Mean Against Median Result, and What the Gap Between Them Admits

A trader submits a backtest showing mean return per trade of 2.1% and median return of 0.7%. The gap is 1.4 percentage points—large enough that it cannot be rounding or noise. Most traders note the discrepancy, assume the mean is what matters for expectancy calculations, and move on. That leaves the diagnostic information on the table. The distance between mean and median is not a curiosity. It describes distribution shape, reveals whether the system depends on rare large wins or hides tail risk in occasional disasters, and determines whether position sizing built on the mean will survive the journey most trades actually take. This article works through constructed examples with real numbers, identifies the patterns that appear in forex and crypto records, and states where the diagnostic breaks down.

What the gap measures and why it appears

A trader submits a year of records showing a mean return per trade of +0.82% and a median of +0.34%. The gap is not noise. It tells you that the distribution is pulling in one direction, and which trades are doing the pulling.

Mean sums every trade’s P&L and divides by the count. Median sorts the same trades from worst to best and picks the middle value. When the distribution is symmetric—gains and losses mirror each other in frequency and magnitude—the two figures land close together. When the distribution tilts, they separate. The size and direction of that separation describe the shape of the outcome set.

Positive skew produces a mean above the median. Most trades cluster around small gains or modest losses, but a handful of large wins pull the average upward. The median stays anchored near the center of the crowd; the mean chases the tail. In the constructed example above, the +0.82% mean suggests occasional outsized winners that the +0.34% median ignores. If you removed the top 5% of trades by profit, the mean would fall sharply while the median might barely move.

Negative skew reverses the pattern: median exceeds mean because large losses drag the average down while most trades post small gains or break even. A record showing a median of +0.40% and a mean of +0.15% points to tail risk—a few trades that lost multiples of the typical gain. This structure is common in strategies that collect small credits repeatedly and occasionally suffer blowups, such as short-volatility positions or naked option selling without defined risk.

The gap itself is neither good nor bad. A positively skewed equity curve can reflect disciplined loss-cutting and patient profit-taking, or it can mean the strategy depends on rare, unrepeatable conditions. Negative skew can signal structural vulnerability or simply the cost profile of the instrument. The gap measures distribution shape. What that shape costs or earns depends on position sizing, frequency, and whether the tail events arrive in clusters or isolation.

Below about one hundred trades, the gap becomes unreliable as a diagnostic. Small samples exaggerate outliers, and a single large trade can flip the relationship between mean and median without revealing anything durable about the process. The measure assumes the sample represents the strategy’s actual behaviour, which breaks when the trader changed method mid-sample or when market regime shifted partway through the period.

A worked example with constructed figures

The gap appears most clearly when you calculate it twice: once for a strategy that relies on outliers and once for a strategy that bleeds through them. The table below presents two constructed sequences of twenty trades each, identical in structure but opposite in shape.

Metric	Strategy A (positive skew)	Strategy B (negative skew)
Seventeen trades	+$40 each	+$40 each
Three trades	+$600, +$750, +$900	-$600, -$750, -$900
Sum of returns	+$2,930	-$1,570
Mean	+$146.50	-$78.50
Median	+$40.00	+$40.00
Gap (mean − median)	+$106.50	-$118.50

Strategy A: positive skew from rare large wins

Strategy A wins seventeen times at forty dollars and hits three outsized gains. Sorting the returns produces a sequence where the tenth and eleventh values—the middle two in a twenty-trade set—are both forty dollars, so the median is forty. The mean divides the twenty-nine hundred thirty dollar total by twenty and reaches one hundred forty-six fifty. The gap of one hundred six fifty is positive, meaning the mean sits well above the median. A trader glancing only at the mean would expect a typical trade to return near one hundred fifty dollars; the median reveals that seventeen of twenty trades returned forty.

Strategy B: negative skew from tail losses

Strategy B wins the same seventeen trades at forty dollars but loses three times at catastrophic size. The median remains forty dollars—the distribution of the bulk has not changed—but the mean drops to negative seventy-eight fifty. The gap is now negative one hundred eighteen fifty. The mean has been dragged below zero by three trades, yet half the sequence still shows a forty-dollar gain. This is the signature of tail risk: the average does not describe the center, and a trader sizing positions to the mean would be calibrating to a figure that most trades never approach.

The Pearson median skewness coefficient formalises the gap: multiply the difference between mean and median by three, then divide by the standard deviation of the series. For Strategy A, assuming a standard deviation of two hundred ninety dollars (constructed), the coefficient is approximately 1.10. For Strategy B, with a standard deviation of three hundred ten dollars, the coefficient is approximately −1.15. Both figures exceed 0.5 in absolute terms, confirming that neither distribution is symmetric and that relying on the mean alone would misrepresent the typical outcome in each case.

Typical patterns in forex and crypto records

A constructed sample of two hundred forex swing trades recorded over eighteen months might show a mean return of −0.3% per trade and a median of +0.4%. The mean trails because three stop-hunts and one flash event produced losses between 8% and 14% each, dragging the average below the center. Most trades closed near breakeven or with modest gains; the distribution has a long left tail. This is negative skew, and it appears in roughly three-quarters of the forex records we see.

Crypto strategies built around trend-following typically reverse the pattern. The same two-hundred-trade sample might yield a mean of +1.2% and a median of −0.1%, with the gap driven by four trades that caught rallies between 20% and 60%. The median sits negative because most sessions chop sideways or bleed slowly while waiting for momentum. Positive skew: a few large wins lift the average well above the middle value. The difference reflects market structure. Forex pairs revert around interest differentials and intervention levels, so outliers tend to be sharp reversals that stop out positions scaled too large. Crypto assets gap on low weekend liquidity and news flow, rewarding the trader still holding when spot moves 30% in four hours.

High-frequency strategies in either market show much smaller mean-median gaps, often under 0.05% in absolute terms. With samples in the thousands and hold times measured in minutes, the distribution compresses toward symmetry. Individual trades lack the room to become outliers, and the law of large numbers pushes the mean toward the median. The gap here reveals less about market character than about execution: if it widens suddenly, slippage or a fee-structure change is usually the cause.

That gap changes again once you subtract transaction costs and model slippage from backtest results. A crypto backtest showing +1.5% mean and +0.2% median might flip to −0.2% mean and +0.1% median in live execution if the strategy enters on market orders during volatile sessions. The positive skew collapses because the large winning trades suffer more slippage in percentage terms than the small losers, and the few trades that provided all the upward pull now provide much less. The median, built from the centre of the distribution where most trades are small, barely moves. This is the point at which a trader discovers that the backtest’s attractive mean return was an artefact of assuming instant fills at the close price.

The common belief that breaks: mean return as the representative trade

Most traders treat mean return as if it describes what will happen on the next trade. It does not. The mean tells you what to expect over many trades, not what a typical trade looks like. In any distribution with outliers—and trading returns are almost always skewed—the mean becomes a poor estimate of the experience you will actually have trade by trade.

Consider a constructed set of fifty trades with a mean return of +1.2R but a median of +0.3R. The mean sits well above the median because three trades returned +15R, +22R, and +31R. Those three pull the arithmetic average upward, but forty-seven of the fifty trades returned less than the mean. A trader entering the fifty-first position has no reason to expect +1.2R; the median tells them that half the trades in the sample fell below +0.3R and half above. That is the number around which the distribution balances, not the one inflated by rare winners.

The mirror case appears in negative-skew systems. A strategy with mean return of -0.4R and median of +0.6R delivers more trades above zero than below, but occasional large losses—say -18R, -24R, -29R—drag the mean into negative territory. The median here reveals that most trades are modest winners; the mean reveals that the system eventually loses money because the tail risk is not managed.

Mean return remains the correct input for position sizing and expectancy calculations, because those depend on long-run totals. But when a trader asks “what does a trade in this system usually look like,” the median answers. The gap between them is not noise. It is the signature of skew, and it tells you whether your system depends on rare large wins or is vulnerable to rare large losses.

How backtesting software masks the gap

Most backtesting platforms report total return, mean profit per trade, win rate, and profit factor. Median return per trade appears in almost none of them. This omission hides distribution shape entirely, and with it the difference between a system with consistent edges and one staking everything on rare outlier wins.

Consider two constructed systems, each tested over two hundred trades. System A produces a mean profit per trade of $47 and a median of $45. System B also shows a mean of $47, but a median of $12. Standard backtest output calls these equivalent: same mean, same total return if trade counts match. The gap between mean and median tells a different story. System A’s returns cluster tightly around the centre; the mean sits close to the typical trade. System B’s mean is pulled upward by a handful of large wins while most trades return far less. The median, resistant to outliers, stays near the experience of the majority of trades.

Grid and martingale systems routinely produce this pattern. They close dozens of small profits—raising the median—then compound position size into occasional catastrophic losses that drag the mean below the median, or close one enormous win that launches the mean far above it. A backtest showing mean profit per trade of $50 with median of $8 is not reporting a $50 edge. It is reporting that half the trades made less than $8 and a few large outcomes distorted the average. Position sizing built on that mean will allocate capital as though every trade carries a $50 expectancy, when the typical trade carries something closer to single digits.

Professional risk assessment examines mean, median, skewness, and kurtosis together for exactly this reason. Skewness quantifies asymmetry: positive skew when mean exceeds median, negative skew in reverse. Kurtosis measures tail weight, the frequency of extreme outcomes relative to a normal distribution. A system with high positive skew and high kurtosis bets on rare large gains; one with negative skew hides tail risk in a few severe losses. Neither is visible in the mean alone, and both have direct consequences for drawdown and required capital.

Without the median, you cannot separate consistency from lottery structure. This limitation becomes acute below roughly one hundred fifty trades, where a single outlier can swing the mean by twenty percent or more while leaving the median nearly unchanged. If your backtesting software does not calculate median return per trade, export the trade list and calculate it separately. The gap between the two numbers is not a footnote. It is the shape of the risk you are about to carry forward.

Where the diagnostic stops holding

Below roughly one hundred fifty trades, the gap between mean and median is usually not measurable in the sense that matters for diagnosis. Random variation dominates. A sample of eighty trades might show mean per-trade profit of $47 and median of $32, a gap of $15 that suggests positive skew. Run the same strategy for another eighty trades and the gap might reverse or vanish entirely, not because the distribution changed but because small samples are noisy. The gap becomes a stable diagnostic property only when the sample is large enough that its sign and rough magnitude persist across successive hundred-trade windows. Below that threshold, you are reading noise.

The gap reveals distribution shape. It does not reveal edge, profitability, or whether the strategy survives costs. A positively skewed distribution—mean above median—can still lose money if the median itself is negative, or if execution costs consume the average gain. A system that wins $200 once in fifty trades and loses $5 the other forty-nine times will show mean above median, but it loses $45 over the sample. The gap tells you the distribution is asymmetric; it says nothing about whether that asymmetry is profitable.

Non-stationarity distorts the gap in ways that are difficult to interpret. If market regime shifts halfway through the sample—a volatility collapse in crypto, a central bank intervention in forex—the gap reflects two mixed distributions rather than one stable process. Mean and median then describe an average of conditions that no longer exist together, and the gap offers little guidance for forward behaviour. The diagnostic assumes the generating process is roughly constant across the sample period.

The gap does not replace other diagnostics. It complements win rate, profit factor, drawdown analysis, and execution-cost accounting. A wide gap with mean below median flags tail risk, but drawdown analysis quantifies how much capital that risk consumed. A narrow gap suggests symmetry, but profit factor and cost accounting determine whether symmetry translates to sustainable returns. Use the gap as one lens among several, not as a summary statistic.

What you do with the information

When the mean sits fifteen percent above the median in a backtest, most traders adjust their expectations downward and move on. That stops too early. The gap tells you how to change position sizing, Monte Carlo parameters, and the terms on which you review third-party systems.

Positive-skew systems: size for the median journey

A constructed example: a trend-following system on EUR/USD shows a mean per-trade return of 1.8R and a median of 0.6R across two hundred trades. The gap points to a small number of large wins carrying the average. Size your positions assuming the median describes most of what you will experience. If your risk model allocates capital based on mean return, you are implicitly betting that the next twenty trades will include at least one of those rare large gains. They might not. Position size calculated from the median keeps you solvent through the stretches where the system grinds along below its long-run average. The mean matters for evaluating the system over a full cycle; the median matters for surviving until that cycle completes.

Negative-skew systems: manage the tail explicitly

Now reverse it. A range-bound scalping system shows a mean of 0.4R and a median of 0.9R. The mean sits below because a few large losses dragged it down. Those losses are not modeling errors; they are part of the system. If your risk limit assumes mean performance, you are underestimating the frequency and size of drawdowns. Set position size and account-level stop-losses based on the lower figure, and model maximum adverse excursion separately. The median tells you what most trades do. The gap tells you how bad the exceptions get.

Monte Carlo simulations: run both parameters

Most equity-curve simulators resample trades around a single central measure, usually the mean. Feed the simulation two separate runs: one using mean return per trade, another using median return with the empirical tail distribution preserved. Compare the resulting drawdown profiles. Systems with wide mean-median gaps produce equity curves that look stable under mean assumptions and catastrophic under median assumptions. The second view is closer to what you will trade through. If the two simulations produce similar results, the gap is narrow enough to ignore in position sizing.

Reviewing third-party systems

When a signal service or system vendor publishes performance, ask for median return per trade alongside mean. If they supply only the mean, calculate the median yourself from the trade log. Refusal to provide the log is its own answer. A system with mean return of two percent per trade and median return of two percent behaves differently from one with the same mean and a median of zero-point-four percent. The first has a symmetric distribution; the second relies on rare wins to stay profitable. Both can work, but the second requires larger capital reserves and longer evaluation periods. The absence of median figures in published results usually means the vendor has not calculated them, which raises questions about what else in the backtest went unexamined.

The gap is a single calculation—mean minus median—but it changes how you read every other performance metric. A win rate of 55% means one thing in a symmetric distribution and something else entirely when the mean sits two standard deviations above the median. Profit factor looks different when you know that three trades out of two hundred provided all the net gain. Maximum drawdown becomes more or less concerning depending on whether it came from one catastrophic loss or a slow grind through many small ones. Calculate the gap early. It does not tell you whether a system is profitable, but it tells you what kind of profitable it might be, and that determines the capital and temperament required to trade it.