Win rate does not determine profitability without outcome sizes and costs. In a hypothetical model, 70% wins of USD 50 and 30% losses of USD 200 give a mean of −USD 25 per trade. Expectancy is the mean of a stated payoff distribution; a journal supplies an estimate from selected past outcomes. A positive estimate alone does not decide future profitability or compounded growth.

What expectancy really is

Expectancy is the mean profit or loss per trade in a specified distribution. Estimate it from recorded net outcomes as win rate × average win minus loss rate × average loss, with losses expressed as positive magnitudes in that formula. Breakeven trades must remain in the total trade count. There is no mathematical minimum of exactly one hundred trades; sample uncertainty depends on dispersion, dependence and representativeness.

Win rate needs the outcome sizes to describe profitability. With 70 percent winners of 1 unit and 30 percent losers of 3 units, expectancy is 0.7 − 0.9 = −0.2 units before any additional costs. A hundred observations do not guarantee that randomness has cancelled out or that the estimated mean will persist.

Why win rate on its own can mislead

The cleanest way to understand this is to compare three hypothetical strategies with different combinations of hit rate and average win-to-loss ratio. Treat all three as running on the same ten thousand euro account, trading the same EUR/USD pair.

Three hypothetical strategies with the same starting capital
Strategy A — high hit rate, poor ratioSeventy percent winners, average win of fifty euros, average loss of one hundred euros. Seven tenths of fifty euros gives thirty-five euros of statistical profit per trade, from which we subtract three tenths of one hundred euros, or thirty euros of statistical loss. The end result is exactly five euros per trade — before execution costs; assess any tax treatment separately under the applicable rules.
Strategy B — low hit rate, strong ratioFifty percent winners, average win of one hundred euros, average loss of fifty euros. Half of one hundred euros gives fifty euros of statistical profit, half of fifty euros gives twenty-five euros of statistical loss. The end result is twenty-five euros per trade — five times more than Strategy A, despite the lower win rate.
Strategy C — deceptively balancedSixty percent winners, average win of eighty euros, average loss of one hundred and twenty euros. Six tenths of eighty euros gives forty-eight euros of statistical profit; four tenths of one hundred and twenty euros gives forty-eight euros of statistical loss. The mean is zero before costs; an individual sequence can still end with a gain or loss.
ConclusionUnder the stated hypothetical payoff distributions, B has mean EUR 25 per trade versus A’s EUR 5, while C has mean zero before costs. These means do not guarantee realised sequence outcomes.

Suppose omitted execution costs average EUR 6 per trade in each hypothetical model. Net means become EUR −1 for A, EUR 19 for B and EUR −6 for C. If a journal already records net outcomes, do not deduct those costs again. Tax treatment is a separate investor-specific calculation, not a fixed universal charge per trade. See discovering a trader’s edge and the risk-reward glossary entry.

R-multiples — outcomes relative to initial risk

In the Van Tharp framework, R is initial planned risk for the individual trade. Calculate its signed outcome divided by that R, with a consistent cost definition. R need not equal 1% or 2% of capital, and an actual loss can exceed −1R. This normalises scale but does not make strategies identical: stop risk, execution and payoff distributions still differ. If risks vary, calculate each R-multiple first and then average.

A concrete hypothetical example: a strategy with a fifty percent win rate, an average win of two R and an average loss of one R. Half of two R gives one R of statistical profit, half of one R gives half an R of statistical loss. The difference works out to half an R per trade — each position earns, on average, half a unit of risk. This figure is independent of account size. Run the same strategy at a forty percent win rate and the sum becomes 0.4 × 2R minus 0.6 × 1R, that is 0.8R minus 0.6R, leaving 0.2R per trade — considerably less, but still on the positive side of zero.

If the assumed mean remains 0.5R, the expected sum across one hundred trades is 50R. With fixed EUR 100 initial risk per trade, this represents expected EUR 5,000 before any costs omitted from the outcome definition. Recalculating risk from changing equity makes R vary in euros. A fixed-fraction binary model can derive final equity from outcome counts, but variable payoffs, costs and sizing constraints need trade-level inputs; sequence also matters for drawdown. Neither model promises a realised profit. See the one-percent sizing rule.

Expectancy in R is the average of individual trade R-multiples. It describes the selected outcomes; using it to estimate future performance requires assumptions about the sample’s representativeness.

Expectancy thresholds — what each number means in practice

The following bands are descriptive sample-mean labels, not validated quality grades, personal trading rules or capital-allocation thresholds. State whether outcomes include costs and how initial R is defined. Future expectancy still requires a justified model of the sample.

Practical interpretation of expectancy in R-multiples
Above 0.5RA large positive sample mean under the stated R and cost definitions; it does not establish durability across future regimes.
Between 0.3R and 0.5RA positive sample mean. Check whether costs are already included and assess uncertainty before drawing a profitability conclusion.
Between zero point one R and zero point three RA positive sample mean between 0.1R and 0.3R. If costs are already included, it is already net of them; its sign does not prove a persistent edge.
Between 0R and 0.1RA small positive sample mean; whether it covers costs depends on whether the outcomes are gross or net and on those costs.
Below zeroThe selected sample has a negative mean under its outcome definition. A future-distribution claim requires additional assumptions; individual profitable periods remain possible.

A mean alone does not reveal loss streaks or maximum drawdown. In a specified model, two distributions can share a mean and have different dispersion or rare losses. A positive arithmetic mean also does not guarantee positive compounded growth, as the diagram illustrates. Record the distribution and path rather than describing a few future trades as certain to deliver the edge.

A positive percentage mean is not compounding

Check units and each result's base before comparing means with final capital. A percentage of changing capital is no fixed cash amount.

Scroll sideways to see the full diagram

A positive percentage mean is not compoundingOur two periods:+50% of100 gives150;then−40% of150 gives90. Arithmetic mean return is+5%, but capital falls10% overall. Cash results+50,−60 hypothetical units average−5. No probabilities or expectation are assigned.Mean of two percentages: +5%Each percentage uses its period's capital.0%+50%−40%Period 1Period 2Capital(hypothetical units)10015010015090StartAfter period 1After period 2100 × 1.50 × 0.60 = 90Total capital return: −10%

Our hypothetical two-period sequence has no deposits, withdrawals or extra costs. Each return uses capital at its period's start. Large+50%,−40% aid arithmetic, not a target or proposed risk. Without a probability model,+5% is only the recorded returns' arithmetic mean, not an estimated next-result expectation.

The mean is positive, but successive multipliers reduce capital.

Mean return is (50%−40%)/2=+5%. Capital instead uses multipliers: 100×1.50×0.60=90, or−10% overall. This does not contradict earlier net-cash averages. Here cash results+50,−60 average−5 units, a different measure and unit from+5%. Do not substitute either in a journal or compounding forecast.

An honest caveat — what expectancy does not promise

Using a historical mean for future inference requires representative selection, consistent rules and an appropriate model of dependence and costs. Stable volatility or correlation alone does not validate that inference. Record rule and fee changes and compare consistently defined windows. Review tools are covered in trading statistics and key metrics.

Sample size — when a calculated expectancy becomes trustworthy

A mean from ten trades describes those outcomes but is usually a weak estimate of future expectancy. Report the small sample and its dispersion; do not treat a short winning streak as proof of a durable edge.

  • Thirty trades provide a limited sample. An uncertainty interval requires the observed dispersion and an appropriate dependence model; the sample count alone cannot turn 0.3R into a specified 0.15R–0.45R interval.
  • One hundred trades give more information than thirty under comparable sampling conditions. With independent observations and equal variance, standard error scales as 1/√n; it remains subject to dependence, selection and changes in the strategy or market.
  • Five hundred trades: under independent, equal-variance sampling, standard error is about 44.72% of its value at 100 trades. This does not establish a universal minimum for comparing variants.
  • Larger samples can improve precision when observations remain representative. Under an independent, stable finite-variance model, halving standard error takes four times as many observations. Trading returns can violate those assumptions.

For independent draws with common finite variance σ², the sample mean has standard error σ/√n, usually estimated with s/√n. A confidence interval also requires its distributional assumptions and critical value. Greater dispersion increases uncertainty; clustering and selection can invalidate the simple formula. Count alone does not give an error interval.

There is no mathematical 100-trade threshold that makes increasing capital justified. Before drawing an inference, examine sample selection, dispersion, dependence, costs and outcomes outside parameter fitting. A short record can reveal an error without proving an edge, and a long one can still be unrepresentative. See risk management basics.

What to do tomorrow

The theory is clear, but its real value appears only once you translate it into your own journal. Below are four concrete steps worth taking over the coming week, to move from the question "is my strategy actually earning?" to a numerical answer expressed in R-multiples.

  1. Choose a clearly defined sample of closed trades and record its size. Calculate the mean net outcome, including breakeven trades and consistent costs. A smaller sample can reveal errors and describe past performance, but any inference about future expectancy should acknowledge its uncertainty.
  2. Calculate each trade’s R-multiple. Divide that trade’s net result by its own initial planned monetary risk, then average those signed R-multiples. Dividing average results by the average risk can produce a different number when trade risks vary. Keep the definition of initial risk and costs consistent.
  3. Use the sign and size as a diagnostic under the recorded assumptions, together with dispersion and drawdown. Check why outcomes changed; 0.3R is not a universal permission to increase size, and reducing trade frequency does not automatically raise expectancy.
  4. Choose a review schedule and record it, but do not treat fifty trades as a universal validity window. A newly calculated mean can be noisy and cannot identify the moment a strategy has permanently stopped working.

Related reading: trader's risk management basics — the wider system into which expectancy fits; the one-percent rule — how to size a single R against the account; trading statistics in practice — which metrics should be recomputed alongside expectancy.

The same sample total, a different mean

About 2 min calculator practice

A separate fictional sample has 30 wins averaging USD 200 net, 50 losses averaging USD 100, and 20 trades exactly at zero after costs. The USD 1,000 result covers 100 trades, so the mean is USD 10. The win rate among non-zero trades is 37.5% and profit factor is 1.2. Change only the number of exactly zero-result trades from 20 to 40. What happens to mean P/L per trade, total P/L and the win rate excluding zero results?

Check the effect of zero-result trades

Check the result and explanation

P/L remains 30 × 200 − 50 × 100 = USD 1,000. The trade count rises to 120, so the mean falls to 1,000 ÷ 120 ≈ USD 8.33. The win rate excluding zero results stays at 30 ÷ 80 = 37.5%, and profit factor remains 6,000 ÷ 5,000 = 1.2. These statistics use different denominators.

This describes the entered sample, not a forecast. The calculator uses net cash amounts, not an average of percentage returns or R multiples. An exit at the entry price with commission is not a zero net result.