Expectancy-Based System Evaluation

What it is

expectancy is the average amount you expect to make or lose per trade, expressed in units of the risk you took (r-multiples): expectancy equals win rate times average win in R, minus loss rate times average loss in R. A strategy with a 40 percent win rate, average winner of 2.5R and average loser of 1R has an expectancy of 0.4 times 2.5 minus 0.6 times 1, which is 0.4R per trade. Expectancy-based evaluation means judging every playbook in this wiki, and every trade you take, by that number and its uncertainty, not by whether the last ten trades felt good. This article is the measurement layer that sits under all the others.

The logic

Win rate alone is meaningless: a 90 percent win rate with 0.2R winners and 5R losers loses money. Profit alone is meaningless: a strategy up 20 percent on 15 trades could be luck. Expectancy in R normalises for position size and combines win rate with payoff into the only number that determines whether a strategy makes money over many trades. Its standard error tells you how many trades you need before you can trust it, which is almost always more than you have.

The "other side" of this evaluation is your own recency-bias and gamblers-fallacy. A trader who evaluates by feel will abandon a positive-expectancy system after a normal losing streak and stick with a negative-expectancy one after a lucky run. Expectancy with a confidence interval is the defence.

Setup rules

Record every trade in a trading-journal with: entry price, initial stop price, exit price, the R result (profit or loss divided by the initial risk), the setup name, the date and time, and the market regime tag you use. Without the initial stop, R cannot be computed and the evaluation is not possible.
Compute per setup, not per account. A profitable account can hide a losing setup and vice versa.
Metrics per setup: number of trades, win rate, average win in R, average loss in R, expectancy in R, profit-factor (gross wins divided by gross losses), maximum consecutive losses, max-drawdown in R, and the standard deviation of R outcomes.
Sample-size rule: treat expectancy as unknown until 30 trades, as provisional until 100, and as a working estimate after 200. A rough standard error is the standard deviation of R divided by the square root of the trade count; with a standard deviation of 1.5R and 50 trades, the error is about 0.21R, which means an observed 0.2R expectancy is indistinguishable from zero.
Rolling review: recompute the last 60 trades' expectancy monthly; a rolling expectancy that has been negative for two consecutive reviews triggers a size reduction, not a rule change.
Regime tags: record whether the index was above its 50-day average, whether volatility was high or low, and the day type if intraday; the per-regime breakdown is where most real findings live.

Entry, stop, target

Evaluation has no entries, but the table shows how the same data produces different conclusions depending on the metric, which is why the metric matters.

Setup	Trades	Win rate	Avg win (R)	Avg loss (R)	Expectancy (R)	Profit factor	Verdict
A	120	72 percent	0.6	1.4	0.04	1.10	Marginal; costs may make it negative
B	85	38 percent	2.8	1.0	0.44	1.72	Promising; sample still small
C	30	60 percent	1.5	1.0	0.50	2.25	Unknown; too few trades
D	210	48 percent	1.1	1.1	0.0	1.02	Break-even; stop trading it

Setup A would look best on a win-rate leaderboard; setup B is the one worth developing; setup C cannot be judged; setup D is where most traders' "main setup" actually sits once measured honestly.

Position sizing and risk

Expectancy connects directly to sizing. Position size should scale with confidence in expectancy, which means small size while the sample is small and larger size only after 100 or more trades confirm a positive number; that is the practical form of the fractional-kelly-criterion argument at /learn/risk-management. /tools/position-size handles the per-trade arithmetic once the risk percentage is chosen. The standard deviation of R and the maximum consecutive losses tell you what the drawdown will look like at a given risk per trade; risk-of-ruin is computed from those, not from the expectancy alone.

What breaks it

Missing stops. Trades without a recorded initial stop cannot be converted to R, and traders who move stops before recording them corrupt every number that follows.
Selection in the journal. Leaving out "that one that didn't count" makes the record fiction. Every trade counts.
Regime mixing. A setup that is 0.5R in trending markets and minus 0.3R in ranges shows 0.1R overall; the average hides the finding.
Small samples and big claims. Thirty trades produce expectancy estimates that can swing from minus 0.5R to plus 0.5R by chance; treating them as truth is how traders scale into a coin flip.
Costs excluded. Backtest expectancy without spread, slippage and commissions overstates real expectancy, sometimes by more than the whole edge for short-term setups.
Edge decay is invisible to a lifetime average and visible in the rolling one; the rolling review exists for exactly that.

How to test it

The evaluation framework is tested by applying it to a known-random series: generate 200 coin-flip trades with a 1R stop and a 1R target and compute all the metrics, then look at how far the rolling 60-trade expectancy wanders (it will show "hot" and "cold" streaks that mean nothing). That calibrates your intuition for noise. Then apply the framework to your own journal, setup by setup, and rank setups by expectancy divided by its standard error (a t-statistic of sorts); anything below 2 is not yet distinguishable from zero. Keep only the setups that clear the bar, trade them at modest size, and re-evaluate quarterly. See walk-forward-testing for the same idea applied to system parameters.

Variations

Expectancy per unit time (R per day or per week of exposure) for comparing fast and slow setups fairly.
Expectancy with regime conditioning as the primary metric, reported as a small table rather than a single number.
Monte Carlo resampling of the trade list to estimate the distribution of drawdowns at a chosen risk per trade; useful once the sample passes 100.