Read the results honestly

Performance measures describe a particular record. Before calculating them, identify the strategy version, dates, product, account currency, position sizes, execution model, costs and outcome definition. A percentage without this context can make two very different studies look comparable. Historical simulation, demo and actual account records must stay distinctly labelled.

Use one net result per completed case under a consistent allocation rule. If a position has partial exits, define how its costs and cash results are combined into that case. Counting each partial exit as a separate win while leaving the rest open changes the denominator. Open positions, pending orders, missing outcomes and skipped candidates need separate entries.

In this lesson the central example is a British learner’s ten invented GBP/USD net outcomes: nine gains of £1 and one loss of £12, after all stated costs. The figures are teaching ledger results, not current trades or a recommended method. The specified order matters when we later calculate drawdown.

A win is a completed case with a positive net result under the stated cost convention; a loss has a negative net result; a net zero is neither. A small positive price result can become a net loss after charges. Write whether the record is price-only, after selected costs or after all modelled costs rather than using “net” ambiguously.

The loss rate uses the same denominator. If there are zeros, win rate plus loss rate is less than 100%; the remaining share is the zero rate. Do not assume loss rate = 1 − win rate unless there are no zero outcomes under the definition. Unresolved results are not zeros and need a disclosed policy.

Average win is total positive net amounts divided by the number of winning cases. Average loss here means the positive magnitude of total negative net amounts divided by the number of losing cases. State the sign convention so the expectancy formula does not subtract a negative loss twice.

A mean can hide variation and outliers. Show the distribution or at least the individual cases when the sample is small. If there are no winners, average win is undefined, not an observed zero-sized win. If there are no losers, average loss is also undefined. The full-sample arithmetic mean can still be calculated from the ledger without inventing the absent category.

When sizes vary, cash averages include that sizing policy. A larger winning position can dominate the cash record. Keep sizes and percentage/R measures alongside cash if useful, with each definition explicit. Do not quietly normalise losses to one size while leaving gains at their larger actual sizes.

The arithmetic mean result of completed cases is their total net result divided by the case count. A common sample expectancy expression gives the same mean when rates and conditional averages use one consistent ledger. It describes what the sample produced; it is not a guaranteed expected return of future trades.

In the British example, 0.9 × £1 − 0.1 × £12 = −£0.30. That equals −£3 divided by ten. A high win fraction does not rescue the negative mean because the loss magnitude matters. Future rates and amounts are unknown and may change with costs, conditions and execution.

An estimated future expectation needs more than this formula: relevant evidence, uncertainty assessment and justified assumptions. Avoid naming a tiny sample’s win fraction “the probability of winning” as if it were established for every later trade. This beginner calculation is deliberately a sample description.

Profit factor divides the total positive results by the magnitude of total negative results in the chosen ledger. In this lesson those are net per-case results after stated costs. Some reports use price-only or other conventions; name yours so a reader knows which results were classified and summed.

For nine £1 gains and one £12 loss, profit factor is £9/£12 = 0.75. In a record with £15 positive and £10 negative totals, it is 1.5: £1.50 of positive results for each £1 of negative results in that sample. Neither ratio says how the gains were distributed in time or guarantees the next sample.

Software may display infinity or a blank for a zero denominator. Explain the convention instead of advertising an infinite edge. For a sample with no losers, the missing category is uncertainty, not proof that losses cannot occur. A high ratio can also come from one exceptional gain or a small selected sample.

The counts are nine wins and one loss. Positive total is £9; loss magnitude is £12; net total is −£3. Win rate is 90%; average win is £1; average loss magnitude is £12; sample mean is −£0.30; profit factor is 0.75. These results all describe the same ten-case ledger after its stated costs.

Do not subtract the same costs again from these already-after-cost outcomes. The later calculator exercise uses a separate price-only scenario plus a separately specified fee. Different conventions can be valid, but mixing them produces incorrect totals.

Ten invented cases are not reliable evidence of future probabilities. The example teaches why a headline win rate can hide a loss. An attractive chart of the first nine cases would omit the final loss and misrepresent the population. Keep all ten and their order visible.

There are two wins, two losses and two zeros. Total is 6 − 4 = 2 units across six cases, so mean is 2/6 ≈ 0.333 unit. Win and loss fractions are each 2/6, and the zero fraction is 2/6. Weighted sample expectancy is (2/6 × 3) − (2/6 × 2) = 1/3. Profit factor is 6/4 = 1.5.

Using 1 − win fraction as the loss fraction would wrongly treat the zero cases as losses. Removing zeros would create a different denominator and give a different per-case mean. State any rounding rule used to classify very small results; do not round losing entries to zero merely because the amounts are inconvenient.

For a real or demo multi-currency record, preserve quote-currency results and the account-conversion convention, rate and timing. Financing and currency conversion can change whether a case is a net win. If some costs are missing, label the reported scope and do not silently compare it with an all-cost ledger.

Balance generally reflects booked account cash results under the platform’s rules; equity includes the current valuation of open positions and relevant account adjustments. Check the product’s own definition. A closed-result ledger like the £100-to-£97 example does not reveal intratrade equity drawdowns or the lowest equity reached while a position was open.

To describe a path, specify observation frequency, valuation convention and whether costs/cash flows are included. Daily closing equity, tick-by-tick equity and equity only after closed trades can produce different measured maximum drawdowns. A coarser record can miss a deeper decline between observations.

Deposits and withdrawals can make naive balance changes look like performance. Keep cash flows separate, use an appropriate return/adjustment convention and disclose it. The simple examples assume no external cash flows, no open positions between ledger points and already-stated costs. Do not apply their percentages unchanged to an account with deposits or withdrawals.

Two records with identical final profit can have different paths, concentration and stress. Read the path beside the final result. A large later gain does not erase the exposure or practical difficulty faced during an earlier decline.

At each observation, keep the highest earlier-or-current account value, called the running peak. Drawdown amount is running peak minus current value. Percentage drawdown divides that amount by the same peak. Maximum drawdown is the largest decline observed under the chosen sampling and valuation convention.

In the British closed-result ledger, the running peak is £109 after case nine and the later trough is £97. The decline is £12, or 12/109 × 100 ≈ 11.01%. The final balance is also £3 below the original £100, but that −3% start-to-end change is a different measurement.

A fall from 100 to 80 is a 20% drawdown, but returning from 80 to 100 needs 25% because the recovery denominator is the smaller 80. For the £97 trough, recovering £12 to £109 needs 12/97 × 100 ≈ 12.37%. Neither calculation says the account will recover.

A cash mean includes the sizing policy used in the record. An R multiple divides a case’s result by its defined initial planned risk reference. If you use R, state whether the numerator is price-only or after costs, whether the denominator includes planned costs and what happens when the risk reference is missing or changes.

For a teaching case with a stored US$10 initial price-risk reference, +US$20 price result is +2R and −US$12.50 price result is −1.25R. If a separate US$0.50 fee applies, the net results become +US$19.50 and −US$13, giving +1.95R and −1.30R when the denominator remains the stored US$10 price-risk reference. Label both conventions.

Do not resize the denominator after seeing the loss to make it appear smaller. A changed stop or added position needs a recorded exposure change, while the initial reference can remain for a clearly named metric. An ordinary protective stop does not guarantee the actual result will be no worse than −1R.

Percent returns, R multiples and currency amounts are not interchangeable. A compound account return depends on the sequence and sizing/cash-flow conventions. Adding trade percentages is not generally the same as a compounded account return. Keep each unit’s calculation and purpose alongside it.

Ask whether the result depends on one exceptional case, one narrow period or one optimistic cost assumption. Report the full result first, then any predeclared sensitivity view with its definition. Removing the largest loss without a legitimate prior rule changes the sample; it is not a repair of the original performance.

Wider spreads or extra fees can change a small positive mean into a negative one. In the six-case net-units example, an additional hypothetical 0.50 unit charge per case would subtract 3 units from the total 2, leaving −1 overall and −1/6 ≈ −0.167 per case. That additional-charge scenario is different from the already-stated-cost baseline.

Compare periods and relevant conditions under definitions chosen before evaluation where possible. If you search many subgroups and show only the profitable one, the subgroup has become another fitted selection. A rule intended for high volatility needs an available-data definition of high volatility, not a label applied after seeing which dates won.

Keep uncertainty, unresolved fills and missing costs visible. A stress scenario can reveal a weakness without being a guaranteed worst case. Historical maximum drawdown is an observed measurement, not a ceiling on future account loss.

Curve fitting or overfitting happens when choices fit historical quirks so closely that apparent performance may not carry to new observations. Many filters, parameter searches and repeated looks at an evaluation set increase opportunities to select noise. A revision can be useful development, but it must not be advertised as if it were decided before the evidence that inspired it.

Prefer clear reasons and a manageable set of assumptions. Keep all tried variants, development/evaluation boundaries and the chosen version. Nearby-parameter or cost checks can expose fragility when planned fairly, but picking the best of those checks is another selection. A strong-looking ratio cannot substitute for a fair research process.

An honest negative or uncertain result is useful. You may decide to stop, simplify the rule, improve data or begin a new development version. Do not keep relabelling the same later sample untouched. Neither a high win rate, positive sample mean nor profit factor above one by itself establishes an applicable future edge.

A learner should be able to trace a headline figure back to the ledger and recompute it. Include the data/assumptions version and enough information to understand corrections. Protect personal account identifiers when sharing. A report can be compact while still being explicit about every denominator and unit.

Read process quality and performance separately. A well-followed rule can have a negative record, and a profitable breach can be unreliable learning. Do not conclude that more discipline alone will make a weak strategy profitable. A quiz pass checks course understanding; it is not certification to trade live or a promise of income.

The next level examines changing relationships, liquidity, positioning and broader review. Those ideas add context without removing the testing requirements here. Keep this ledger, specification and journal for later learning rather than replacing them with a single attractive screenshot.