Skip to content
The FavAcademy Briefing

Systems · Free

What Your Trading Statistics Can and Cannot Prove

Read win rate, payoff, expectancy, and drawdown together, with clear costs, sample definitions, and uncertainty.

Published · Updated

Public summary

A 60% win rate can accompany a losing record. Work through an original twenty-trade example to see why payoff and costs matter, then learn how small samples, changed rules, and missing observations limit what a performance report can establish.

A 60 percent win rate appears alongside a net loss of 60 dollars and a sample-uncertainty graphic.
A 60 percent win rate appears alongside a net loss of 60 dollars and a sample-uncertainty graphic.
Image credit

A 60 percent win rate appears alongside a net loss of 60 dollars and a sample-uncertainty graphic. Original AI-generated educational illustration; diagrams are illustrative.

FavAcademy · Original commissioned AI illustration; public use authorized

Article

A trading report summarizes the observations included in it. Before interpreting any percentage, ask what was recorded, what was excluded, which rules were used, and whether costs are reflected. A precise-looking number can still describe an incomplete or unrepresentative sample.

Statistics are useful for organizing evidence. They cannot transform a short record into proof of a durable edge or a guarantee about future returns.

Define the sample

Record the period, product, account, strategy version, and inclusion rule. Distinguish live executions, demo executions, and historical simulations. Do not combine them without preserving those labels.

Include losing trades and permitted setups you skipped when those omissions affect what you are testing. Keep canceled or rejected orders separately if they matter to execution analysis. A report cannot reveal observations that were never recorded.

Specify whether “trade” means one order, one position, or a sequence of entries and exits expressing one idea. Counting partial exits as separate wins can make two reports appear different even when they describe the same activity.

Win rate does not include win size

Imagine a fictional twenty-trade sample containing twelve winners and eight losers. The win rate is twelve divided by twenty, or 60%.

Suppose the average winner is $15 and the average loser is $25 before separately counted costs. Gross winning outcomes total $180, while gross losing outcomes total $200. The sample has lost $20 despite winning more often than it lost.

If additional costs total $2 per completed trade and have not already been included, subtract $40. The net sample result is then minus $60, or minus $3 per trade. Do not subtract spread again if your fill-based outcomes already incorporate it.

Payoff and expectancy answer different questions

In that example, average winning size divided by average losing size is 15/25, or 0.6. State the notation explicitly because labels such as “risk-reward” are used in opposite orders by different platforms.

Sample expectancy can be calculated as the average net result. Using these assumptions, 0.60 times $15 minus 0.40 times $25 minus $2 equals minus $3 per trade.

That is a description of the sample. It is not a known value for the strategy’s future distribution. A planned target-to-stop ratio is also different from the ratio of actual average outcomes.

Drawdown adds the sequence

Now consider a different fictional equity path after all costs: $2,000, $2,100, $1,995, and $2,040, with no deposits or withdrawals. From the $2,100 peak to the $1,995 trough, drawdown is $105, or 5% of the peak.

The account finishes above its initial value, yet it experienced a loss from its high. Final return alone does not describe that path. A different order of outcomes can change drawdown even when the total profit is identical.

State whether your drawdown uses closed-trade balance, marked equity, intraday readings, or end-of-day observations. Sparse observations can miss a deeper interim decline.

Respect uncertainty and changing conditions

Twelve wins in twenty trades do not establish a true 60% future win probability. NIST’s material on proportion confidence limits explains that an observed success proportion has statistical uncertainty under the relevant model assumptions.

Trading observations can also be dependent, and conditions may change. Several positions exposed to the same event are not necessarily twenty independent experiments. A confidence calculation based on an unsuitable model can look rigorous while answering the wrong question.

There is no magic sample size that makes a strategy proven. More well-recorded observations help, but do not repair biased selection, unrealistic costs, or repeatedly changed rules.

Use a review checklist

Before trusting a report, verify: complete inclusion; consistent definitions; reconciled costs; retained strategy versions; deposit and withdrawal treatment; realistic simulation assumptions; and the period covered.

Then ask which question remains unanswered. The report might show an execution problem or a cost-sensitive result. It might simply be too limited to distinguish competing explanations.

Keep conclusions proportional to the evidence. “This sample lost $3 per trade under these assumptions” is clear. “The strategy can never work” or “the next trade is due to win” goes beyond what that arithmetic establishes.

Risk disclosure

For education only. Historical samples and hypothetical examples do not establish future returns; trading involves risk.