Durable reference map
A three-stage method to reuse whenever conditions change
Keep the sequence of inputs, calculation and exception testing stable instead of relying on a market forecast.
- Align inputs and unitsGo to equations and definitionsObserved win rate・Expectancy per trade・Maximum drawdown
- Reconcile the worked exampleGo to table and calculation stepsCompare eight wins and two losses under two sequences
- Test exceptions and next checksGo to rules and counterexampleMeasure win rate, average win and average loss over the same sample.
Three facts missing from a win percentage
An 80% figure does not reveal whether each win is USD 100 or USD 10, whether each loss is USD 100 or USD 1,000, or where the two losses occur in the sequence.
A small number of large losses can erase many small gains. Aggregate every trade after execution costs in account currency or a common R unit.
- Retain both winning and losing trade counts.
- Review maximum and percentile losses as well as the average.
- Include commissions and slippage at trade level.
Expectancy and the equity path answer different questions
Expectancy summarizes the average result per trade for an assumed distribution. Reordering a fixed set of outcomes does not change that average.
Maximum drawdown depends on the path. Losses after an accumulated peak create a deeper peak-to-trough decline than the same fixed outcomes in another order.
Replay the same 80% under two orders
The example is hypothetical and exists only to verify the arithmetic. Eight wins are USD 100 each, two losses are USD 500 each, and costs are set to zero for clarity.
Both sequences finish at minus USD 200 with expectancy of minus USD 20 per trade. Maximum drawdown is USD 1,000 when losses follow the eight wins and USD 600 in the separated sequence.
Separate observed outcomes from confidence and planned reward-to-risk
Article 4 tests whether a subjective probability recorded before the trade is calibrated. This article starts from realized win and loss counts and their payoff distribution.
Article 6 examines planned target distance relative to stop distance. Average realized win and loss in this article are outcomes; a planned 3R target must not be substituted for a realized 3R average win.
Use win rate for monitoring, not automatic leverage
Calculate quantity from the stop loss and account budget. Monitor win rate beside expectancy, sample count, maximum drawdown and losing-run length. If the distribution changes, revalidate the premise before changing size.
- Show rolling and full-history windows together.
- Do not delete a large loss merely because it is inconvenient.
- Normalize to equal size or a common R before comparison.
Calculation framework
Observed win rate
Read the role of each equation first, then follow the numerical example to check the decision path.
Observed win rate
p = N_win / N_total- p: observed win rate
- N_win: profitable trades
- N_total: all evaluated trades
In plain language: It is a count ratio and contains no payoff magnitude.
When this conclusion does not apply: Calculate only when N_total > 0 and 0 ≤ N_win ≤ N_total, after defining treatment of cancellations, flat results and partial exits.
Expectancy per trade
E = p × W − (1 − p) × L- W: average net win after costs
- L: absolute average net loss after costs
- p: observed win rate
In plain language: A high win rate can still produce negative expectancy when loss magnitude is large.
When this conclusion does not apply: Calculate only with 0 ≤ p ≤ 1 and non-negative W and L measured on one currency and cost basis. It is an in-sample average and does not guarantee the future distribution or independence.
Maximum drawdown
MDD = max_t(Peak_t − Equity_t)- Peak_t: max_{0≤s≤t} Equity_s, including the zero starting baseline Equity_0 = 0
- Equity_t: cumulative P&L at time t
- MDD: largest peak-to-trough decline
In plain language: The value changes with outcome order even when counts and payoffs are fixed.
When this conclusion does not apply: This example starts cumulative P&L at zero and excludes cash flows.
Ask what the win percentage leaves out
Win rate counts successful outcomes and divides by the number of defined trades. It says nothing about the size of a typical gain, the size of a typical loss, costs, or the order in which results arrived. An 80-percent label can therefore coexist with positive, zero, or negative net performance depending on the omitted payoff distribution.
The denominator needs a rule for flat results, partial exits, canceled orders, and multiple fills. Changing that rule after seeing outcomes can raise the displayed rate without improving the economic record. A defensible report retains every eligible event and shows how one trade is defined before presenting the percentage.
Quantity adds another layer. A high recent win rate may have occurred at small size, while a larger next order changes the money attached to a future loss. Multiplying lots by the percentage assumes persistence and affordability that the count statistic does not establish. The next position must still begin with its own stop and authorized loss budget.
Combine frequency with the magnitude of wins and losses
A simple expectancy calculation multiplies win probability by average win and subtracts loss probability times average loss. All amounts need one currency and a consistent cost basis. Mixing gross winners with net losers, or using planned targets for wins and realized values for losses, creates an average that does not describe any actual sample.
In the hypothetical set, eight wins of USD 100 total USD 800, while two losses of USD 500 total USD 1,000. Net P&L is negative USD 200 across ten trades, so the sample expectancy is negative USD 20 per trade. The 80-percent win rate is accurate as a count and still economically incomplete.
Expectancy is an average, not a guarantee or a time path. It does not show how results cluster, whether the sample is stable, or whether the next loss can exceed the historical average. The calculation is most useful as a consistency check that forces the win percentage to share the page with the amounts it omitted.
Preserve realized order before calculating drawdown
Maximum drawdown depends on the cumulative path from an earlier peak to a later trough. Reordering the same eight gains and two losses leaves win rate, average win, average loss, expectancy, and final net result unchanged, yet it can change the funding strain experienced along the way. Trade order is therefore data, not presentation formatting.
When all eight USD 100 wins arrive first, cumulative P&L reaches a USD 800 peak before two USD 500 losses take it to negative USD 200. The peak-to-trough decline is USD 1,000. In the separated sequence, the path’s peak remains the initial zero and the deepest trough is negative USD 600, so maximum drawdown is USD 600.
A report that shuffles outcomes or shows only an average can hide this difference. The USD 400 drawdown gap does not make either invented sequence a forecast; it demonstrates that identical counts and payoffs can impose different interim capital demands. That path dependence is one reason recent winning runs should not trigger automatic leverage.
Define a trade so partial exits do not manufacture extra wins
A parent position can generate several executions and partial exits. Counting each profitable exit as a win while treating the final losing remainder as one loss can inflate frequency. The unit of analysis should be specified, such as a completed parent decision, and all fills, costs, and residual P&L should reconcile to that unit before success is classified.
Flat outcomes also require a rule. Excluding them from the denominator can raise or lower the percentage depending on the sample. The report should disclose whether zero after costs is a flat, loss, or separate category. The choice is methodological; changing it after reviewing the result is a form of outcome-dependent selection.
Canceled orders generally are not realized trades, but excluding submitted orders from execution-quality analysis can still matter. Maintain separate populations for strategy outcomes and order handling rather than forcing every operational event into the win-rate denominator. Clear event identities prevent duplicate fills from becoming duplicate successes.
Use one cost basis from entry through aggregation
Spread, commission, financing, and conversion can turn a small gross win into a net flat or loss. Average win and average loss should therefore state whether they are gross or net and identify the account currency and conversion time. Comparing gross headline wins with net account losses systematically favors the winning side of the equation.
Costs can depend on quantity and holding time. Scaling the next lot after a strong count record may increase both the stop loss and transaction charges, so historical expectancy at one deployment is not automatically invariant to size. If execution quality changes with quantity, the payoff distribution can change even when the analytical strategy does not.
The example sets costs to zero solely to expose the relationship among counts, amounts, and ordering. That assumption must not be carried into a live estimate without evidence. A sensitivity review can add plausible stated cost scenarios, but it should label them as assumptions rather than observations from the hypothetical ten-trade set.
Separate an observed win rate from an ex-ante confidence forecast
An observed 80 percent summarizes completed outcomes under a defined sample. An 80-percent confidence forecast is issued before one event and can later be calibrated across comparable cases. Substituting one for the other creates hindsight: the realized count is treated as though it had been known when the position was sized.
A short recent run is especially vulnerable. Eight wins in ten cases can arise in many processes, and the same ten observations are too few to establish a stable future rate. The page should display count and period and avoid translating the point estimate into certainty. A longer record can still change when market conditions or strategy implementation changes.
Confidence calibration also omits payoff magnitude, while win-rate expectancy adds magnitude but remains in-sample. Both can inform monitoring, yet the next loss budget belongs to a separate capital rule. This separation prevents a favorable historical summary from silently becoming permission to expose more account currency.
Do not substitute a planned target for the average realized win
A strategy can place a target at three times its stop distance and still realize much smaller average wins because of time exits, partial exits, missed fills, or costs. Inserting the planned 3R amount into expectancy while using realized losses makes the positive side hypothetical and the negative side observed. The resulting number is not a coherent sample statistic.
Record target attainment separately from the distance placed at entry. Realized average win should be reconstructed from executions in the same currency and event unit as average loss. The difference between planned and realized reward is itself useful diagnostic information, but it should not be erased to make a high win rate appear more profitable.
A farther target also does not authorize more lots. It changes the possible gain and may change attainment; the stop loss per lot remains tied to stop distance. If quantity is multiplied, both target gain and stop loss scale. The historical win percentage does not eliminate the extra amount exposed when the next outcome is a loss.
Inspect dependence, streaks, and concentration
The simple expectancy equation treats the sample as a collection of outcomes but does not establish independence. Losses can cluster when several trades share an instrument, market condition, or strategy factor. A high overall win rate can therefore coexist with episodes in which multiple positions fail together and consume more capacity than isolated averages imply.
Losing-run length and maximum drawdown preserve some path information. They still do not reveal every dependency or tail mechanism, so they should be reported beside, not in place of, trade-level records. A single large loss may be a data error, an execution gap, or a genuine strategy outcome; deleting it as an inconvenient outlier requires evidence, not aesthetics.
Concentration also matters across strategies. Two systems with attractive individual win rates can hold correlated exposures. A per-trade lot multiplier based on each recent success record can expand aggregate loss at the same time. Portfolio capacity and reservations must therefore remain separate constraints even when every component calculation passes alone.
Test whether the apparent improvement survives a frozen rule
A win-rate increase found in development data should be evaluated on later observations without changing filters, outcome definitions, or strategy parameters. Testing many variations and selecting the one with the highest percentage creates a winner’s curse. The number of trials and discarded candidates belongs in the evidence record.
The validation period should retain every eligible trade in realized order and apply the same cost treatment. If the strategy version or execution method changes, mark the break rather than blend records. A lower later win rate may reflect ordinary variation or a changed process; neither possibility is resolved by increasing size on the development result.
Comparing confidence intervals or resampling can describe uncertainty, but no statistical method turns a finite sample into a guaranteed future frequency. The decision consequence should remain proportional to the evidence: monitor whether payoff and path remain within expectations, while calculating each order from its current stop and active budget.
Attack the headline with adversarial reconstructions
One test keeps eight wins and two losses but changes USD amounts, showing that the same 80 percent can have positive or negative expectancy. The source counterexample uses USD 200 average wins and USD 100 average losses, which would produce USD 140 expectancy. That favorable arithmetic still does not require a larger loss budget.
A second test keeps all amounts and changes only the order, producing the USD 1,000 versus USD 600 maximum drawdowns. A third adds fees until some USD 100 gross wins become smaller net wins or flats. These perturbations reveal which claims depend on counts, magnitude, cost basis, and sequence rather than treating one percentage as a complete score.
A sizing test applies a proposed multiplier and recomputes loss at the next stop. The control should reject any lot whose checked account-currency amount exceeds the active budget, regardless of recent win rate. If the system instead raises the budget automatically, it has converted an outcome statistic into capital authority without testing tail or aggregate exposure.
Create a record that supports both payoff and path review
Each parent trade should retain decision time, strategy version, instrument, quantity, budget, stop, target, every fill, fees, conversion, partial exits, and final net account-currency result. The sequence key must survive export so cumulative P&L can be rebuilt. Summaries should be generated from these events rather than manually entered totals.
The report should show wins, losses, flats, denominator rule, average win, average loss, net expectancy, total P&L, maximum drawdown, and longest losing run over one stated period. If R units are used, each trade needs the loss budget recorded before entry; dividing by a realized loss after the event corrupts normalization.
Missing trades and corrections need versioned events. Replacing a result in place can alter both counts and path without leaving evidence. A reconciliation total should connect executions to the summarized parent set, and a data-quality flag should prevent a polished win-rate chart when fills, costs, or ordering do not balance.
Interpret warnings about hypothetical and consecutive losses
The CFTC warning supports caution that hypothetical results can omit actual market conditions, execution, spreads, fees, and the ability to withstand consecutive losses. It does not specify the two sequences or their dollar amounts. Those are invented here to make path dependence and payoff arithmetic transparent.
CME educational material supports beginning with a predefined account loss cap and deriving quantity from stop location and monetary value. It also discusses consecutive losses in the context of risk control. The evidence does not establish that the hypothetical 80-percent process will continue or that a particular loss budget is suitable.
The sources therefore support a bounded control conclusion: win percentage should not replace loss magnitude and sizing inputs. They do not prove that every high-win-rate strategy is unprofitable. The counterexample with smaller losses makes that boundary explicit while preserving the rule that favorable statistics do not automatically enlarge the next order.
Know when the historical percentage is not comparable
A change in instrument, session, strategy logic, holding period, or execution method can make the old population a poor baseline. Combining pre-change and post-change trades may yield a stable-looking percentage that describes neither regime. The record should mark breaks and state when no sufficiently comparable sample exists.
A quantity increase itself can break comparability if it changes fills, costs, or trader behavior. Using the smaller-size win rate to justify larger size and then assuming the payoff distribution is unchanged is circular. Capacity effects need direct observation at the relevant deployment, and early evidence should be labeled limited rather than extrapolated.
When comparability fails, the proper output is not a replacement universal rate. Report the data gap, retain the existing loss-budget control, and collect a new sample under the defined process. No historical headline is more informative than an explicit statement that its denominator no longer matches the decision being considered.
Translate the analysis into a bounded decision consequence
In the constructed ten trades, the correct summary is 80 percent wins, negative USD 20 expectancy per trade, negative USD 200 total result, and sequence-dependent maximum drawdown. Omitting any of those fields changes the reader’s understanding. None of them, alone or together, selects a suitable lot for a future setup.
The next quantity still comes from its price-distance loss per lot and the account-currency amount authorized before the trade. Historical statistics can trigger review of the strategy premise, costs, or capital policy, but they should not act as a hidden multiplier. Any budget change requires separate reasoning and should be visible in the calculation record.
This educational reconstruction cannot promise future expectancy or drawdown. It shows why a count ratio is not sufficient evidence for oversizing and how to preserve the omitted dimensions. The goal is not to condemn a high win rate; it is to stop that appealing number from concealing payoff magnitude, path, and money at risk.
Decision and control rules
- Measure win rate, average win and average loss over the same sample.
- Calculate drawdown and losing runs in realized order.
- Compare after costs in account currency or a common R unit.
- Do not make win-rate improvement an automatic lot multiplier.
Common failure modes
- Displaying win rate while hiding maximum loss.
- Deleting a small number of large losses as inconvenient outliers.
- Looking only at average results after shuffling and ignoring the funding path.
Evidence and specifications
- CFTC — Commodity Trading Systems Sold on the Internet
What this source supports: The CFTC warns that hypothetical results may not reflect the ability to absorb consecutive losses and may omit market conditions, spreads, execution and fees.
- CME Group — The 2% Rule
What this source supports: The exchange lesson begins with a predefined account loss cap and discusses consecutive losses, supporting size control from loss magnitude rather than win percentage alone.
- CME Group — Proper Position Size
What this source supports: CME derives quantity from stop location, account risk and tick value, providing a sizing basis that does not use win rate as a direct multiplier.
Questions to resolve
Is an 80% win rate a good strategy?
Win rate alone cannot answer that. You need payoff magnitude, costs, maximum loss, sample size and the equity path.
Can size rise when expectancy is positive?
Expectancy estimates an average. A larger loss allowance needs separate evidence about tail loss and account capacity.
Does order change the final result?
Not for the same fixed-dollar outcomes. It changes the path, and real sizing, stop-out or compounding rules can then make path consequential.
Which win-rate window should be used?
Show the full record and a predeclared rolling window, accounting for strategy changes. Do not choose a favorable interval after seeing results.
Recalculate from current inputs
Review win rate with payoff size and outcome order, then calculate quantity from a separate account loss budget.
Important: This is educational material about outcome statistics. It does not recommend a strategy, win rate, budget or quantity. Hypothetical outcomes are not performance, and historical expectancy does not guarantee future results.