Trading Expectancy: The One Number That Shows Your Edge (2026)

The Expectancy Formula in Plain Text

Trading expectancy is the estimated average net outcome of one trade. It combines how often each outcome occurs with the average size of that outcome. It is more informative than win rate or planned reward-to-risk on its own, but it is still an estimate from a sample—not a promise that the next trade, month, or market regime will be profitable.

The complete discrete-outcome form is E = p_winW − p_lossL + p_flat0. Use the same unit throughout: account currency, points, or R-multiples. Do not combine a currency-denominated average win with an R-denominated average loss.

There is an even simpler audit formula:

An estimated positive expectancy means the sample average is above zero. It does not prove the underlying strategy will remain positive. An estimated negative expectancy is evidence that the sampled process lost per trade on average; it may reflect a weak strategy, execution errors, costs, an unfavorable regime, or a mixture of different setups. Zero is merely the sample breakeven point after whatever costs are included.

Expectancy in R-Multiples

R normalizes every trade by its initial planned risk. If one trade risked one unit and finished at +2.4 units, its outcome is +2.4R. If another finished before the stop at −0.7 units, its outcome is −0.7R. This lets you compare trades taken at different position sizes.

Use initial risk measured at entry, and preserve the realized result of partial exits, slippage, commissions, and stop changes. A planned 3R target does not make a trade a 3R winner if the position was scaled out for a realized +1.2R. For the interaction between hit rate and realized payoff, see the guide to win rate versus risk-reward.

Example 1: Positive Sample Expectancy

Suppose a trend-following sample contains 200 closed trades: 82 winners, 108 losers, and 10 scratches. Average realized winner is +2.4R and average realized loss is −0.9R.

• Win contribution: 0.41 × 2.4R = 0.984R
• Loss contribution: 0.54 × 0.9R = 0.486R
• Scratch contribution: 0.05 × 0R = 0R
• Sample expectancy: 0.984R − 0.486R = +0.498R per trade

The sample total is 99.6R, and 99.6R ÷ 200 produces the same +0.498R arithmetic mean. That identity is a useful spreadsheet check. It does not establish that the next 200 trades will produce 99.6R: outcome probabilities, payoff sizes, costs, and market conditions can change.

Example 2: A High Win Rate With Negative Expectancy

Now consider 300 reversal trades: 186 winners, 108 losers, and 6 scratches. The win rate is 62%, but the average winner is only +0.45R while the average loss is −1.1R.

Winning more often did not overcome the payoff imbalance. The diagnosis is not automatically “widen every target” or “tighten every stop.” First inspect whether large losses came from rule breaks, gaps, a particular session, or a distinct setup. Changing exits without that diagnosis can reduce the win rate and leave expectancy unchanged—or worse.

Example 3: Costs Turn a Thin Edge Negative

Assume a fixed-outcome strategy wins 50% of trades at +1.02R and loses 50% at −1R. Before costs, expectancy is (0.50 × 1.02R) − (0.50 × 1R) = +0.01R. If the all-in average execution cost is 0.04R per trade, net expectancy becomes −0.03R.

Where possible, calculate directly from broker-confirmed net outcomes so commissions and fees are already included. Add spread, slippage, financing, borrow, data, or platform costs only when they are economically attributable and not already counted. FINRA’s investor guidance likewise emphasizes that fees and transaction costs reduce returns; the practical lesson is to reconcile the same cost once, not omit it or double-count it.

Break-Even Win Rate

For a two-outcome setup with a fixed average win W, fixed average loss magnitude L, no scratches, and costs already embedded in outcomes, break-even win probability is:

If W and L are gross outcomes and a constant cost c is subtracted from every trade, solve pW − (1 − p)L − c = 0. Then p = (L + c) ÷ (W + L). With W = 2R, L = 1R, and c = 0.05R, the break-even rate rises to 35%. Real costs and payoffs usually vary, so this is a scenario model, not a substitute for the sample mean.

Expectancy Table: Win Rate × Realized Payoff

The table below is purely mathematical. It assumes two outcomes, every loss equals −1R, the stated reward is the average realized winner, and costs are zero. It is useful for sensitivity analysis, not as a benchmark for what a trading style “should” achieve.

A strategy does not improve merely because you select a more attractive cell. Increasing the target can reduce the hit rate; tightening a stop can change both the loss distribution and execution quality. Model the coupled change, then test it on unseen trades.

How to Calculate Expectancy From Your Journal Data

Add a Confidence Interval

A point estimate such as +0.18R hides sampling error. One practical approach is a nonparametric bootstrap:

1. Keep the chosen cohort fixed and resample its complete trade outcomes with replacement.
2. For each resample, calculate the mean net outcome.
3. Repeat many times and report a preselected interval, commonly the central 95% of bootstrap means.
4. Repeat the analysis on a chronological holdout or later forward sample before changing risk.

If the interval crosses zero, the sample and method have not resolved the sign of expectancy. That does not prove the strategy has no edge; it means “positive” is not yet a stable conclusion from this evidence. NIST’s guidance on confidence intervals explains why a sample mean should be accompanied by an interval estimate rather than treated as the population value.

Ordinary resampling assumes observations are exchangeable. Trading outcomes often cluster by volatility, market regime, or execution conditions. If sequence dependence matters, preserve chronology in a walk-forward test or use a defensible block-resampling method. Do not optimize the setup and estimate its final confidence on the same data. Research on backtest overfitting shows why repeated strategy selection can make historical results look stronger than they are.

Can Positive Expectancy Still Lose?

Yes. A strategy with a positive underlying expectation can produce a losing sample, and a strategy with a negative expectation can produce a winning sample. Variance does not disappear because the mean is positive.

For a prespecified block of ten independent trades with a 40% win probability, the probability that all ten are losses is 0.6¹⁰, about 0.605%. That calculation is not the same as saying such a streak occurs “once every 165 trades.” Rolling sequences overlap, real trades may not be independent, and changing win probabilities alter the result. Use streak math as a scenario, not a calendar prediction.

Position size must survive plausible adverse sequences and estimation error. Expectancy alone does not specify safe risk, maximum drawdown, ruin probability, or liquidity needs.

Expectancy vs Profit Factor

Expectancy is net outcome divided by trade count. Profit factor is gross profit divided by the absolute value of gross loss. Both summarize the same trade set from different angles, and both can be distorted by outliers, cost omissions, or mixed regimes. The guide to what a profit factor means explains its denominator and failure modes.

Multiplying sample expectancy by a future trade count gives a planning expectation under strong assumptions: the distribution, execution, sizing, and opportunity rate remain comparable. It is not a forecast or an income promise.

How to Improve Negative Expectancy

The formula identifies four numerical levers—win frequency, loss frequency, average win, and average loss—but it does not tell you which behavior caused them. Diagnose before changing rules.

1. Find the Cohort Creating the Drag

Split results by predefined setup, session, instrument, direction, volatility state, and rule compliance. Do not mine dozens of filters until one looks good. Start with a plausible mechanism, make one comparison, and retain the losing cohort in the record.

2. Separate Strategy From Execution

Compare planned stop and target with realized exits. If rule-breaking losses drive the result, the intervention is execution discipline. If compliant trades remain negative, the setup itself needs revision or retirement. A structured trade review keeps that distinction visible.

3. Inspect Exit Evidence

Use maximum adverse and favorable excursion to test whether stops are routinely exceeded before recovery or winners repeatedly travel farther after exit. The MAE/MFE analysis guide explains the measurement. Excursion data can suggest a hypothesis; it cannot by itself prove that a new stop or target would have filled at the modeled price.

4. Test One Change Out of Sample

Write the rule before the next sample, including what would falsify it. Keep position size stable, log exceptions, and compare the new chronological cohort with the baseline. If you tune repeatedly on the same trades, the apparent improvement includes selection luck.

Expectancy and the Kelly Criterion

John L. Kelly Jr.’s 1956 paper addressed long-run growth when probabilities and payoff conditions are known. In the simple binary, fixed-payoff case, the familiar fraction is f* = p − (q ÷ b), where p is win probability, q = 1 − p, and b is net payoff per unit lost.

Trading outcomes are usually variable, serially dependent, capacity-constrained, and estimated from limited data. Substituting sample win rate and average payoff into the binary formula is therefore a sensitivity scenario, not proof of an optimal or safe account fraction. Estimation error can make a positive Kelly result dangerously high, and the formula does not know your drawdown limit, obligations, leverage mechanics, gap risk, or broker liquidation rules.

Model the simple binary Kelly scenario, but treat the output as a stress-test input—not a position-size recommendation. There is no universal “professional” Kelly fraction or default percentage that this guide can prescribe.

A Weekly Expectancy Review That Leads to Action

1. Reconcile: match closed-trade count and net outcome to the source statement.
2. Freeze the cohort: name the setup, date range, market, session, account, and rule version.
3. Calculate: sample mean, conditional formula, counts, median, extremes, and chosen uncertainty interval.
4. Compare: current chronological cohort versus the prior comparable cohort and an untouched holdout where available.
5. Explain: identify one mechanism supported by linked trades, not just a changed metric.
6. Act: write one behavior or rule to test, its stop condition, and the review date.

Trader’s Second Brain calculates expectancy from imported closed-trade results and exposes setup-level expectancy alongside sample size and drift. That removes arithmetic work, but you still need to verify grouping, costs, tags, and the question being tested. Import a reconciled sample and review expectancy by setup; the product is ours, and no journal can guarantee that a historical edge will persist.

If you want a quick two-outcome sensitivity check before working with the full journal, use the break-even and expectancy calculator. Keep the calculator assumptions beside the result.

Primary Sources and Method Notes

• Interactive Brokers webinar material states the win-probability × average-win minus loss-probability × average-loss expectancy form.
• NIST’s confidence-interval guidance explains why a sample mean is an estimate with sampling uncertainty.
• Kelly’s original 1956 paper is the primary reference for the long-run growth criterion.
• Bailey and coauthors on backtest overfitting documents the selection problem created by repeated strategy trials.
• FINRA’s fees and commissions guidance supports using net rather than selectively gross outcomes.

Last fact review: September 7, 2026. Formulas are educational measurement tools, not individualized investment advice or a prediction of returns.