You have a system. It has rules. You ran a backtest and the equity curve slopes upward. The next question most traders ask is “how much should I risk per trade?” That question is premature. The first question should be: does this system actually have a positive expectancy, and how do I know?
Expectancy and R-multiples are the measurement layer that sits between backtesting and position sizing. Skip them, and you are sizing positions on a system you have not actually validated. I have seen traders jump straight from a profitable backtest to Kelly criterion sizing without ever measuring their expectancy per unit of risk. The Kelly formula requires an edge as input. If you feed it garbage numbers, it outputs garbage position sizes.
This article covers what expectancy and R-multiples actually measure, how to calculate both correctly, where the common mistakes hide, and how the output connects to position sizing decisions you will make later.
What R-Multiples Measure
An R-multiple expresses a trade’s result as a ratio of the initial risk. R is the dollar amount you planned to lose if the trade went wrong. If you buy a stock at $50 with a stop at $47, your R is $3 per share. A trade that makes $9 per share returned 3R. A trade that lost $3 returned -1R. A trade that hit the stop and slipped to a $3.60 loss returned -1.2R.
The formula is straightforward:
R\text{-multiple} = \frac{\text{Trade P\&L}}{\text{Initial Risk (R)}}
The value of this framing is normalization. A $500 profit on a $250 risk and a $5,000 profit on a $2,500 risk are both 2R trades. Once you convert every trade to R-multiples, you can compare results across different position sizes, different price levels, and different instruments without the dollar amounts distorting the picture.
I track all my swing trades in R-multiples because the raw P&L is misleading. A $2,000 winner sounds impressive until you realize the initial risk was $1,800. That is a 1.1R trade. Barely worth the screen time.
One mistake traders make here: using a theoretical stop that was never actually in the order system. If your planned stop was $47 but you “would have” exited at $48.50, your R is not $1.50. R must reflect the actual risk you accepted when you entered. Retroactively tightening stops to make R-multiples look better corrupts every downstream calculation.
Calculating Expectancy
Expectancy tells you how much you expect to make per unit of risk, on average, across many trades. It combines your win rate with the size of your wins and losses, all expressed in R.
The formula:
\text{Expectancy} = (W \times \overline{R_{\text{win}}}) + (L \times \overline{R_{\text{loss}}})
Where W is win rate (as a decimal), L is loss rate (1 – W), \overline{R_{\text{win}}} is the average R-multiple of winning trades, and \overline{R_{\text{loss}}} is the average R-multiple of losing trades (a negative number).
A system with 40% wins averaging +2.5R and 60% losses averaging -1.0R:
\text{Expectancy} = (0.40 \times 2.5) + (0.60 \times -1.0) = 1.0 - 0.6 = 0.40R
That system expects to earn 0.40R per trade over time. If R is $500, each trade is worth $200 on average. A system with negative expectancy loses money regardless of how you size it.
Notice what a 40% win rate looks like in raw terms: you lose six out of ten trades. Most traders would abandon that system within a month. But the 2.5R average winner more than compensates. This is exactly why expectancy matters. Win rate alone tells you nothing about profitability.
A Worked Example With a Real Trade Log
Suppose you have 20 trades from a breakout system. Your initial risk on every trade is 1R. Here are the R-multiples:
Winners (8 trades): +1.2R, +3.1R, +2.0R, +1.5R, +4.2R, +1.8R, +2.6R, +1.4R
Losers (12 trades): -1.0R, -1.0R, -0.8R, -1.0R, -1.1R, -1.0R, -0.9R, -1.0R, -1.0R, -1.2R, -1.0R, -1.0R
Win rate: 8/20 = 0.40
Average winner: (1.2 + 3.1 + 2.0 + 1.5 + 4.2 + 1.8 + 2.6 + 1.4) / 8 = 17.8 / 8 = +2.225R
Average loser: (-1.0 -1.0 -0.8 -1.0 -1.1 -1.0 -0.9 -1.0 -1.0 -1.2 -1.0 -1.0) / 12 = -12.0 / 12 = -1.0R
\text{Expectancy} = (0.40 \times 2.225) + (0.60 \times -1.0) = 0.89 - 0.60 = 0.29R
The system has a positive expectancy of 0.29R per trade. Over 100 trades at $500 per R, the expected gross profit is $14,500. That number is not a prediction. It is a statistical estimate that depends on the distribution holding.
What you should check next: is the 4.2R outlier carrying the result? Remove it and recalculate. Without the 4.2R trade, average winner drops to (13.6 / 7) = 1.943R. New expectancy: (7/19 x 1.943) + (12/19 x -1.0) = 0.715 – 0.632 = 0.08R. The system barely breaks even without that one outlier. That is critical information before you commit real capital.
Why Expectancy Must Come Before Position Sizing
The Kelly criterion takes win probability and payoff ratio as inputs to output optimal bet size. If expectancy is negative, Kelly outputs zero or negative. You should not be trading that system at all, let alone sizing positions for it.
But the failure mode I see more often is subtler. Traders calculate expectancy from a backtest, get a positive number, and immediately plug it into Kelly. The problem: backtest expectancy is not the same as forward expectancy. Curve-fitted parameters inflate backtest wins. Transaction costs and slippage eat into R-multiples. Market regime shifts change the distribution entirely.
This is where walk-forward analysis connects. Walk-forward testing gives you out-of-sample expectancy estimates. In-sample expectancy is a hypothesis. Out-of-sample expectancy is evidence. The difference between the two is often the difference between a system that looks profitable and one that actually is.
Position sizing amplifies whatever your system produces. If the system has 0.4R expectancy, aggressive sizing makes you rich slowly. If the system has -0.1R expectancy, aggressive sizing makes you poor quickly. Sizing is a multiplier, not a source of edge. Expectancy is the source.
Common Mistakes When Measuring Expectancy
The first mistake is using too few trades. Twenty trades is a starting point for illustration. It is not enough for confidence. With 20 trades and a 40% win rate, the standard error on the expectancy estimate is wide enough to include zero. I would want 60 or more trades before treating an expectancy number as reliable, and 100+ before sizing positions aggressively around it.
The second mistake is ignoring the distribution of R-multiples. An average is a summary statistic. It hides the shape. If your system produces fifty -1R losers and one +55R winner, your expectancy is positive but your actual experience is 50 consecutive losses punctuated by one windfall. Most traders cannot psychologically survive that. The median R-multiple and the standard deviation of R-multiples matter as much as the mean.
The third mistake is not accounting for slippage and commissions in the R calculation. If your planned risk was $3 per share but slippage on entries and exits costs $0.15 total, your effective R is $3.15 on losers and your effective gain is reduced by $0.15 on winners. Small per-trade friction compounds across hundreds of trades and can turn a marginally positive expectancy negative.
The fourth mistake is treating expectancy as fixed. Markets change. A trend-following system with 0.5R expectancy in trending markets might have -0.3R expectancy in choppy, range-bound conditions. Measuring expectancy across a single market regime and assuming it holds forever is how traders blow up. I recalculate expectancy on a rolling 50-trade window to catch deterioration early. When it crosses below 0.10R, I reduce position sizes before it potentially turns negative. That connects directly to hard drawdown stops as a capital preservation tool.
Connecting Expectancy to Position Sizing Decisions
Once you have a reliable expectancy estimate from out-of-sample data, you can make rational position sizing decisions. The chain works like this:
Expectancy tells you the expected R per trade. Multiply by trade frequency to get expected R per period. Multiply by your chosen R-value in dollars to get expected dollar return per period. Your R-value in dollars is the position sizing decision.
A system with 0.30R expectancy and 15 trades per month produces 4.5R per month in expectation. At $500 per R (1% of a $50,000 account), that is $2,250 per month before drawdowns. At $1,000 per R (2% of the account), it is $4,500 per month. The larger R also means larger drawdowns. A string of 8 losers at 2% per trade is a 16% drawdown before any winners arrive.
This is where VIX-based regime sizing becomes practical. Instead of using a fixed R-value, you scale R down when volatility is high (because losing streaks tend to cluster in high-volatility regimes) and scale R up when volatility is low and your system’s expectancy historically improves.
What expectancy does not tell you: the maximum drawdown, the longest losing streak, or the probability of ruin. For those, you need to look at the full distribution of R-multiples and run scenario analysis. A positive expectancy system can still produce drawdowns that exceed your tolerance if you size too aggressively. Expectancy is necessary for profitable trading. It is not sufficient for survival.
The R-Multiple Distribution Matters More Than the Average
Two systems can have identical expectancy and feel completely different to trade. System A: 60% win rate, average winner +1.2R, average loser -1.0R. System B: 30% win rate, average winner +4.0R, average loser -1.0R. Both have 0.32R expectancy. System A delivers small, frequent wins. System B delivers long losing streaks interrupted by occasional large winners.
The R-multiple distribution determines which system you can actually stick with. Most traders say they want high reward-to-risk ratios, but when System B puts them through 12 consecutive losses, they abandon it three trades before the 4R winner arrives.
Plotting your R-multiples as a histogram reveals the shape. A healthy trend-following system typically shows a cluster of -1R losers, a spread of small winners between 0R and 2R, and a thin tail of large winners above 3R. If that tail disappears, your expectancy collapses. If the cluster of losers shifts from -1R to -1.3R (stops getting hit with slippage), your expectancy degrades from below.
I keep a running histogram updated after every 10 trades. When the shape changes, the system is telling me something. Either the market regime shifted, or my execution is drifting. Both require attention before the expectancy number catches up to the deterioration.
When Expectancy Turns Negative
A negative expectancy means the system loses money per trade on average. No position sizing scheme fixes this. Not Kelly. Not fixed fractional. Not martingale. Nothing. You cannot size your way out of a losing system.
The common response is to add filters, tighten stops, or change the indicator parameters. That is fine as a research exercise, but every change restarts the expectancy measurement. The new configuration needs its own out-of-sample test and its own trade count before you trust the revised number.
What does NOT work: optimizing parameters until backtest expectancy turns positive and then trading the optimized version. That is curve fitting. The walk-forward approach exists specifically to guard against this. If the optimized parameters produce negative expectancy in the out-of-sample window, the system does not have an edge. Full stop.
The hardest decision in systematic trading is accepting that a system you spent months building has no edge. Expectancy gives you the number. What you do with it is a discipline problem, not a math problem.
Measuring Edge, Not Predicting Returns
Expectancy is a backward-looking measurement. It tells you what the system did produce per unit of risk, not what it will produce. The assumption that past expectancy continues into the future is an assumption, not a fact. Market structure changes. Liquidity shifts. Strategies that worked in low-rate environments behave differently in high-rate environments.
The correct use of expectancy is as a monitoring tool, not a forecasting tool. Calculate it. Track it over rolling windows. Compare in-sample to out-of-sample estimates. When the two diverge significantly, investigate. When out-of-sample expectancy drops toward zero, reduce exposure before it goes negative.
Combine expectancy monitoring with drawdown-based circuit breakers. If the account hits a predefined drawdown threshold, stop trading regardless of what the expectancy number says. The drawdown is real. The expectancy estimate might be stale.
Expectancy and R-multiples do not make trading predictable. They make it measurable. That is the prerequisite for every position sizing decision that follows.
Educational content only. Not investment advice. Trading involves risk. You are responsible for your decisions.