Why a high win rate can still lose money: expectancy, payoff and costs

Win rate answers one question only: what share of closed trades made money? It says nothing about how much those winners made or how much the losers cost. A record with a strikingly high win rate can still produce a net loss if the losing trades are large relative to the winning ones. This guide shows why that happens, how to compute expectancy from a trade record, and how to audit a performance claim before treating it as evidence.

A 100-trade example where 90% wins lose money

Consider a hypothetical record of 100 completed trades. In 90 of them the account gains +1R, where R is the amount initially risked under the record's own consistent definition. In the remaining 10 the account loses -10R each. The 90 winners contribute +90R in total. The 10 losers contribute -100R in total. The net result is -10R before costs, which is -0.1R per trade on average.

Every number here is labelled arithmetic on a constructed example, not data from any provider. The point is structural: count and magnitude are different axes. The win rate describes the first axis. Expectancy, described below, combines both.

Definitions used throughout this guide

• Win rate: the share of counted closed trades that produced a positive result. Loss rate: the share that produced a negative result. Both depend entirely on which trades are included in the denominator.
• Average win: total gross profit divided by the number of winning trades. Average loss: total gross loss divided by the number of losing trades, expressed as a positive magnitude.
• R-multiple: a trade outcome expressed as a multiple of R, the amount initially risked under the record's own consistent definition. A +2R trade gained twice the initial risk; a -1R trade lost exactly the initial risk.
• Gross profit: the sum of all positive trade results. Gross loss: the sum of all negative trade results as a positive magnitude. Net result: gross profit minus gross loss, then minus costs.
• Expectancy: the average net result per trade over the sample, computed with the formula given below.
• Payoff ratio: average win divided by average loss. Profit factor: gross profit divided by absolute gross loss.
• Drawdown: the decline from a peak equity value to a later trough within the same record. It measures path, not final outcome.
• Sample size: the number of observations behind a statistic. Outlier: a single outcome far from the rest of the distribution, capable of dominating averages.
• Realised versus hypothetical result: realised results come from executed trades. Hypothetical results are simulated, backtested or paper outcomes that did not involve the reported execution record.
• Closed versus open trade: a closed trade has been fully exited and its result is fixed. An open trade carries unrealised profit or loss that can change until exit.

The expectancy formula

Expectancy per trade = (win probability x average win) - (loss probability x average loss), where average win and average loss are entered as positive magnitudes. Applied to the opening example: 0.90 x 1R minus 0.10 x 10R equals 0.9R minus 1.0R, giving -0.1R per trade, matching the direct calculation.

This figure is an arithmetic summary of the observed sample. It is not a forecast. Future trades may follow different distributions, and the sample itself may be incomplete or selected. Treat expectancy as a description of what happened, conditional on the inclusion policy used to build the record.

Break-even win rate before costs

Before costs, the break-even win rate is: average loss divided by (average win plus average loss). If average win and average loss are equal, break-even sits at 50%. If average win is twice average loss, break-even falls to about 33%. If average loss is ten times average win, as in the opening example, break-even rises above 90%, which is why a 90% win rate there still loses money.

Costs shift this threshold. If a constant cost per trade is deducted from every outcome, the labelled approximation becomes (average loss + cost per trade) divided by (average win + average loss). Real costs vary by trade because spread, commission, financing and slippage depend on the instrument, session and holding period, so calculate the exact break-even point from trade-level net results.

Three labelled hypothetical profiles

Each profile covers 100 trades and uses no provider data. None is presented as good, realistic or recommended. They exist to show how the same statistics interact differently.

1. Profile A: 90 wins at +1R and 10 losses at -10R. Gross profit +90R, gross loss -100R, net result -10R before costs. Win rate 90%, expectancy -0.1R per trade.
2. Profile B: 40 wins at +2R and 60 losses at -1R. Gross profit +80R, gross loss -60R, net result +20R before costs. Win rate 40%, expectancy +0.2R per trade.
3. Profile C: 55 wins at +1R and 45 losses at -1R. Gross profit +55R, gross loss -45R, net result +10R before costs. Win rate 55%, expectancy +0.1R per trade.

Profile A has the highest win rate and the worst net result. Profile B has the lowest win rate and the best net result. Profile C sits between them. The ordering changes once costs enter, which the next section shows.

Cost stress test applied to Profile C

Assume an illustrative average total cost of 0.12R per completed trade, covering spread, commission, financing and slippage combined. Across 100 trades that subtracts 12R. Profile C's +10R gross result becomes -2R net. These cost figures are invented arithmetic for demonstration, not a fee estimate for any broker or instrument. Actual costs must be taken from the trader's own statements.

Fees and expenses reduce investment returns, and regulators advise reading disclosures and checking statements to see what was actually paid. Costs are not optional adjustments; they are part of the net result and must be reconciled against records. [4]

Payoff distribution: what an average hides

An average win or average loss collapses many outcomes into one number. A set of trades could contain dozens of small gains and one much larger loss, with the mean sitting between them. Two records with identical average win and average loss can still have different largest losses and different peak-to-trough paths.

Alongside averages, ask for the median outcome, the largest single win, the largest single loss, and percentile or ordered outcome data such as the best and worst deciles. This does not assume any particular statistical distribution. It simply prevents one tail outcome from disappearing inside a mean.

Profit factor and its limits

Profit factor is gross profit divided by absolute gross loss. Profile A gives 90R divided by 100R, or 0.90. Profile B gives 80R divided by 60R, about 1.33. Profile C gives 55R divided by 45R, about 1.22, all before costs.

If gross loss is zero, profit factor is undefined, not proof of safety. The observation that a selected sample contains no losing trades says only that no loss is present under that sample's inclusion policy. Profit factor inherits weaknesses in the underlying record, including selection bias and missing trades.

Drawdown and sequence risk

Take the same 100 trades from any profile and rearrange their order. The arithmetic mean is unchanged, but the path differs. If the ten -10R losses in Profile A arrive first, the running total reaches -100R before any winning outcomes; if they arrive later, the same final sum follows a different drawdown path. Sequence therefore belongs in the record alongside the final total.

This guide states no safe drawdown limit, because appropriate tolerance depends on capital, leverage, objectives and constraints that vary by individual. What matters for auditing is that drawdown is reported from peak to trough within the actual record, with dates, rather than omitted or smoothed away.

Sample-size sensitivity

Percentages move quickly on small samples. Over ten trades, a single additional winner moves the win rate by ten percentage points. Over one hundred trades, the same change moves it by one point. Small samples therefore make headline rates unstable and easy to improve by luck or by timing the start date.

More observations help stability, but they do not repair deeper problems. Selection bias in which trades were recorded, missing trades from failed executions or omissions, and changed rules mid-sample all distort a record regardless of its length. No sample size converts a curated list into independent evidence.

Denominator manipulation

The win rate is a ratio, so whoever controls the denominator controls the headline. Common adjustments include excluding breakeven trades, excluding open losers while keeping open winners, dropping cancelled calls, counting partial exits inconsistently, duplicating entries scaled into several positions, and selecting convenient periods while omitting drawdown months.

Any credible claim states an explicit inclusion policy: which trades count, how breakeven and open positions are treated, whether partial exits aggregate or split, and which date range applies. Without that policy, two people computing the same account history can publish different win rates, both technically derived from the data.

What a screenshot proves and what it does not

A screenshot or equity-curve image establishes only the values visible in that image. It may omit deposits, withdrawals, costs, leverage, open exposure and the completeness of the series. Without the underlying export and cash ledger, those omitted fields cannot be reconciled independently.

Stronger evidence is a complete trade export covering the stated period, paired with a cash ledger reconciling deposits, withdrawals, fees and ending balance. Those artefacts allow an independent reader to recompute gross and net results and check the claimed figures.

Hypothetical and backtested results

Regulators have long warned about simulated performance in promotion. The NFA's Interpretive Notice on Compliance Rule 2-29 states that hypothetical performance results have repeatedly produced misleading promotional material, do not represent actual trading, are generally designed with hindsight, may under- or over-compensate for liquidity and slippage, and cannot completely account for financial risk or the ability to follow a programme through losses. [1]

That warning concerns promotional use of hypothetical results. It does not imply that every backtest is fraudulent. Backtests can be useful analytical tools when assumptions are disclosed, costs are modelled and out-of-sample behaviour is tested. The audit question is whether the presenter distinguishes clearly between simulated and realised results, and whether hindsight bias, liquidity assumptions and slippage treatment are documented. [1]

Guaranteed returns and AI trading bots

The CFTC warns that fraudsters use AI language to promote trading bots and signals, sometimes with unreasonable or guaranteed-return claims, and states that AI cannot predict the future or sudden market changes. Guaranteed or implausibly consistent returns attached to automated systems are claims to verify, not facts to accept. [2]

This warning targets deceptive promotion. It does not establish that every automated system is fraudulent. Automation changes execution mechanics, not the arithmetic of expectancy, and any bot performance claim should face the same audit as a manual record: complete export, defined denominator, reconciled cash flow and separated realised versus hypothetical results. [2]

Past-performance rules in financial promotions

In the United Kingdom, the FCA Handbook contains scope-specific rules for retail financial promotions involving past, simulated past or future performance. COBS 4.6 addresses matters including prominence of warnings, reference periods, identification of sources, and the statement that past performance is not a reliable indicator of future results. [3]

Applicability depends on the communication, the product and the business conducting it. COBS governs regulated firms' communications within its scope; it is not automatically a rule for every educator, signal seller or social media screenshot. Nothing here is legal advice. For a specific communication, the relevant rules and any applicable exemptions need professional assessment. [3]

Costs belong in the calculation

The SEC advises that fees and expenses reduce investment returns and recommends reading disclosures and checking statements. Trading records can include spread, commission, financing and adverse slippage, so reconcile the applicable components to statements and fills before describing a result as net. TraderJury's separate cost guide explains the ledger method. [4]

How to audit a performance claim

1. Archive the exact wording and dates of the claim, including screenshots, posts and any edited versions, so later comparisons have a fixed reference.
2. Obtain a complete trade export for the stated period, not a summary table or image.
3. Define the denominator: which trades count, how breakeven and open positions are treated, how partial exits aggregate, and which date range applies.
4. Reconcile deposits and withdrawals against the equity curve so account growth is not confused with external funding.
5. Calculate gross profit, gross loss, net result after costs, win rate, expectancy and profit factor directly from the export.
6. Check the largest single loss and the maximum drawdown from peak to trough, with dates.
7. Separate realised results from hypothetical, backtested or paper results, and label each portion explicitly.
8. Record any strategy, rule, instrument or sizing changes during the period, since they break the assumption of a single stable process.
9. Date the conclusion and state the inclusion policy alongside it, so the finding is reproducible by another reader.

A reproducible monthly performance audit

Once a month, run the same procedure on your own record or on any record you are evaluating. Export all closed trades for the month. Apply the written inclusion policy without exception. Compute win rate, loss rate, average win, average loss, gross profit, gross loss, expectancy and profit factor. Subtract itemised costs reconciled to statements. Record the largest win, largest loss and maximum drawdown with dates. Note open positions separately, marked to market, and exclude them from closed-trade statistics. Log strategy changes and archive the output with its date.

This produces a description of what happened under a stated policy. It does not recommend a win rate, a strategy, a payoff ratio or a risk limit, and it does not predict next month. Its value is consistency: the same definitions, applied repeatedly, make deterioration and improvement visible instead of arguable.