Trading Performance Statistics Glossary: Expectancy, Sharpe, Sortino and Drawdown

Trading performance statistics describe different properties of a return or trade history. One measure rarely captures profitability, downside exposure, tail risk, implementation costs and the reliability of the evidence simultaneously.

This glossary explains commonly reported performance indicators, their calculations and the dangers of overinterpreting small samples. It separates descriptive realised results from claims about a strategy’s future edge or an individual’s investment suitability.

Why this specialist terminology matters

Market commentary often compresses a complex mechanism into one familiar expression. The right question is not only what a term means, but which contract, accounting basis, measurement date or operational framework applies. A number without this context can be precise and still misleading. The definitions below include both the basic concept and a caution about practical interpretation.

Trading Performance Statistics Glossary: essential terms

Net return

Change in account or portfolio value after the specified costs, cash flows and valuation conventions are incorporated. A performance record must state whether returns are gross or net of commissions, spread, borrowing and other charges.

Gross return

Performance before a defined set of fees or expenses are deducted. Gross return can appear attractive even if implementable net results are poor. Define which costs are excluded.

Arithmetic mean return

The simple average of a series of period returns. It does not incorporate compounding and can be materially higher than the geometric average for a volatile return series.

Geometric return

The compound growth rate consistent with successive period returns over the specified horizon. It reflects sequence and compounding. It differs from a simple average when returns fluctuate.

Annualisation

Converting period statistics into an annual scale under stated assumptions. Multiplying returns or scaling volatility using square-root-of-time rules can be misleading with autocorrelation, leverage changes or irregular samples.

Win rate

The proportion of closed trades classified as winners under the reporting definition. A high win rate does not guarantee profitability; loss size, position sizing and transaction costs also matter.

Average win

Mean positive realised result across winning trades under a specified unit of account. It should be compared with average loss after costs and adjusted for different position sizes.

Average loss

Mean absolute loss among losing trades under the chosen convention. Large rare losses can dominate returns despite frequent small winners. Excluded open positions may distort the estimate.

Expectancy

The average outcome estimated per trade: win probability × average win minus loss probability × average loss, under consistent units. It is a sample estimate, not a guarantee. Dependence between trades and regime changes can invalidate a stable-distribution assumption.

Profit factor

Gross sum of winning trade profits divided by absolute sum of losing trade losses under defined cost treatment. It omits the sequence of outcomes, drawdown behaviour and the uncertainty around a small number of trades.

Payoff ratio

Average winning outcome divided by average losing outcome, normally using absolute loss magnitude. It combines with win rate to determine simple expectancy but does not specify time, capital usage or tail losses.

Sharpe ratio

Mean excess return relative to a defined risk-free reference divided by total return standard deviation, with consistent periodicity. It treats upside and downside variation symmetrically and can be distorted by non-normal distributions and smoothed returns.

Sortino ratio

Mean return above a chosen minimum acceptable return divided by a downside deviation measure under the stated convention. Choice of threshold and estimation method matters. It should not be compared without those definitions.

Maximum drawdown

The largest observed peak-to-trough portfolio decline during the measurement window. It is historical and path-dependent, not the maximum possible future loss. Adding more history may expose deeper declines.

Drawdown duration

Time spent below an earlier portfolio high or the time from peak to recovery under a chosen definition. Two strategies with identical maximum drawdowns can impose very different waiting periods before capital recovers.

Calmar ratio

A ratio often computed as annualised return divided by the absolute maximum drawdown over a specified history. The denominator is an extreme sample observation and is unstable across different window lengths.

Volatility

Standard deviation of returns over the selected time series and frequency, commonly annualised. Volatility does not capture all loss risk, especially gaps, illiquid marks and extreme tails.

Value at Risk (VaR)

A model-based loss quantile over a defined confidence level and horizon under specified assumptions. VaR does not describe average loss beyond the quantile and may miss severe model errors.

Expected shortfall

A tail-risk measure describing the average loss in the worst specified portion of a modelled return distribution. Its usefulness depends on the loss model, scenario set, liquidity assumptions and data quality.

Information ratio

Mean active return relative to a benchmark divided by the standard deviation of active returns over matching periods. It compares benchmark-relative performance rather than absolute reward per unit of return volatility.

Sample size

The number of observations in a performance estimate, such as trades or independent return periods. A large nominal number of highly dependent intraday trades may contain far less independent information than the count suggests.

Look-ahead bias

Using data in a historical simulation that would not have been known when the decision supposedly occurred. It can create impossible backtest performance. Correct event timestamps and available-data versions are necessary.

Survivorship bias

Excluding securities or strategies that disappeared or failed before the end of the research period. It can make historical results look better than the real opportunity set. Delisted stocks and abandoned strategies may need inclusion.

Overfitting

Selecting a strategy or its parameters to match noise in a historical sample rather than a stable economic relationship. Repeated optimisation without genuinely unseen testing can generate impressive but fragile simulated results.

Worked example and interpretation

Consider 100 hypothetical completed trades: 45 winners averaging £150 each and 55 losers averaging £100 each, after the costs used in the record. Win rate is 45%; mean win-to-loss ratio is 1.5; estimated expectancy is 0.45 × £150 − 0.55 × £100 = £12.50 per trade. Gross positive outcomes total £6,750 and losses total £5,500, so profit factor is approximately 1.23. Neither figure establishes a durable edge: results may be dependent on a single market regime, a few unusually large trades or choices about excluded open positions. If a later loss is £2,000, average loss and expectancy can change materially. Performance should be examined alongside drawdowns, execution capacity and out-of-sample evidence.

Analytical extension: compare attractive ratios with adverse paths

Suppose a hypothetical strategy has annualised historical return of 10%, an annual risk-free benchmark of 2% and return volatility of 16% using a consistent measurement convention. Its simplified Sharpe ratio is (10% − 2%) ÷ 16% = 0.50. Another strategy with a different return distribution could report the same ratio yet experience deeper tail losses. A realised Sharpe ratio is a sample statistic whose uncertainty depends on the length and dependence structure of the data.

Drawdown arithmetic: If an account falls from £100,000 to £60,000, the peak-to-trough drawdown is 40%. Recovering to £100,000 requires a gain of £40,000 on £60,000, or approximately 66.67%, not 40%. Sequence therefore matters: two strategies with the same arithmetic average return can deliver different compound outcomes when losses and gains occur in different proportions. Financing costs and cash withdrawals can further alter the result.

Profit factor and expectancy are complementary. Profit factor compares total gains with total losses among the included trades; expectancy estimates the average gain or loss per trade. Neither on its own describes capital at risk, holding time, loss clustering or the ability to implement the strategy at scale. A handful of outlier winners can make profit factor unusually high, while removing abandoned or still-open losing positions can artificially improve the reported picture.

Statistical robustness matters more than decorative precision. Specify whether observations are independent, whether prices were available when the simulation assumed, whether transaction costs rise with size and whether the tested parameters were selected after viewing the sample. Hold-out periods, cross-validation where appropriate, and adverse parameter sensitivity can improve confidence but do not prove future profitability. Present realised performance as historical evidence with clearly described limitations, not as a promise.

Concept comparison

Applying the terms in a structured research process

First identify the exact exposure or decision being studied. Record the relevant market or institution, contract specification, valuation basis and observation date. Separate directly observed data from estimates and from judgements. If the research question concerns the future, state more than one possible scenario rather than implying that a past relationship must continue.

Second, translate the terminology into its economic consequences. Ask which party receives or pays cash, what changes if the quoted price moves, what financing or collateral may be required and which risk remains after any hedge or mitigation. Whenever figures are compared, match units, timing and direction before doing the arithmetic.

Third, explain what would make the conclusion unreliable. A market can become less liquid, an estimated coefficient can shift, operational rules can change and the available data can be incomplete. Checking these limitations is part of professional analysis, not a weakness in the explanation.

Common interpretation mistakes

• Ignoring costs, financing, taxes or open losses when presenting a performance history.
• Presenting annualised results from an unrepresentative short history as if they were a reliable forecast.
• Optimising strategy parameters on a dataset and then calling the same dataset an independent validation test.

Frequently asked questions

Is a strategy with a 70% win rate necessarily good?

No. If average losses greatly exceed average wins, net expectancy can be negative. Costs and tail events can worsen outcomes further.

Is a high Sharpe ratio proof of skill?

No. It is an estimated summary sensitive to the return history, frequency, leverage, pricing and risk-free benchmark. It should be combined with other evidence.

How much backtest data is enough?

There is no universal count. Independent market regimes, turnover, structural changes and the number of model decisions matter at least as much as the number of observations.

Related education

Explore Risk–Reward, Win Rate and Trading Expectancy; Trading Strategy Backtesting; Value at Risk vs Expected Shortfall, the Trading Glossary or the complete Education Hub.

Sources and further reading

• CFA Institute — Portfolio Performance Evaluation
• CFA Institute — Analysis of Active Portfolio Management

Important notice: General financial education only, not personalised advice or a recommendation to buy, sell or hold any financial instrument. All numerical scenarios are hypothetical. Investments and derivatives may incur significant losses and applicable product terms determine actual outcomes.