A hypothetical journal contains 117 trades and a roughly 62% win rate, while its average loss is twice its average win. Those summary figures can coexist with a negative result: with 62% wins and a two-to-one loss size, the simplified expected result is 0.62−0.38×2=−0.14 average-win units per trade. A specific account-percentage loss requires the actual trade sizes and equity path.

Why win rate alone proves nothing

Win rate alone omits payout sizes and costs. In simplified hypothetical models, 80% wins of one unit against 20% losses of five units give −0.2 unit per trade; 35% wins of four against 65% losses of one give +0.75 unit. These assumed probabilities are not estimates of a particular strategy.

For a binary payoff model, expectancy is win probability times mean win minus loss probability times mean absolute loss. Include zero outcomes and costs consistently when estimating it from a sample. Mean net R divides each trade result by its defined initial planned risk before averaging. Positive historical expectancy, 0.3R or 0.5R is not a universal future-profitability threshold. See strategy expectancy.

Aggregate metrics do not reveal loss order

Take eight results: four +1 and four −1 units. Two orders share win rate, profit factor and mean, but their longest losing runs differ. Lines show cumulative sums; colours show each step's sign.

Scroll sideways to see the full diagram

Two orders with identical aggregate metricsEight fictional net results use the same unit. A alternates +1, −1, giving eight sign runs and at most one consecutive loss. B has four +1 then four −1, giving two sign runs and four consecutive losses. Both have 50% wins, profit factor 1 and mean 0.SAME TOTALS, DIFFERENT ORDERA: alternating +1 and −1+1−1+1−1+1−1+1−1B: four +1, then four −1+1+1+1+1−1−1−1−1Both: win rate 50%, PF = 1, mean = 0.A: 8 sign runs, longest loss run 1.B: 2 sign runs, longest loss run 4.

Our fictional list has eight net results in a common unit, fixed model conditions and no zero outcomes. Paths add results without compounded returns. This is not a statistical test or a prediction from a small sample. NIST — Runs Test defines consecutive runs in a two-category sequence; here we compare only the signs in our own list.

A sign run is an uninterrupted block with the same sign: A has eight one-result blocks, B two four-result blocks.

Retain chronology and transaction IDs alongside aggregate metrics. Run count and longest losing run describe another property of the same sample. Matching win rate and profit factor do not prove matching paths or dependence.

The four metrics you must read together

Read win rate, mean result, profit factor and maximum drawdown together, with their definitions and the underlying records. They omit other information, including open exposure, cash flows, costs not included in the result, and concentration in a few trades. An aggregate report is a starting point for investigation rather than a complete test of the account.

Profit factor is the sum of positive trade results divided by the absolute sum of negative results under the same cost convention. It is undefined when the loss sum is zero. A value of 1 means the included gains equal included losses; values of 1.5 or 3 do not establish out-of-sample validity or prove overfitting.

For a stated balance or equity path, maximum percentage drawdown measures the largest fall from a prior peak to a later trough. A hypothetical €12,000 peak followed by €8,400 is 30% down; recovery requires €3,600/€8,400≈42.86%. Separate cash flows from trading performance. Pause and sizing limits need an explicit policy; 25% and 40% are not universal rules. See maximum drawdown and risk management.

Average win, average loss and the high win rate trap

Compare average net winners and absolute losers with the actual exit policy. A planned target-to-stop ratio of two does not force the realised average ratio to equal two: partial exits, differing risk sizes, costs and slippage can change it. Investigate an illustrative €110 mean win and €180 mean loss using the records rather than assuming a particular behavioural cause.

A high win rate can coexist with negative net results when losses are large. The average ratio does not itself establish when losses cluster or whether a single week offsets gains; inspect the chronology as well.

Include zero outcomes explicitly

Our hypothetical ten net results are +200, −100, 0, +200, −100, −100, +200, −100, +200 and −100 PLN, each with PLN 100 initial risk. Four winners give PLN 800, five losers cost PLN 500, and one trade is zero. Including the zero in the count gives a 40% win rate, PLN 300 total, PLN 30 mean (0.30R), and a net profit factor of 800/500 = 1.60. Excluding the zero changes the reported win rate to 4/9 ≈ 44.44%; state which convention is used. The mean is 0.40 × 200 − 0.50 × 100 + 0.10 × 0 = PLN 30. Do not replace the 50% loss share with 1 minus win rate when zero outcomes are present.

Variable cash risk can reverse the sign of the monetary result relative to mean R. A hypothetical +2R at PLN 10 risk earns PLN 20, while −1R at PLN 100 risk loses PLN 100. Mean R is +0.5R, but the combined cash result is −PLN 80. Keep the initial risk denominator even after moving a stop, and report missing risk data rather than dividing by a new zero.

Exposure and an illustrative example

Define time exposure as the share of the specified observation interval with qualifying open positions. That measure does not capture position size or concentration, and lower time exposure does not automatically make matching returns stronger evidence of skill.

In a hypothetical binary model, 55% wins averaging €165 and 45% losses averaging €145 give €25.50 expectancy per trade. In a separate hypothetical segmentation, 40% of entries have a 31% win rate and the other 60% a 67% win rate. The combined rate is 0.40×31%+0.60×67%=52.6%, with €18.06 expectancy if the same €165/€145 payouts apply to both segments. The latter cannot simultaneously be the 55% model. A six-month or 132-trade study still needs actual counts, costs and uncertainty before a changed rule is evaluated on untouched data.

How to keep a journal that can be segmented at all

Define the unit being counted: one trading idea with partial exits can produce several execution records. Retain the fills and their shared idea identifier, initial risk, net result, dates, instrument, direction, setup and adherence to rules. Distinguish closed results from open positions and transfers from profit. These fields support reconciliation; an arbitrary count of a dozen fields does not ensure completeness. See the trading journal and the journal template.

Keep every relevant trade and reconcile the journal with the account history. Review candidate segments with their sample sizes and data completeness. Selecting the worst historical group can fit noise; evaluate a changed rule prospectively rather than assuming that removing one category solves losses in eight out of ten cases.

Check concentration without deleting the original sample

In a separate hypothetical journal, nine net losses of 1R and one net gain of 20R give +11R, a +1.1R mean and profit factor 20/9 ≈ 2.22. Omitting the largest winner gives −9R, a −1R mean and profit factor 0. This is a diagnostic contrast, not a reason by itself to delete a valid trade or reject a strategy whose rules include rare large gains. Retain the complete sample and label every omission. Repeating the check for each observation produces related views of one history, not new independent samples or an estimate of future rare-event frequency.

What to do tomorrow

  1. Reconcile the last hundred trades if available and calculate win rate, mean net winner, mean absolute loser and mean net result/R under defined conventions. Neither a negative sample estimate nor 0.1R alone proves the true future expectancy; investigate the evidence and exposure before the next decision.
  2. Calculate drawdown on a stated path and compare it with a predefined exposure policy. Halving size after 25% or moving to demo after 40% are possible policy examples, not universal safeguards. Future observations do not guarantee recovery.
  3. Segment by predeclared weekday, time and setup definitions. Record counts and uncertainty, preserve the initial analysis and test any exclusion on later untouched observations. A six-week exclusion does not by itself isolate an improvement.
  4. Schedule a review that first reconciles records and costs, then examines risk, open exposure, drawdown and rule deviations. Record the data cut-off and strategy version, preserve the original history and choose a stated hypothesis for further testing. Review duration is a planning choice, not a guarantee of progress.