Trading expectancy: why win rate can mislead you

Trading expectancy estimates the average outcome per trade from the win rate, average win, average loss, and costs of a defined process. Win rate alone can mislead because frequent small wins can be outweighed by fewer large losses, while a lower win rate can remain positive when the average win is large enough.

The result is an estimate from a specific sample. It does not predict the next trade, prove that the rules have an edge, or promise that the same distribution will continue.

Start with the outcome distribution

Imagine two hypothetical strategies evaluated over 100 completed trades. The first wins 70 times. The second wins only 40 times.

The 70% strategy sounds stronger until the size of each outcome is included. If its average winner is +1R but its average loser is -3R, the 30 losses remove more than the 70 wins produce. If the 40% strategy averages +2R on winners and -1R on losses, its fewer wins can still produce a positive average before costs.

Here, R is a neutral risk unit. One R represents the amount the process planned to lose if the original invalidation was reached, before trading costs. It is not a recommended account risk. Gaps, slippage, and changed exits can also make the actual loss different from the planned 1R.

How trading expectancy is calculated

For a sample containing only winning and losing outcomes, the basic equation is:

expectancy = (win rate × average win)
           − (loss rate × average loss)
           − average cost per trade

Use the average loss as a positive magnitude in this form because the equation already subtracts it. Win rate and loss rate should be decimals that add to 1. Average win, average loss, and costs must use the same unit.

If breakeven trades are recorded as a separate category, include their probability and net outcome rather than forcing them into the win or loss group. A zero price result may still be negative after fees and slippage.

Expectancy is a probability-weighted average. A result of +0.15R per trade does not mean the next trade should return +0.15R, or that every ten trades should return +1.5R. Individual results arrive in an uneven sequence.

Worked example: 70% wins versus 40% wins

This comparison is hypothetical. It assumes each strategy's observed win rate and average outcome came from 100 trades. It then applies a uniform cost of 0.05R to every completed trade so the arithmetic can be reproduced.

If those averages described exactly 100 trades, the first sample would total -25R after the assumed costs. The second would total +15R. These totals restate the sample arithmetic. They are not forecasts for the next 100 trades.

The comparison shows why asking only “How often does it win?” leaves out too much. Outcome size and cost can reverse the conclusion suggested by win rate.

What expectancy tells you

Expectancy compresses three useful facts about a recorded process into one comparable unit:

1. How frequently the trades were profitable.
2. How large the average profitable and losing outcomes were.
3. How much the included costs reduced the average result.

Expressing outcomes in R can make samples with different prices or account sizes easier to compare. That comparison is valid only if R is defined consistently. If one trade uses planned initial risk, another uses actual maximum loss, and a third changes its invalidation after entry, the shared label hides different measurements.

The calculation can also expose where a result depends on one fragile input. A strategy with a thin gross expectancy may become negative after realistic costs. A strategy whose average win depends on one unusual outlier may look very different when that trade is removed or capped under live liquidity constraints.

What expectancy does not tell you

A positive historical expectancy does not establish that the next trade will win. It does not describe the order of wins and losses, the depth of a possible drawdown, or whether the position size is survivable. It also does not show that the estimate is statistically reliable.

The equation cannot repair weak inputs. It will produce a precise answer from ten trades, but precision in the arithmetic is not confidence in the estimate. It also cannot detect that entry rules, exits, eligible symbols, or market conditions changed halfway through the sample.

Most importantly, expectancy is conditional on a complete rule set. Combining trades from different setups and calling the total “my strategy” can create an average that describes none of the underlying processes.

Why sample size needs more than a magic number

There is no universal trade count that makes expectancy reliable. The evidence depends on the distribution and on how the sample was produced.

A small sample has wide uncertainty because a few outcomes can dominate both win rate and average outcome. A larger sample can still mislead if it contains many near-duplicate trades from one market period, excludes eligible losses, or follows rules chosen after inspecting the same history.

The number of strategy variations tried also matters. Research on backtest overfitting shows that repeatedly testing parameters against the same historical data raises the chance of selecting a configuration that captured quirks of that sample rather than a pattern that survives new data.

Instead of treating one count as sufficient, record:

• the number of trades and the period covered;
• the markets, symbols, sessions, and regimes represented;
• the exact rule version used;
• how many alternative rules or parameters were tested;
• whether results held on data that was not used to choose the rules;
• the distribution of outcomes, not only their average.

An estimate can remain uncertain after this review. Unknown should remain unknown.

Costs belong inside the equation

“Zero commission” does not mean zero cost. Depending on the instrument, broker, account, and order, results can include commissions, fees, spreads, slippage, and other charges. FINRA and the US Securities and Exchange Commission both advise investors to understand transaction and account costs because they reduce returns.

Use actual fills and charges where available. If a backtest uses an estimate, state the assumptions and test how sensitive the result is to less favourable execution. Do not calculate a gross expectancy, discover a small positive number, and leave costs for later.

In the worked example, changing the assumed average cost from 0.05R to 0.20R would move the second strategy from +0.20R gross to zero. The setup did not change. The economic conclusion did.

Common ways expectancy gets overstated

Win rate is reported without outcome size. The reader cannot tell whether the losses outweigh the wins.

The average hides the distribution. One extreme winner or loss can dominate the estimate. Inspect the individual outcomes and concentration.

Costs are missing or too tidy. A fixed estimate may not represent changing spreads, liquidity, order size, or slippage.

Different rules share one sample. Changed exits, discretionary filters, and multiple setups are pooled into an average that cannot be repeated.

The sample was selected after the result was known. Excluded losses, chosen dates, and repeated parameter searches can make historical expectancy look stronger than it is.

A market regime changes. Relationships observed in one volatility, liquidity, or trend environment may not persist in another. A larger old sample does not automatically describe current conditions.

Positive becomes permanent. Historical expectancy is evidence about the measured sample, not a licence to stop monitoring execution and results.

What to review next

Before relying on an expectancy estimate, ask:

1. Are the setup, entry, exit, and eligibility rules defined precisely?
2. Are all eligible trades included under the same rule version?
3. Are wins, losses, breakevens, and costs classified consistently?
4. Do average wins or losses depend on a small number of outliers?
5. Does the sample cover relevant conditions, rather than one favourable period?
6. Were the rules tested on data that did not help create them?
7. Is the estimate still positive under realistic cost and slippage assumptions?
8. What remains unknown about the sample or live execution?

That review does not confirm a strategy. It makes the evidence and its limitations easier to inspect.

Where HeraldGoat fits

HeraldGoat is pre-launch. It is being built to monitor named playbooks, follow each Setup through its changing state, and place first-pass context in an inspectable Goat Queue. More consistent monitoring can help preserve which trigger and surrounding facts were present for later review.

HeraldGoat does not calculate or guarantee strategy expectancy. It does not define exits, size positions, execute orders, or replace a trading journal. Better monitoring cannot repair negative expectancy, and a cleaner alert cannot turn an uncertain estimate into a promise.

Continue with the system around the number

• See why profitability belongs to the complete system
• Keep a passed checklist separate from outcome probability
• Audit whether every alert has a defined response
• Browse all practical trading guides

If inspectable first-pass context would help you monitor a defined playbook, you can join the launch waitlist. Joining does not imply immediate product access.

Sources

• FINRA: Fees and commissions: explains transaction costs, ongoing expenses, and why zero-commission trading does not make investing cost-free.
• SEC Investor.gov: How fees and expenses affect your investment portfolio: explains how transaction and ongoing fees reduce investment returns and where investors can inspect those charges.
• Bailey et al.: Backtest overfitting in financial markets: describes how testing many strategy variations on the same historical data can select models that fail on new data.

Method note: the expectancy equation and two-strategy comparison are educational models. The examples are hypothetical. Their risk units, costs, trade counts, and outcomes are not recommended settings or historical performance.

This guide is for educational information only. It is not investment advice, a recommendation, or a promise of trading results. Trading and investing involve risk, including the possible loss of capital.