Trading expectancy and profit factor explained

Expectancy: expected value per trade

Expectancy answers: “If I take this setup a thousand times with the same rules, how much do I expect to make or lose per trade?” The classic formula in dollar terms:

Expectancy = (Win% × Avg Win) − (Loss% × Avg Loss)

Where Loss% = 1 − Win%. Positive expectancy means the strategy has a mathematical edge before costs; negative expectancy means you are paying the market to trade, regardless of how often you win.

Systematic traders often express expectancy in R-multiples: profit or loss relative to initial risk (1R = the distance from entry to stop at trade open). If you risk $500 per trade and net $700 on a winner, that is +1.4R. Expectancy in R-units is portable across position sizes and pairs naturally with fixed-fractional sizing.

E(R) = (Win% × Avg Win in R) − (Loss% × Avg Loss in R)

A strategy with E(R) = +0.20 and 200 trades per year expects +40R before compounding — but only if the distribution is stable and costs are included (see pitfalls below).

Profit factor: gross wins over gross losses

Profit factor (PF) is the ratio of total winning dollars to total losing dollars over a sample:

Profit Factor = Sum(Gross Wins) / Sum(|Gross Losses|)

Interpretation bands practitioners use (always sample-dependent):

• PF < 1.0: net losing; gross wins do not cover gross losses.
• PF = 1.0: break-even before costs; any friction pushes you negative.
• PF 1.25–1.5: modest edge; common for diversified systematic books.
• PF > 2.0: strong on paper; verify sample size and regime stability.

Profit factor and expectancy measure the same underlying edge from different angles. PF ignores trade count scaling; expectancy scales directly with frequency. A low-frequency trend system might show PF 2.5 with E(R) 0.8; a high-frequency scalper might show PF 1.15 with E(R) 0.05 but hundreds of trades per month.

Payoff ratio and the win-rate matrix

The payoff ratio (reward-to-risk ratio, win/loss ratio) is:

Payoff Ratio = Average Win / Average Loss

Win rate alone is misleading. The break-even win rate for a given payoff ratio is:

Break-even Win% = 1 / (1 + Payoff Ratio)

Examples:

• Payoff 1:1 (avg win equals avg loss) → need 50% wins to break even.
• Payoff 2:1 → need only 33.3% wins.
• Payoff 0.5:1 → need 66.7% wins — Harbor’s pre-refactor trap.

Rearranging expectancy in R-terms when losses are capped at 1R:

E(R) = Win% × Payoff − (1 − Win%) × 1

This is why trend followers accept 35–45% win rates: payoff ratios of 2:1 or 3:1 on winners compensate for frequent small losses. Mean-reversion systems often show 55–70% win rates but require payoff ratios above 0.6–0.8 to stay positive depending on costs.

Harbor’s pre-refactor row: 68% win rate at ~0.36 payoff → E(R) ≈ 0.68×0.36 − 0.32 ≈ −0.08R. Post-refactor: 54% at 0.92 payoff → E(R) ≈ +0.14R.

How these metrics connect to sizing and backtests

Kelly criterion uses win probability and payoff ratio to suggest maximum geometric-growth bet size. Full Kelly is volatile; desks typically use half-Kelly or fixed fractional risk (0.5–1% of equity per trade) once expectancy is positive out-of-sample. Expectancy tells you whether to trade; Kelly and heat limits tell you how much.

Backtests should report expectancy, profit factor, and payoff ratio on held-out data — not just CAGR and Sharpe. A backtest with PF 3.0 on 40 trades is noise; the same PF on 800 trades after walk-forward validation is evidence. Pair summary stats with MAE/MFE scatter plots to see whether stops are too wide (inflating payoff on paper while destroying live fill quality) or targets too tight (high win rate, tiny payoff).

Portfolio-level metrics like Sharpe ratio and maximum drawdown aggregate many trades and paths. Expectancy is trade-level; Sharpe is return-level. A positive-expectancy system can still produce a bad Sharpe if variance is high or correlation to your book spikes in crises.

Harbor Capital mean-reversion sleeve refactor

The sleeve traded RSI(2) oversold bounces in liquid sector ETFs with a fixed 1% equity risk per signal. Pre-refactor diagnostics on 412 trades:

• Win rate: 68.2%
• Average win: +0.40R; average loss: −1.10R (stops slipped on gap opens)
• Payoff ratio: 0.36; profit factor: 0.78; expectancy: −0.08R

Root causes: stops placed at arbitrary ATR multiples instead of MAE percentiles of historical winners; profit targets at fixed 0.5R while MFE showed 70% of winners exceeded 1.2R before reversing; no cost model in reported PF.

Changes applied on a walk-forward basis:

1. Stop width set to 85th percentile MAE of prior in-sample winners (reduced avg loss to 0.95R).
2. Scale-out: 50% at 0.8R, runner trail from MFE − 0.4R (raised avg win to 0.88R).
3. Subtracted 0.03R per round-trip for spread and slippage in reported stats.
4. Skipped signals when sector ETF spread exceeded 0.08% (reduced trade count 18%).

Out-of-sample (214 trades): win rate 54.2%, payoff 0.92, PF 1.31, E(R) +0.14R, annualized return +6.8% vs −4.2% prior. Lower win rate felt worse in dashboards but matched how profitable mean-reversion actually works: many small wins, fewer controlled losses, occasional outsized runners.

Metric decision table

Common pitfalls

• Celebrating win rate: 70% wins at 0.4:1 payoff is a slow bleed.
• Ignoring transaction costs: PF 1.05 in backtest becomes 0.92 live; always haircut expectancy.
• Tiny samples: 30 trades cannot support PF > 2 claims; require hundreds or bootstrap confidence intervals.
• Outlier dependence: one +10R trade can make PF look healthy; report median R and trimmed expectancy.
• Mixing R definitions: changing stop distance mid-sample invalidates R-multiple comparisons.
• Conflating profit factor with Sharpe: PF is gross win/loss; Sharpe is return volatility adjusted.
• Optimizing win rate with tighter targets: raises win% but crushes payoff and expectancy.
• Regime blindness: positive expectancy in low-vol years may vanish when spreads widen in crises.

Practitioner checklist

• Compute expectancy, profit factor, and payoff ratio on out-of-sample trades only.
• Express results in R-multiples with a fixed, documented stop rule per setup.
• Subtract realistic per-trade costs before declaring PF > 1.
• Plot MAE/MFE scatter to justify stop width and target scale-outs.
• Check break-even win rate for your payoff ratio; know your margin of safety.
• Report trade count and confidence: flag PF when N < 100.
• Segment by regime (volatility quartile, trend vs range) — expectancy is rarely uniform.
• Size positions from positive OOS expectancy via fractional Kelly or fixed risk, not win rate.
• Recompute monthly; edge decay shows up in falling PF before drawdowns spike.

Related reading

• Maximum adverse excursion and MFE explained — stop and target calibration from trade paths
• Backtesting trading strategies explained — walk-forward validation before trusting PF
• Kelly criterion explained — sizing from win rate and payoff ratio
• Risk management and position sizing explained — heat limits and R-based risk budgets