The Backtest Metrics That Flatter a Strategy (and What to Read Instead)

Net profit and win rate flatter more than any other numbers on a backtest report. A big net-profit figure tells you the equity curve ended higher than it started; it says nothing about the path, the risk taken to get there, or how many trades were involved. A high win rate is worse, because it feels like skill while hiding the shape of the losses. A strategy can win 90% of its trades and still bleed to zero if the 10% it loses are large enough. These two figures are the ones sellers put in the headline precisely because they are the easiest to make look good.

The honest way to read a report is to treat net profit and win rate as context, not evidence. They tell you what happened; they do not tell you whether what happened was robust, repeatable, or distinguishable from luck. To get at that, you read a different set of numbers, and you read them together. No single metric survives on its own. And before any of them mean anything, you check how many trades the backtest contains, because a metric computed on 20 trades is a rumor, not a measurement. This piece is educational and makes no claim about future returns; the point is only to read the past honestly. For the deeper failure modes behind flattering numbers, see [why most backtests fail](/learn/why-most-backtests-fail).

Net profit is a single endpoint. Two strategies can post the same net profit while one rode a smooth climb and the other survived a 60% drawdown that would have triggered every stop-out and margin call along the way. The endpoint erases the path, and the path is where the risk lives. Compounding makes this worse: a late run of luck can inflate the final figure far beyond what the typical trade justifies.

Win rate misleads through a different mechanism. It ignores the size of wins and losses entirely. Consider two systems:

| Metric | System A | System B | |---|---|---| | Win rate | 80% | 40% | | Avg win | 1R | 3R | | Avg loss | 5R | 1R | | Expectancy per trade | -0.2R | +0.6R |

System A wins four times as often and loses money every trade on average. System B looks unimpressive and is the sounder of the two on this sample. Win rate pointed you at exactly the wrong one. This is why traders who lead with win rate are, knowingly or not, steering your eye away from the numbers that would contradict them. A high win rate is not a lie by itself; it is a lie by omission, and the omitted part is the payoff structure. None of this predicts what either system does next. It only says that on the trades observed, win rate ranked them backwards.

Read the metrics that account for the size and shape of outcomes, not just their direction. The core set:

• Expectancy — average profit or loss per trade, in currency or in R (risk units). This is the single most honest per-trade summary. Positive expectancy on a large sample is the minimum bar; it is not a promise.
• Profit factor — gross profit divided by gross loss. Above 1.0 means wins outweighed losses on this sample. Values far above 2.0 on a small sample usually mean too few trades, not a great system.
• Payoff ratio — average win divided by average loss. Read it alongside win rate; the two only mean something together.
• Maximum drawdown (%) — the deepest peak-to-trough fall. This is the number that tells you whether you could have survived the equity curve psychologically and financially.
• Recovery factor — net profit divided by max drawdown. It asks: how much return did you extract per unit of pain? A strategy that made 30% but suffered a 50% drawdown has a recovery factor below 1 and is fragile regardless of the headline.
• Sample size (trade count) — the gate on all of the above.

Expectancy times trade count reconstructs net profit, which is why net profit alone is redundant once you have the components. Read these together and a flattering report loses its shine fast. You can run this reading automatically with the free [Backtest Health Check](/backtest).

Every metric on a backtest is an estimate, and the uncertainty of an estimate shrinks slowly with sample size, roughly with the square root of the number of trades. Thirty trades is not a small version of a good sample; it is noise wearing the costume of a measurement. A profit factor of 3.0 over 25 trades and a profit factor of 1.4 over 1,200 trades are not comparable claims, and the second is the more trustworthy one despite the smaller number.

This is why ForexCodes gates metric readouts by trade count. Below a threshold, we will show you the numbers but flag them as under-powered rather than let them stand as conclusions. The honest response to a 40-trade backtest is not to compute its Sharpe to two decimals; it is to say the sample cannot support a confident statement, full stop. There is a related trap: even a large sample can flatter if the strategy was tuned on that same data, which inflates every in-sample metric at once. Marcos Lopez de Prado and David Bailey have written extensively on how selecting the best-looking configuration across many trials manufactures impressive backtests that carry no real edge, and their deflated Sharpe ratio adjusts a Sharpe reading for exactly this. For how many trades is enough, see [how many trades is enough](/learn/how-many-trades-is-enough); for the selection problem, see [overfitting and curve-fitting explained](/learn/overfitting-curve-fitting-explained) and the [deflated Sharpe ratio](/learn/deflated-sharpe-ratio).

Work in a fixed order so the flattering numbers cannot set the frame. First, check the trade count and decide whether the sample can support any conclusion at all. If it cannot, stop reading metrics and go get more data or a longer test window. Second, look at expectancy and profit factor together; a system that fails here fails, regardless of win rate. Third, read max drawdown and recovery factor to understand the cost of the returns. Only last, glance at net profit and win rate as sanity checks, never as evidence.

Then ask the question that no single report can answer: were these numbers produced on data the strategy had never seen? A metric measured only on the data used to build the strategy is describing memorized noise as much as any real pattern. The metrics in this article are diagnostic, not predictive. They can tell you a strategy is fragile or under-tested; they cannot tell you it will make money, and anyone who reads them that way has skipped the part that matters. This whole exercise is about integrity and robustness, never about a return you should expect. To pressure-test how much of a result depends on the single best-fit configuration, run the free [Overfitting Check](/backtest/overfitting), and read [in-sample vs out-of-sample](/learn/in-sample-vs-out-of-sample) before you trust any of these figures.