Win rate, profit factor, expectancy — the three numbers and which one to ignore

Open any backtest report or signal-service track record and you'll see the same three numbers: win rate, profit factor, expectancy. They sound like they're describing the same thing from different angles. They aren't. Two of them can be excellent while the strategy bleeds money. One of them, calculated honestly, tells you the actual answer to "should I size into this." Most retail traders look at the first, sometimes the second, and never properly understand the third.

The short version: win rate tells you how often the strategy is right but not how much that rightness pays. Profit factor tells you the ratio of total winnings to total losses but hides what each individual trade looked like. Expectancy tells you the dollar (or R) value of running the strategy one more time. Of the three, only expectancy survives the move from spreadsheet to live account. The other two are useful as supporting context — but if you have to ignore one of them entirely, it's never expectancy.

This post walks the math of each metric, the specific combinations that look great but lose money, why expectancy is the truth-teller, and the deeper metrics (Sharpe, Sortino, Calmar) that matter when you're sizing the strategy rather than just running it. The interactive at the end takes a list of trade outcomes — wins and losses you can paste from anywhere — and outputs all three numbers plus a verdict on whether the strategy is genuinely profitable, optically profitable, or just lucky.

Win rate — the most-cited and least useful metric

Win rate is the percentage of trades that closed positive. A strategy that takes 100 trades and wins 65 has a 65% win rate. The arithmetic is simple. The interpretation is where things go wrong.

What win rate doesn't tell you: how big the wins were, how big the losses were, or whether the strategy is profitable. A 90% win rate strategy that wins $1 nine times and loses $20 once is losing money — net result is a loss of $11 over those 10 trades. A 30% win rate strategy that wins $10 three times and loses $1 seven times is making $23 over the same period. Both can be described accurately as their win rates and both descriptions are useless without the other numbers.

Win rate's appeal is psychological, not mathematical. Frequent feedback that "I was right" feels good. The traders who optimise for it end up at one specific point on the stop-placement seesaw — wide stops, tight targets, lots of small wins, occasional large losses. That shape can be profitable, but it isn't profitable because of the high win rate. It's profitable in spite of the structural cost of small targets, which the edge-after-cost post covers in detail.

The single most common retail trading mistake is treating win rate as the headline result. It isn't. It's a feature of the equity curve's shape, not a measure of its quality.

Profit factor — better, still incomplete

Profit factor is the ratio of total dollars won to total dollars lost:

PF = Σ(winning trades) / |Σ(losing trades)|

A strategy that wins $5,000 across all winners and loses $2,500 across all losers has a profit factor of 2.0. A profit factor above 1.0 means the strategy made money over the period. Below 1.0 means it lost. Equal to 1.0 means break-even.

Profit factor is more honest than win rate because it can't hide the size of wins and losses — they're both in the math. A high-win-rate strategy with tiny wins and one big loss won't have a high profit factor, because that one big loss bloats the denominator. Conversely, a low-win-rate strategy with big winners can show a high profit factor even with most trades being losers.

The general rule of thumb in the literature: a long-run profit factor above 1.5 is reasonable, above 2.0 is very good, above 3.0 is suspicious until verified across enough trades that it can't be a small-sample artefact (more on this in the win-rate sample-size post).

Where profit factor still lies: it doesn't account for trade frequency. A strategy with profit factor 2.0 over 50 trades has produced very different stress on the equity curve than profit factor 2.0 over 5,000 trades. The first is a few large bets; the second is a high-frequency strategy with many overlapping commitments. The risk profiles are completely different even though the headline metric agrees.

The other place profit factor mis-signals: a strategy with one outlier winner can show a glamorous profit factor that, removed of that one trade, becomes break-even or worse. Most retail systems with profit factors above 4.0 over fewer than 200 trades are showing this artefact, not real edge.

Expectancy — the one that doesn't lie

Expectancy is the average dollar (or R-unit) outcome per trade across the full sample:

E = (Win rate × Avg win) − ((1 − Win rate) × Avg loss)

If a strategy has a 60% win rate, an average win of $100, and an average loss of $80, the expectancy per trade is:

0.6 × 100 − 0.4 × 80 = 60 − 32 = +$28 per trade

That number is the expected value of running the strategy one more time. It's the only metric of the three that directly answers the question "if I take 100 more trades from this system, what's my expected outcome?"

Multiply expectancy by trade frequency to get expected period return:

Expected annual return = E × (trades per year)

For the example above, at 200 trades per year: $28 × 200 = $5,600 expected. That's the number that survives the move from spreadsheet to live trading, before costs are subtracted (after which the edge-after-cost framework takes over).

The reason expectancy is the truth-teller: it directly encodes both the size and the frequency of wins and losses. Every other metric loses information by aggregating in a way that can hide the underlying distribution. A strategy with positive expectancy will, over a long enough sample, deliver close to E × N total return. A strategy with positive win rate or positive profit factor might deliver positive return — depending on the other variables.

The catch with expectancy: it's an expected value calculation, which assumes the future is drawn from the same distribution as the past. For real markets, this is approximately true within a regime and dramatically false across regime changes. Expectancy is the right metric to optimise within a strategy, but it gives you no protection against the market itself shifting underneath the strategy.

Where the three diverge

Three strategies with the same profit factor but completely different headline numbers everywhere else. Strategy A (the scalper) has the highest win rate but the lowest expectancy per trade — the small wins barely cover the larger losses. Strategy B (the balanced day-trader) has the best per-trade expectancy. Strategy C (the trend follower) has the lowest win rate but the largest per-trade payout. Multiplied by trade frequency, the expected monthly return is also completely different — Strategy B wins despite middle-of-the-pack metrics on every other dimension. None of this is visible from profit factor alone.

The lesson: profit factor is comparable across strategies only when frequency is similar. Win rate is comparable across strategies only when R:R is similar. Expectancy multiplied by frequency is comparable across all strategies, period. That's why it's the metric that decides which strategy you should actually run.

The combinations that look great but fail

Three patterns retail traders fall into when over-relying on one metric:

High win rate, negative expectancy. A strategy wins 85% of the time but the average loss is 6× the average win. Net per-trade expectancy is negative, but the equity curve looks beautiful for stretches because losses are rare. The first time the strategy hits a few losses in a row, three months of small wins are erased. Reality check: compute expectancy. If it's negative, no win rate saves you.

High profit factor, tiny sample. A strategy has profit factor 4.5 over 30 trades. Looks like genius. The 4.5 number is largely driven by one or two outlier wins that probably won't repeat. Reality check: compute profit factor with the largest 10% of wins removed and see what's left. If the residual is below 1.0, the strategy's "edge" is one or two lucky trades.

Strong expectancy, frequency too low to live on. A strategy has expectancy of $200 per trade but takes 6 trades per year. Annual expected return is $1,200. Even if the strategy has real edge, the sample is too small for the statistics to settle and the absolute return is too small for the time invested. Reality check: multiply expectancy × frequency and ask whether the expected annual return is worth the operational complexity.

Curve-fit metrics on optimised parameters. All three numbers can look great on a strategy that's been heavily optimised against historical data. Reality check: run the strategy on out-of-sample data (a different period, a different asset, ideally both). Real edge survives the move; curve-fit edge collapses.

The interactive

Try the four presets to see how each style produces very different metric profiles. The scalper preset shows the win-rate trap — high percentage, low per-trade payout. The balanced preset shows the cleanest combination of all three metrics. The trend preset shows a low win rate paired with strong expectancy, the shape of professional managed-futures returns. The losing preset is the trap — the metrics look fine on the headline but the rare losses dwarf the wins.

The most useful experiment: paste your own actual trade results and see what the calculator says. Most retail traders are surprised that their reported "65% win rate" strategy has expectancy near zero or slightly negative once the actual win/loss sizes are accounted for. That gap is the difference between feeling like you're winning and actually being profitable.

Beyond the three: when Sharpe and friends matter

For sizing decisions and capital allocation, expectancy is necessary but not sufficient. Two strategies with the same expectancy can have very different risk-adjusted returns — one might deliver its expected return smoothly, the other in jagged jumps. The metrics that capture this:

• Sharpe ratio = (return − risk-free rate) / standard deviation of returns. Penalises volatility on both sides — a strategy that swings wildly even to the upside scores worse. Above 1.0 is reasonable for retail, above 2.0 is institutional-grade, above 3.0 is suspiciously high without a long sample.
• Sortino ratio = like Sharpe but only penalises downside volatility. Closer to what retail traders actually care about — most of us don't mind upside variance.
• Calmar ratio = annual return / max drawdown. The most psychologically relevant metric for traders deciding how much to allocate. Above 1.0 means the worst drawdown is smaller than the annual return, which is the threshold most retail traders need to stay with a strategy through bad stretches.

These are the metrics professional fund evaluators look at. They're not relevant for "should I take this trade?" — that question is answered by the three above. They're relevant for "should I scale my position size?" — which only matters once expectancy has cleared zero.

For the broader picture of how these numbers interact with actual trade execution, the edge-after-cost post covers the cost-floor adjustment that has to be subtracted from gross expectancy before any of these metrics describe what your account will actually do. To compute the metrics on your own trade list, the R-multiple calculator gives you the per-trade R, and the annualized return projection wraps win rate, R:R, and trade frequency into a Monte-Carlo of what a year actually looks like. Both metrics also slice cleanly by tag once the underlying log carries them — the trade-tagging post lays out the four-label system that lets you read profit factor and expectancy per setup instead of as one blended number.