher the bar; a deflated Sharpe ratio accounts for that.
  • Survives the path: drawdown within the account's limit and no death on the adverse periods.

Edge is conditional

An edge measured on one session, one bar construction and one regime is a claim about those conditions. The same rule can lose in another regime, which is why regime filters exist and why a walk-forward test moves chronologically instead of pooling everything. It is also conditional on execution: a bar-based test that enters at the next open assumes a fill the live queue may not give.

In Senzoukria

The Auto backtest panel's Gauntlet section is titled "is this edge real?" and lists its criteria by name: Account survives the whole run; Walk-forward: enough windows, most profitable out of sample; Permutation test: edge distinguishable from chance; Deflated Sharpe: survives the number of trials; Probability of backtest overfitting; Max drawdown within the account's limit; Survives the red periods; Prop-firm cycle. The previous-sessions view reports Expectancy / trade, defined as net P&L divided by the number of trades, and an R-based summary when trades carry a stop. Its verdict text explicitly states that "solid on this sample" is the most a backtest can ever prove.

Common mistakes

  • Calling a winning week an edge.
  • Quoting gross expectancy and forgetting that the fees are per contract and per order.
  • Optimising on the holdout, which turns the evidence for the edge into training data.
  • Assuming an edge in a simulated account transfers unchanged to a funded account with its own drawdown and consistency rules.

This page in other languages

Frequently asked questions

Can a strategy with a low win rate have an edge?
Yes. Expectancy depends on the average win and average loss as much as on the win rate. A rule that wins less than half the time can have positive expectancy if winners are large relative to losers, and a rule that wins most of the time can have negative expectancy if the occasional loss is large. Costs then decide whether the net figure stays positive.
How many trades are needed to claim an edge?
There is no universal number; it depends on the variance of outcomes relative to the mean. What matters is that the sample is large enough that the average is not dominated by a few trades, which is why the desktop's verdict flags results that are too good for their sample size and refuses to compute ratios on very few trades.
Does passing the Gauntlet prove future profitability?
No. The Gauntlet summarises a protocol run on the supplied data. Passing it means the result survived a set of named challenges on that data. Market conditions, execution and account rules can still differ going forward.

Keep reading