Risk–Reward Ratio, Win Rate and Trading Expectancy Explained

Trading expectancy connects win rate with the sizes of actual gains and losses. An approach that wins most trades can still lose money if occasional losses are large, while one with a lower win rate can have positive hypothetical expectancy when average gains sufficiently exceed losses. Trading costs and rare adverse outcomes complicate both statements.

This guide explains expectancy, break-even win rates, reward-to-risk ratios, profit factor and R-multiples through worked examples, then examines uncertainty, clustering, drawdowns and backtesting bias.

What is trading expectancy?

Trading expectancy is an estimate of the average profit or loss generated by a strategy per trade under a specified set of probabilities and outcome sizes. In a simple two-outcome model it equals the probability of a win multiplied by the average win, minus the probability of a loss multiplied by the average loss. The values should account for trading costs to be economically useful.

Expectancy is not a guaranteed next-trade return. Probabilities and average outcomes estimated from past data can change with market conditions, transaction costs, strategy rules and execution. A small positive historical mean may be statistically uncertain or driven by a few unusual results.

The expectancy formula

A simplified calculation uses E = p × W − (1 − p) × L, where p is the assumed win probability, W is average positive outcome and L is the positive magnitude of average negative outcome. If outcomes include breakeven trades, fees or multiple categories, the model must be adjusted or represented as a weighted average across all outcomes.

The units should be consistent: pounds, dollars, percentage of capital or R-multiples. Mixing average winner percentage on margin with loser pounds on notional exposure produces a meaningless result. A coherent model must also specify what counts as a trade and whether partial exits are combined.

Worked expectancy example

Suppose a hypothetical strategy wins 40% of its trades, with an average winner of +2R and average loser of −1R, where R is the chosen initial model risk amount. Its simplified expected result is 0.40 × 2R − 0.60 × 1R = +0.20R per trade before any costs not already included.

If expected additional execution and financing costs average 0.15R per trade, estimated net expectancy becomes +0.05R. Such a small estimated advantage may not be robust to sampling error, changes in slippage or a few extreme losses.

What is the risk–reward ratio?

The term risk–reward ratio can be expressed in different orders, creating confusion. A plan with £100 of intended loss distance and £200 of hypothetical target distance may be described as risk-to-reward 1:2 or reward-to-risk 2:1. State the numerator and denominator rather than assume the notation is universal.

A chart showing twice as much potential target distance as stop distance does not establish a favourable expected outcome. The target may be rarely reached, the stop may gap, and the eventual average winner and loser may differ substantially from the original distances.

Planned versus realised reward-to-risk

A trader can plan a target at +3R, but if positions are routinely closed at +0.8R due to early profit-taking, the actual average winner is not 3R. Similarly, a planned −1R stop may produce −1.5R average losses after gaps or poor fills. Expectancy calculations should use achieved outcomes rather than decorative chart ratios.

For a strategy backed only by hypothetical tests, state that fills are simulated and include uncertainty. A claimed risk–reward ratio should distinguish trade planning from historical or live execution evidence.

Break-even win rate

In a simplified model with fixed average gain W and average loss magnitude L, the gross break-even probability is p = L ÷ (W + L). For average win 2R and average loss 1R, the break-even win rate is 1 ÷ 3 ≈ 33.3% before extra costs.

If trading costs reduce net winners and enlarge net losers, the required win rate rises. This is why an apparently favourable target multiple cannot be assessed independently of spread, fees, slippage and actual exit behaviour.

Worked win-rate comparison

Consider hypothetical Approach A with 70% winners averaging +0.5R and 30% losers averaging −2R. Its expectancy is 0.70 × 0.5 − 0.30 × 2 = −0.25R. Frequent winners conceal an unfavourable loss distribution.

Approach B with 35% winners averaging +3R and 65% losers averaging −1R produces 0.35 × 3 − 0.65 = +0.40R before additional costs. This does not establish that Approach B is viable; a realised market history may contain rare losses much larger than its assumed −1R.

Profit factor explained

Profit factor is gross profit from winning trades divided by the absolute gross loss from losing trades over a specified sample, under a consistent calculation. A value above one indicates that realised gains exceeded realised losses for that sample before any costs excluded from the calculation. It does not establish future profitability or the severity of interim drawdowns.

Suppose four hypothetical wins of +2R and six losses of −1R produce gross gains of 8R and gross losses of 6R. Profit factor is 8 ÷ 6 ≈ 1.33. The average outcome is (8R − 6R) ÷ 10 = +0.20R, consistent with the earlier 40% win example.

Risk multiples and the meaning of R

An R-multiple expresses the realised outcome relative to a chosen initial risk reference. If initial intended model risk is £100 and a trade earns £250 net under the same convention, its outcome is +2.5R. A realised loss of £180 is −1.8R. The denominator should be fixed and recorded at entry rather than redefined later to make outcomes look favourable.

Because planned stop distance is not a guaranteed loss limit, actual losses can be worse than −1R. A performance report that artificially caps losses at −1R by construction can conceal the economic effects of gaps and market-impact costs.

Why sample size changes confidence

A strategy with ten winning trades may have been lucky, and estimates of average winners, losers and hit rate become uncertain when observations are few. Even a hundred trades can provide a weak basis for a rare-event or highly skewed strategy, particularly if all occurred in one market regime.

Independence matters: trades driven by the same macroeconomic event are not necessarily separate observations of one underlying edge. Counting multiple correlated positions as independent evidence overstates effective sample size. Statistical confidence should reflect clustering and changing market conditions.

Losing streaks and probability intuition

A positive estimated expectancy can coexist with long losing sequences. Even if each trade independently has a 60% chance of losing under a simple model, several consecutive losses are not extraordinary. Real markets often violate the independence assumption because returns cluster during regimes.

A strategy should be evaluated for the ability to withstand drawdowns and funding needs, not merely a positive average. An expected return does not tell the investor when losses will occur or whether capital will remain available long enough for any theoretical average to emerge.

Variance, skew and tail losses

Two approaches with identical expected return can have very different risk. One may deliver relatively small, balanced outcomes, while another earns frequent small amounts but occasionally suffers a very large loss. Variance, skewness, tail dependence and liquidity risk help explain why mean expectancy is insufficient.

For example, a strategy selling insurance-like exposure may show a high win rate until a crisis causes a single unusually large loss. A sample that excludes the crisis overstates the reliability of its average. Scenario analysis and worst-case economic exposures are necessary complements to expected value.

Expectancy after transaction costs

Costs need to be associated with actual trade frequency, order size and market conditions. Commissions, spreads, market impact, funding, borrowing and taxes where applicable can meaningfully change results. High-turnover strategies with small expected gains may be particularly sensitive.

Suppose a strategy has estimated gross expectancy of £8 per trade and £10 of realistic round-trip costs. Net expectancy is −£2 under those assumptions. If market impact increases with order size, scaling exposure may further reduce net outcomes. A backtest using free instantaneous execution would misrepresent the expected economics.

Expectancy across changing market regimes

The same rules can have different outcome distributions during strong trends, volatile reversals or quiet trading ranges. A model trained only on an extended equity bull market can appear to have a positive expectation because the market itself rose, not because the rule added value. Risk-matched benchmark comparisons are essential.

Do not divide historical data into dozens of regimes and select the best-performing filter without correcting for the search. Definitions chosen retrospectively can create false evidence. A claim that a strategy works only under special conditions requires independent tests of those conditions.

Payoff targets and probabilities interact

Moving a profit target farther away can increase the potential average win while reducing the chance the target is reached. Placing a stop closer can reduce the nominal loss distance while making it more likely to activate under normal price variation. The two inputs cannot be optimised independently of the market’s actual return distribution.

An appealing target-to-stop ratio is therefore not proof of favourable expectancy. A reliable evaluation uses the full distribution of realised exits, includes trades that never reached either level, and tests what happens during gaps and exchange closures.

Partial exits and average outcome calculation

When a trade is reduced in stages, a consistent trade-level performance convention is needed. A position half-closed at +1R and half-closed at +3R produces +2R overall before costs, assuming each half represents equal starting units and the R denominator is defined consistently.

Reporting two winning “trades” from one underlying idea can distort the number of independent observations. Partial exits, pyramiding and re-entry need transparent accounting rules so win rates and expectancy can be compared meaningfully over time.

Compounding and sequence risk

Average percentage outcomes do not completely determine long-run wealth because losses and gains compound. A 50% portfolio decline requires a 100% subsequent return to recover the original starting value. Sequence matters when leverage or withdrawal requirements can force reductions at unfavourable times.

Growth-optimal sizing formulas rely on estimated return distributions that may be inaccurate. Overconfidence in a positive expected value can create ruin risk if exposure is scaled aggressively. Risk constraints and capital survival remain critical even when a model suggests an advantage.

Expectation versus actual outcomes

A probability-weighted expectation is a property of an assumed distribution, not the return of any single future trade. A positive hypothetical expectancy cannot prevent a long negative sample or a structural change in the market. Equally, an observed short profitable sequence does not verify that the expectation is truly positive.

Researchers should report estimation uncertainty, the sample period and whether the results are live, paper-traded or simulated. Hypothetical outcomes often omit the psychological and operational difficulties that arise when real capital and margin demands are involved.

Assessing slippage and order quality

A strategy’s order assumptions should be compared with observed quotes, trading volume and executable market depth. Slippage is often worse during volatility and when many participants respond to the same signal. A position size that appears feasible in a backtest may overwhelm available liquidity when traded live.

The difference between signal price and actual fill can dominate a small theoretical edge. An evaluation should also consider whether exits would be available at the assumed price during a gap or market halt, rather than subtract only a token flat commission.

Comparing strategies using common units

Comparisons need a consistent definition of capital, leverage, currency, trade outcome and transaction costs. A strategy reporting +0.5R per trade with a large risk budget is not automatically preferable to one with +0.2R and lower portfolio drawdown. R is a model-based reference, not an identical monetary amount across systems.

Report drawdown, exposure, turnover and worst observed losses alongside expectancy and profit factor. Two approaches with the same historical average can have very different chances of surviving an adverse sequence without a forced liquidation or other funding problem.

How backtesting can inflate expectancy

Choosing the best entry and exit parameters from thousands of candidates can make a random pattern look profitable. Survivorship bias removes failed assets from the sample, while look-ahead bias uses information unavailable at the simulated decision time. Unrealistic fills, ignored costs and rare-event omissions can further inflate apparent expectancy.

Use pre-specified methods, truly untouched evaluation periods and stress tests. The CFA Institute’s 2026 curriculum and the CFTC’s warnings about hypothetical trading systems provide appropriate methodological context. See Backtesting Explained.

A practical expectancy research checklist

Define the exact trade unit and accounting of partial exits. Calculate all outcomes net of estimated or observed costs, then report average winner, average loser, frequency and number of observations. Examine extreme losers, clustered periods and changes in market regime. Compare results with a reasonable risk-matched benchmark.

Do not conclude that a high target multiple or win rate alone establishes a viable strategy. Any quoted positive expectancy is a historical or hypothetical estimate subject to sampling, execution and structural uncertainty, not a guarantee of financial return.

Worked expectancy and profit factor table

Illustrative ten-trade sample; a positive historical average does not demonstrate a robust edge.

Why average results should include rejected and unfilled signals

A model can generate a signal without an actual completed trade because the instrument is illiquid, the limit order is not filled or a pre-existing risk constraint rejects the opportunity. Ignoring those cases can exaggerate the strategy’s apparent scalability. The set of executed outcomes is not necessarily a random sample of all signals.

A reliable research log distinguishes the theoretical signal, the order actually submitted and the quantity filled. It should record implementation costs and whether rejected signals would have produced different returns. The purpose is not to force every signal to become a trade, but to assess a model as it could truly operate with market access and risk constraints.

Trading expectancy and uncertainty around rare losses

Some strategies have frequent modest winners with occasional very large losers. A limited backtest may observe few or no such events and therefore materially overstate estimated expectancy. A strategy with an attractive average should still be challenged using severe price gaps, changing correlations and loss magnitudes outside the observed sample.

Without adequate tail-risk information, a positive mean is not enough to infer suitable sizing or future capital growth. Stress tests can make the uncertainty visible, but they cannot assign reliable probabilities to events that have not occurred in the available data.

Frequently asked questions

What matters more, win rate or risk–reward ratio?

Neither in isolation. The full distribution of net winners, net losers and trade costs determines historical expectancy.

What is break-even win rate for 2R winners and 1R losers?

One-third, about 33.3%, in the simplified gross two-outcome model with no additional costs.

Can a strategy be profitable with only 40% wins?

Mathematically yes, if average gains sufficiently exceed losses after costs, but real-world evidence is still required.

Does profit factor above one guarantee future returns?

No. It describes one sample and may be dominated by a few trades or selection bias.

Can a positive expectancy still lead to large drawdowns?

Yes. Outcome order, tail losses, leverage and clustered market regimes can create severe drawdowns.

Related Education

Explore position sizing, portfolio drawdowns, strategy rules and backtesting bias.

Sources and further reading

• CFA Institute — Backtesting and Simulation
• CFTC — Hypothetical Performance Limitations
• SEC Investor.gov — Investment Risk

Further structured learning: Explore the Risk Management Masterclass for more systematic education in position sizing, trading risk and disciplined decision processes. This course provides education, not trading signals or personalised financial advice.

Important notice: General educational mathematics only. Hypothetical expectancy and R-multiples are not live or verified performance claims. Estimates may be wrong, costs and gaps may be larger than expected and trading can lead to substantial or total loss.