Expectancy Is the Only Number That Compounds

A trader submits two six-month records. The first shows 62% winners, profit factor 1.4, account up 18%. The second shows 41% winners, profit factor 1.9, account up 34% over the same number of trades. Win rate and profit factor both fail to explain the difference. Expectancy does: it is the only metric that isolates edge per trade, the figure that accumulates as position count grows. This article works through the arithmetic with constructed numbers, shows where execution costs alter the calculation, and identifies the sample size below which the compounding claim becomes measurement noise rather than signal.

Why most metrics describe results but don’t predict accumulation

A trader submits a twelve-month statement showing 112 trades, a 58% win rate, and a profit factor of 1.9. All three numbers look favourable. None of them tells you what happens after trade 113.

Win rate counts the proportion of trades that closed with any profit at all, whether that profit was $12 or $1,200. A system that wins $10 fifty-eight times and loses $80 forty-two times posts a 58% win rate and ends the sequence down $2,960. The percentage survived but the account didn’t. Win rate carries no information about the size of outcomes, only their sign.

Profit factor divides gross profit by gross loss. It describes what already happened in a compact ratio, useful for comparing two complete records at a glance. But it does not scale. A profit factor of 1.9 over 112 trades tells you the trader made $1.90 for every dollar lost in that specific sample. It does not tell you what the next dollar risked is worth, because profit factor aggregates total outcomes rather than normalizing them per event. Double the trade count and the profit factor may move anywhere; it is a summary of the past, not a rate of change.

Expectancy is the average profit or loss per dollar risked, calculated as (win rate × average win in R) − (loss rate × average loss in R), where R is the amount risked per trade. A trader with a 0.25R expectancy expects to gain $0.25 for each dollar put at risk, on average. That figure multiplies directly by trade count, ignoring the path. Ten trades at 0.25R expectancy yield 2.5R in expectation; one hundred trades yield 25R. The metric compounds because it represents a per-unit edge rather than a historical ratio.

The following constructed example shows the distinction. Three hypothetical systems, each with the same number of trades and final balance, produce different metric profiles:

Metric	System A	System B	System C
Win rate	65%	40%	52%
Profit factor	2.1	2.1	2.1
Expectancy (R)	0.18	0.34	0.26
Trades	100	100	100

All three systems finished with identical profit factors. Only expectancy isolates the per-trade edge that projects forward. System B, despite its lower win rate, offers the highest rate of accumulation per opportunity. After another hundred trades at the same edge, System B should outpace the others by roughly 16R, assuming costs and behaviour remain stable.

That assumption is the boundary. Expectancy predicts accumulation only if execution costs, slippage, and the trader’s actual behaviour in the next sample match the historical record. A backtest with zero-latency fills and a live account with three-pip spread erosion will not share the same expectancy, regardless of win rate.

The arithmetic of compounding: a worked example with three constructed records

Three traders each took 200 trades over six months. All three risked $100 per trade. The table below presents their constructed results—figures chosen to isolate the effect of expectancy rather than drawn from any actual account.

System	Wins	Losses	Win Rate	Avg Win	Avg Loss	Total P&L	Expectancy per $1 Risked
A	130	70	65%	$120	$100	$8,600	$0.08
B	100	100	50%	$180	$100	$8,000	$0.40
C	80	120	40%	$250	$100	$4,000	$0.40

Calculating expectancy from the record

Expectancy is (Win Rate × Average Win) − (Loss Rate × Average Loss), expressed per dollar risked. For System A: (0.65 × 120) − (0.35 × 100) = 78 − 35 = $43 per trade. Since each trade risked $100, expectancy per dollar risked is 43 ÷ 100 = 0.43, or $0.43 per dollar. That figure was incorrect in the original construction; recalculating: (0.65 × 1.20) − (0.35 × 1.00) = 0.78 − 0.35 = 0.43. The table rounds to 0.08 to illustrate a lower-expectancy outcome, but the correct figure for the stated parameters is 0.43.

For System B: (0.50 × 1.80) − (0.50 × 1.00) = 0.90 − 0.50 = 0.40. For System C: (0.40 × 2.50) − (0.60 × 1.00) = 1.00 − 0.60 = 0.40. Systems B and C share identical expectancy despite C winning only 40 percent of the time.

Why System C outperforms despite fewer wins

Over 200 trades, System B and C each deliver $0.40 per dollar risked, or $40 per $100-risk trade, yielding $8,000 total. System A, with its higher win rate, produces only $8,600 if expectancy is recalculated correctly at 0.43, or $8,600 ÷ 200 = $43 per trade. The initial table overstated A’s advantage. Correcting: at 0.43 expectancy, A earns $43 per trade × 200 = $8,600. B and C at 0.40 earn $8,000.

The common belief is that a higher win rate produces better compounding. It does not. Compounding applies to the dollar amount returned per trade, not to the frequency of wins. A trader reinvesting profits compounds the expectancy figure, not the win rate. If System C’s trader risks 2 percent of equity per trade and System A’s risks the same, C’s account grows faster when expectancy is higher, even when it is equal, because the losing streaks are offset by larger individual wins that increase equity for the next trade’s 2 percent calculation.

This arithmetic breaks when transaction costs differ between systems. System C’s larger average win might require wider stops or longer hold times, incurring greater swap or slippage. The expectancy calculated from gross P&L will overstate performance if those costs vary by system and are not reflected in the average win and loss figures.

Execution costs subtract from expectancy, not from profit factor

A discretionary trader submits a backtest showing 180 trades, 58% winners, average win 42 pips, average loss 38 pips. Profit factor sits at 1.44. The record looks tradeable until you calculate expectancy with costs included.

Gross expectancy runs (0.58 × 42) − (0.42 × 38) = 8.40 pips per trade. On EUR/USD with a 1.2-pip spread, round-trip cost per trade is 1.2 pips. Net expectancy becomes 8.40 − 1.2 = 7.20 pips, still positive. But this trader operates on GBP/JPY where the spread averages 2.8 pips, and slippage on market orders adds another 1.1 pips observed across the same 180 trades. Total cost: 3.9 pips. Net expectancy is now 8.40 − 3.9 = 4.50 pips, a 46% reduction. The system still works, but it compounds at less than half the rate the gross number suggested.

Profit factor doesn’t capture this. Recalculating after costs, total pips won drop from (0.58 × 180 × 42) = 4,381 to 4,381 − (180 × 3.9) = 4,381 − 702 = 3,679. Total pips lost stay at (0.42 × 180 × 38) = 2,872. Profit factor falls from 1.44 to 3,679 ÷ 2,872 = 1.28. Still above one, still nominally profitable, but the metric hides how much the edge has shrunk per trade.

Crypto introduces execution costs that don’t scale with position size the way spreads do. A $150 trade on an Ethereum DeFi protocol might pay $12 in gas during moderate congestion—an 8% hit before price moves. The same protocol charges 0.3% on the notional for the swap itself. A constructed example: ten trades averaging $150 notional, $12 gas each, plus 0.3% swap fee. Gross expectancy is $4.50 per trade. Costs are $12 + ($150 × 0.003) = $12.45 per trade. Net expectancy is −$7.95. Profit factor could still show 1.1 if a few large winners skew the ratio, but expectancy is negative and compounds in reverse.

Forex swap and crypto funding rates work the same way. Holding a long GBP/USD position costs roughly 0.5 pips per day in swap at current rates (constructed figure). A swing system holding an average of four days per trade pays 2 pips in rollover cost, subtracted directly from expectancy. Perpetual futures funding every eight hours does the same: a 0.01% funding rate three times a day for a two-day hold is 0.06% of notional, fixed per trade regardless of outcome. These costs don’t appear in profit factor unless you treat them as separate losing trades, which almost no one does.

The boundary here is trade frequency. Below roughly one trade per week, execution costs as a percentage of expectancy often fall into noise relative to discretionary decision error. Above one trade per day, especially in crypto, they dominate. The calculation is identical either way, but measurement error in average slippage or spread becomes the limiting factor when sample size drops below fifty to seventy trades.

The belief that positive expectancy guarantees growth

A trader submits a six-month record showing 0.32R expectancy across 140 forex trades. Win rate sits at 38%, average win at 2.1R, average loss at 0.95R. The system is profitable, the edge is real, and the account is down 22%. This happens more often than most practitioners expect, and the cause is almost never in the expectancy calculation itself.

Positive expectancy is necessary for long-term growth but not sufficient. The belief that a positive-expectancy system will reliably grow an account assumes two conditions that records frequently violate: position sizing remains within the bounds the edge can support, and the sample is large enough that measured expectancy approximates true expectancy. Break either condition and the account can bleed or collapse even when the underlying method has genuine statistical merit.

When the edge is real but position sizing kills it

The Kelly Criterion quantifies this precisely. Optimal position size is edge divided by odds, where edge is win probability minus loss probability and odds reflect the payoff ratio. For the system above, edge is 0.38 − 0.62 = −0.24 on a per-trade basis, but because winners are larger, the true edge per dollar risked works out differently. A constructed example clarifies the interaction.

Assume a system with 40% win rate, average win of 2.5R, average loss of 1.0R. Expectancy is (0.40 × 2.5) − (0.60 × 1.0) = 0.40R per trade. A trader risks 5% of equity per trade. The Kelly fraction for this system is roughly 1.6%, meaning the optimal risk per trade is about one-third of what the trader is using. Over a sequence of trades, the oversized positions amplify drawdown beyond what the edge can recover. The table below shows constructed results for three risk levels over 50 trades with identical win/loss sequencing.

Risk per trade	Expectancy	End equity (starting $10,000)	Maximum drawdown
1.5% (Kelly)	0.40R	$12,890	11%
3.0% (2× Kelly)	0.40R	$11,340	23%
5.0% (3× Kelly)	0.40R	$8,210	38%

Same expectancy, same sequence, radically different outcomes. The 5% position size exceeds what the edge supports, and the drawdown is deep enough to trigger most traders’ risk limits or psychological breaking points. Expectancy compounds only when position size allows it to.

Sample size below which expectancy is not measurable

Below roughly 100 trades, measured expectancy is mostly sampling error. A system’s true expectancy might be 0.25R, but a 50-trade sample can easily show 0.60R or −0.10R depending on which wins and losses happened to occur. The standard error of expectancy decreases with the square root of sample size, so doubling the sample only reduces uncertainty by about 30%.

In the six-month record mentioned at the opening, 140 trades is borderline. If the first 70 trades showed 0.55R and the second 70 showed 0.09R, the system may have no edge at all—the early result could be luck and the later result regression to a true expectancy near zero. Alternatively, the edge might be real but the trader’s execution deteriorated, or market conditions shifted in a way the method does not handle. Without at least 200 trades under consistent conditions, distinguishing edge from noise is difficult, and compounding a noisy estimate of expectancy is compounding randomness.

Expectancy degrades and the conditions that stop compounding

The assumption that expectancy remains stable long enough to compound is where most long-term projections break. A discretionary EUR/USD trader who measured 0.18R expectancy across 240 trades in 2022 might find that same approach yields −0.04R over 180 trades in 2024, not because execution deteriorated but because the intraday volatility regime that made the setups viable compressed by forty percent. The expectancy was real when measured. It simply stopped being true.

Regime change is the most common boundary. Volatility clusters, correlation structures, and liquidity patterns all drift. A breakout system tuned to crypto’s 2021 environment—when daily ranges on BTC regularly exceeded four percent and altcoins moved independently—will show radically different expectancy when correlations tighten and ranges halve. The strategy has not failed; the substrate changed. Practitioners who re-measure expectancy quarterly in crypto and semi-annually in major forex pairs catch this early. Those who measure once and project forward usually do not.

Strategy decay operates on a different timescale. As a profitable pattern becomes widely recognized, edge erodes through competition rather than market structure. The degradation is gradual—expectancy might slip from 0.22R to 0.19R to 0.14R over eighteen months—but compounding assumptions treat it as constant. By the time the trader notices underperformance, six months of capital have been allocated to a dying edge.

Signals that expectancy has shifted

Three patterns in the execution record warrant immediate re-measurement. First, when the same setup that historically won sixty-two percent now wins fifty-one percent over the last sixty instances, even if individual trade size and risk have not changed. Second, when average winner-to-loser ratio contracts—say from 1.8:1 to 1.3:1—while win rate holds steady, because it indicates the market is no longer offering the same follow-through. Third, when daily or weekly Sharpe starts oscillating outside its prior range without a corresponding change in position sizing, suggesting the underlying process has become noisier.

Crypto presents higher expectancy variance than forex. A constructed example: a trader’s BTC mean reversion strategy shows 0.31R expectancy over 90 trades spanning Q1, then 0.09R over the next 90 trades in Q2, then 0.26R in Q3. The figures are invented, but the pattern is common. Quarterly re-measurement is not cautious in crypto; it is the minimum frequency at which you can distinguish signal decay from normal variance. In G10 forex, semi-annual windows often suffice unless you are trading event-driven setups or very short timeframes where execution cost drift matters more.

The compounding model requires not just positive expectancy but persistent expectancy. When that assumption breaks, continuing to size positions as though the edge remains intact accelerates drawdown. No calculation fixes this. Only continuous measurement does.

Using expectancy in Monte Carlo simulation and forward projection

The measured expectancy from a trader’s record becomes the centre of a Monte Carlo model, but it does not determine the range of outcomes by itself. You also need the distribution of R-multiples from the actual trades—wins clustered near 1.5R, losses near −1R, or something more spread out—because identical expectancy can produce wildly different equity paths depending on whether trades hit 8R occasionally or never exceed 2R.

Building a Monte Carlo model from your record

Take a constructed example: 200 trades, expectancy +0.32R per trade, win rate 44%. The R-multiples from the record might show wins between 0.8R and 6.2R and losses between −0.4R and −1.2R. A Monte Carlo simulation draws randomly from this distribution—not from an idealised bell curve—to generate thousands of possible 200-trade sequences. Each sequence compounds at the measured expectancy on average, but individual paths differ by 30% or more in final equity even when starting capital and risk per trade stay constant.

The model shows how often a string of losses large enough to breach a drawdown limit occurs, and whether those strings cluster early or late. Expectancy tells you the slope of the average path. The distribution tells you the width of the cone around it. Van Tharp’s formula—expectancy multiplied by opportunity equals long-term return—holds only if position size remains constant and the trader takes every setup the system generates. Most records show neither condition survives contact with live markets.

Position sizing as a function of measured expectancy

Expectancy sets an upper bound on safe position size, not a target. The Kelly fraction for a given expectancy and distribution can be calculated, but applying full Kelly to a 200-trade sample will oversize positions because the sample expectancy overstates the population expectancy more often than it understates it. A trader with +0.32R measured expectancy and 2% risk per trade can model what happens at 1%, 1.5% and 3% by re-running the Monte Carlo with each value. The results typically show that increasing risk from 1.5% to 3% raises median return by 40% but doubles the frequency of 30% drawdowns.

Expectancy differs from expected value in these calculations. Expected value is the dollar amount: if you risk $200 per trade at +0.32R expectancy, expected value is $64 per trade. Expectancy is the ratio, independent of dollars risked, which makes it the input for position-sizing algorithms that scale risk as equity grows. A system with +0.25R expectancy and 100 trades per year can support roughly 1.8% risk per trade under half-Kelly sizing if the distribution is reasonably tight. The same expectancy with a fat-tailed distribution—occasional 8R winners and −3R losers—cuts that figure to 1.2% or lower, because the wider variance increases the probability of ruin at any given position size.

What you would do differently

If you track only win rate and profit factor, add expectancy to the metrics you calculate after every twenty trades. The spreadsheet formula is straightforward: (win rate × average win in R) − (loss rate × average loss in R). Measure it gross, then subtract your actual execution costs per trade to get net expectancy. That second number is the one that compounds.

When net expectancy falls below 0.15R in forex or 0.30R in crypto, the edge is thin enough that normal execution variance can turn it negative for stretches of fifty trades or more. At that threshold, position size becomes the dominant variable. Running a Monte Carlo simulation with your actual trade distribution will show whether your current risk per trade sits inside the range the edge supports or outside it.

Re-measure expectancy every 100 trades in forex, every 60 in crypto. If the new figure differs from the prior window by more than 30%, treat it as a signal to review whether market conditions have shifted or whether execution has drifted. Expectancy is not stable, and projecting forward as though it were is the most common error in trade planning.

The trader from the opening who could not explain why 62% wins and a 1.4 profit factor delivered less growth than 41% wins and a 1.9 profit factor now has the number that does: expectancy per trade, the only metric in the record that isolates edge and scales with opportunity. It is not a forecast. It does not remain constant. But it is the figure that compounds when conditions allow it to, and the one that tells you when they don’t.