Crypto trading expectancy is the average net outcome per completed trade under a stated counting and cost convention. Win rate records only how often an outcome is positive. A strategy can win frequently and still lose money when its losses or costs outweigh its gains. Calculate both from a reconciled trade ledger, then inspect sampling uncertainty, open positions and the capital required to produce those outcomes.
“Eighty percent winners” sounds precise. Without the size of those winners, the size of the losses and the treatment of unfinished trades, it does not tell you whether the account gained money.
A high win rate with a negative average
The following two samples are original hypothetical constructions. They are not market observations, KLINEO results or proposed trading systems. Each has 20 completed round trips. Gross outcomes exclude costs, and each round trip has a chosen fixed total cost of $1. There are no open positions or external cash flows in this first illustration.
| Sample | Gross winning trades | Gross losing trades | Average gross win | Average gross loss magnitude | Total gross P&L | Total costs | Total net P&L |
|---|---|---|---|---|---|---|---|
| A | 16 | 4 | $6 | $30 | −$24 | $20 | −$44 |
| B | 9 | 11 | $18 | $10 | +$52 | $20 | +$32 |
Sample A wins 80% of the time but averages −44 / 20 = −$2.20 net per trade. Sample B wins 45% of the time but averages 32 / 20 = +$1.60 net per trade. Costs leave the signs of individual outcomes unchanged in these chosen samples: A's net winners are $5 and net losses are $31; B's are $17 and $11.
For A, the gross arithmetic is 16 × 6 − 4 × 30 = −24. For B, it is 9 × 18 − 11 × 10 = 52. Subtracting 20 one-dollar costs produces the reported net totals. These calculations are identities for these invented observations, not forecasts of the next trade.
CME's educational discussion of mathematical expectation likewise combines win frequency with average winning and losing amounts. It concerns trading mathematics; using the identity does not establish that historical frequencies and payoffs will remain stable. CME's mathematical-expectation lesson
Use one cost convention throughout the calculation
For a sample with no breakeven trades, net expectancy can be calculated as win fraction × average net win − loss fraction × average net loss magnitude. If the win and loss amounts are gross, subtract average total trade cost separately. Do not use net amounts and then subtract the same costs again.
If costs vary by trade, calculate them in the ledger rather than assuming one fixed dollar charge. Fees may depend on filled notional, while slippage is reflected in the difference between the chosen reference price and the actual execution price. Funding and borrowing, where applicable, follow their own exposure and timing rules.
An execution price already adjusted for slippage should not receive a second cash deduction for the identical price effect. The backtest evidence and cost reconstruction works through this distinction with explicit buy and sell prices.
Zero-P&L trades also need a stated rule. If ten positions gain, five lose and five finish at exactly zero after costs, the win fraction among all 20 is 50%, not the 66.67% obtained by excluding breakeven positions. Either statistic can be reported if labeled, but their denominators are different. The sample mean remains total net P&L divided by all completed trades under the chosen trade definition.
Retain rejected orders and unfilled signals in a separate operational count. They are not completed trades, yet excluding them from the execution review can hide why an attractive theoretical signal rarely became a position.
Derive the break-even rate instead of guessing it
Under a deliberately simplified model, assume every gross winner is W dollars, every gross loser loses L dollars and every completed round trip costs c dollars. Let p be the probability of a gross win. The modeled average net outcome is pW − (1 − p)L − c. Setting that expression to zero gives p = (L + c) / (W + L).
For sample A's chosen payoffs, the modeled break-even rate is (30 + 1) / (6 + 30) = 31 / 36, approximately 86.11%. Without the chosen costs, it would be 30 / 36, approximately 83.33%. The difference comes entirely from the stated cost assumption.
This threshold is not an estimate that A's true win probability is 86.11%. It is the frequency required for a model with those fixed payoffs and costs to break even. Real strategies can change payoff distributions with volatility, order size and exit behavior. If W, L and c vary together, a single fixed threshold may describe the sample poorly.
When costs exceed a small gross winner, that trade changes classification after costs. Recompute the win and loss groups from the net ledger. Copying a gross win percentage beside net average payoffs can produce an internally inconsistent report.
Define what counts as one trade
Partial exits, position additions and multiple entry fills create several reasonable reporting units. One fill, one completed position and one strategy campaign are different units. A ladder that takes five profitable partial exits and one large final loss can appear to have an attractive win rate when each exit is counted separately.
Choose the grouping rule before comparing strategies. Retain parent position identifiers and link every fill and cost to them. If a position remains open at the end of the report, keep its marked value in portfolio equity even if it is excluded from completed-trade expectancy.
An unresolved losing position is particularly important. Closed-trade win rate can rise simply because losing positions are not closed. The drawdown and recovery guide explains why marked equity reveals a path that a closed-profit total may omit.
Dollars per trade also do not measure capital efficiency. A strategy using $20,000 for a $2 average outcome differs from one using $200 for the same outcome. Report exposure, holding time and available capital alongside the trade statistic. Overlapping positions prevent a naive assumption that the entire account can fund every observed opportunity independently.
Normalize risk without making losses look capped
An R multiple expresses an outcome relative to a declared initial risk amount. If sample A assigns a hypothetical $20 reference risk per trade, its net average of −$2.20 equals −0.11R. This is a scale conversion, not evidence that each loss was limited to $20. Its four gross losses are already $30 each under the example's assumptions.
Record what the reference risk means: planned price distance, position quantity, included costs and whether a stop was assumed. Stops can fill differently from their trigger or fail to complete under some conditions. Calling an outcome “one R” cannot guarantee an actual maximum account loss.
Compare net dollar expectancy with R expectancy when trade sizes vary. The average of individual R multiples is not generally the same as total dollars divided by the sum of planned risk. Each weights observations differently. State which question the chosen summary answers.
A Sharpe or Sortino ratio answers another question using periodic returns. Do not substitute trade-level expectancy into a daily-return ratio or compare ratios computed on incompatible sampling intervals.
Twenty outcomes are still only twenty outcomes
The sample mean and observed win fraction are descriptive results. Inferring an enduring expectancy requires a defensible sampling model and evidence about selection, dependence and changing conditions. Trades sharing a market shock or overlapping exposure need not behave as independent trials.
NIST documents confidence intervals for binomial proportions and warns that common symmetric approximations can perform poorly with small samples or few failures. Such intervals address a specified binomial model. They do not, by themselves, estimate the distribution of dollar payoffs or establish that a changing trading process meets that model. NIST exact binomial confidence limits
Selection adds another issue. If sample B was the best of hundreds of settings, its positive sample average is not an isolated prospective result. Preserve the trial history and evaluation boundaries before treating a favorable number as evidence of an edge.
There is no universal number of trades that validates every strategy. Twenty varied independent observations, twenty overlapping exits from one position and twenty trades selected from a favorable week contain different information. Report the actual concentration and dependencies instead of promising a magic threshold.
Turn the metric into an audit action
Request a ledger with grouping identifiers, gross outcomes, all modeled costs, net outcomes, position sizes, holding times and open-position marks. Recompute the sample mean directly from the net column and reconcile it with the account-level result. Investigate any difference before interpreting the headline.
For a forward check, compare the frozen rule's intended decisions with a supported paper-trading validation. Use that observation to inspect logic and assumptions; simulated fills remain a separate limitation. A positive mean can support further research but does not make a trade suitable or ensure a future profit.
Method and limits
This article was prepared with AI assistance. The two 20-trade samples, break-even calculation and reporting worksheet are original hypothetical illustrations, checked independently by arithmetic. The cited primary education and statistical documentation support narrowly identified concepts; no historical KLINEO results, live trades or confidence intervals were computed. Trading can lose money despite a favorable historical average, as described in the risk disclosure.