An edge is a statistical asymmetry: a repeatable reason your average outcome over many trades is positive after costs. The metric that quantifies it is expectancy — the mean profit or loss you can expect per trade, usually expressed in units of risk (R). The core tension is that an edge is a property of a distribution of many trades, not of any single trade: a system can have a genuine positive expectancy and still lose on the next dozen trades, while a system that wins 9 times in 10 can be a guaranteed loser if the rare loss is large enough. Defining an edge therefore means separating the two axes that actually drive profitability — how often you win (win rate) and how much you win versus lose (payoff) — and measuring whether their combination clears zero with enough certainty to bet on.

How it's calculated

The general expectancy formula, popularized for traders by Van Tharp, is:

> Expectancy = (Win% × Average Win) − (Loss% × Average Loss)

where Win% + Loss% = 100% (scratch trades go in whichever bucket the data shows). Expressed in dollars, this is the average dollar result per trade. The more robust form normalizes everything to R-multiples.

• R is the initial risk on a trade — the dollar distance from entry to the planned stop, times position size. It is the unit of account.
• An R-multiple is any trade's result divided by its initial R. A trade risking $200 that nets +$600 is +3R; one stopped out at the planned stop is −1R.
• Expectancy (in R) = the mean of all your R-multiples. This is Van Tharp's key reframing: expectancy is simply the average R generated per trade. An expectancy of +0.4R means that, on average, every dollar risked returns $0.40 over a large sample.

R-normalization is what makes the metric portable: it strips out account size and position size, so you can compare a $500 risk trade with a $5,000 one, or one system against another, on the same scale. (Sources: Van Tharp, Trade Your Way to Financial Freedom; TraderLion; pnlledger.)

Expectancy ≠ profitability rate. Two systems with identical +0.4R expectancy are not equally good if one trades 20 times a year and the other 200. Van Tharp captures this with opportunity (trade frequency) and the composite Expectancy Score ≈ Expectancy × Opportunity (he calls the combined idea "expectunity"). His worked illustration: a system at 1.0R × 50 trades returns ~50R/year, while one at 0.2R × 500 trades returns ~100R/year — the lower-expectancy system is the better business. (Sources: Van Tharp; Hightick; easylanguagemastery.)

How it's used in practice

1. Pass/fail screen. Positive expectancy after realistic costs (commissions, spread, slippage) is the minimum bar for trading a system live. Negative expectancy cannot be salvaged by position sizing — sizing controls how fast you go broke, not whether you do. 2. Decoupling win rate from payoff. Expectancy shows that a low win rate is fine if the payoff is large (trend-following systems often win 30–40% with large multi-R winners), and a high win rate can still lose (selling premium / mean-reversion systems with rare large losses). This kills the beginner instinct to chase win rate. 3. Position sizing input. Expectancy plus win rate feed the Kelly criterion, which gives the bet fraction that maximizes long-run geometric growth. A common form is Kelly% = Win% − (Loss% ÷ payoff ratio), equivalently Expectancy ÷ Average Win. Because Kelly's inputs are estimates, practitioners almost universally use fractional Kelly (half or quarter) — half-Kelly is commonly cited (after MacLean, Ziemba & Blazenko, 1992) as capturing roughly three-quarters of full Kelly's growth rate at about half the variance/drawdown — or fall back to a fixed 1–2% risk per trade. 4. Comparing and forward-tracking systems. Expectancy (and Expectancy Score) is the apples-to-apples yardstick for ranking strategies, and tracking realized vs expected R over time is an early-warning that an edge is decaying.

Adoption, debate & evidence

The expectancy/R-multiple framework is the dominant vocabulary among discretionary and systematic retail traders and is taught nearly everywhere (Van Tharp, TraderLion, prop firms). The underlying mathematics — expected value, Kelly sizing — is uncontested and is the same machinery used in gambling and quantitative finance.

The genuine debate is not about the formula but about measurement reliability. Expectancy is an estimate from a sample, and the sample is usually too small and too contaminated to trust:

• Sample size. Roughly 30 trades is a bare statistical minimum; commonly cited practitioner guidance puts a meaningful evaluation at 100–200+ trades, and even then confidence intervals on expectancy are wide. A handful of profitable trades is not evidence of an edge.
• Backtest overfitting is the dominant failure mode. Bailey and López de Prado show that running even a few backtest variants reliably produces strategies that look significant but aren't, and that the more a strategy is tuned, the larger the gap between backtest and live results. Their Deflated Sharpe Ratio and Minimum Backtest Length formulas exist specifically to penalize this selection bias — a raw Sharpe of 2.0 across 100 trials can deflate toward noise.
• Non-stationarity. Estimated win rate and payoff drift with market regime; an edge measured in one regime can vanish in another. This is the main reason full Kelly is dangerous — it assumes the inputs are exact.

The honest summary: positive measured expectancy is necessary but not sufficient evidence of a real edge. The edge is real only if the positive expectancy survives realistic costs, out-of-sample data, a large sample, and a correction for how many variants you tried.

Strengths & limitations

Strengths. Single number that integrates win rate and payoff; R-normalization makes systems and accounts comparable; directly drives sizing; positive-vs-negative is an unambiguous go/no-go.

Limitations. It is an average — it hides the variance and the worst-case drawdown that actually blow up accounts, so it must be read alongside the R-distribution, max drawdown, and worst losing streak. It is only as good as its inputs: in-sample, cost-free, or small-sample expectancy is routinely optimistic. The single most common misuse is treating a positive expectancy from a short or overfit backtest as proof of an edge — and the second is conflating high win rate with high expectancy.

Sources

• Van Tharp, Trade Your Way to Financial Freedom — expectancy, R, R-multiples, opportunity/expectancy score (summaries: vantharpinstitute.com/tharp-think-trading-concepts; hightick.com/van-k-tharps-expectancy-function)
• TraderLion — "R and R-Multiples" (traderlion.com/risk-management/r-and-r-multiples)
• P&L Ledger — "Expectancy & R-multiples: the plain-English guide" (pnlledger.com/expectancy-r-multiples-the-plain-english-guide)
• EasyLanguage Mastery — "Rank Your Trading System With Expectancy Score" (easylanguagemastery.com)
• Kelly criterion references — Zerodha Varsity (zerodha.com/varsity/chapter/kellys-criterion); fractional-Kelly practice (enlightenedstocktrading.com/kelly-criterion)
• MacLean, Ziemba & Blazenko, "Growth versus Security in Dynamic Investment Analysis" (Management Science, 1992) — half-Kelly ≈ ¾ growth at ≈ half variance (see also stat.berkeley.edu/~aldous/157/Papers/Good_Bad_Kelly.pdf)
• Bailey & López de Prado, "The Deflated Sharpe Ratio" (davidhbailey.com/dhbpapers/deflated-sharpe.pdf); backtest-overfitting and Minimum Backtest Length
• Sample-size guidance — backtestbase.com/education/how-many-trades-for-backtest; medium.com/@trading.dude statistical-significance guide

Confidence note: formula and R-multiple definitions are well-corroborated (medium). The "100–200+ trades" and fractional-Kelly "~75% of growth" figures are commonly-cited practitioner heuristics, not single canonical results — qualified as such.