Mean, Median and Distribution

ID: E-002

Introduction

E-001 established the first principle of quantitative trading analysis: before trying to prove what a strategy can do, the trader must understand what the observed data actually look like. E-002 takes the next step by examining three closely related ideas: the mean, the median, and the distribution of observations.

These concepts are often taught as simple statistical definitions. For a professional trader, that is not enough. The important question is not merely how to calculate an average or locate a median. The important question is what those numbers reveal about the structure of trading outcomes, what they hide, and how the shape of the distribution changes the interpretation of strategy performance.

A strategy can have a positive mean while most individual trades are negative. It can have a median near zero while a small number of large winners create most of the total return. Two strategies can have the same average outcome and completely different distributions of risk and reward.

The Expert-level objective is therefore to stop treating a summary statistic as the dataset itself. The mean and median are summaries. The distribution is the broader structure in which those summaries must be interpreted.

1. The Mean as a Measure of Central Tendency

The arithmetic mean is calculated by adding all observations and dividing by the number of observations.

For n observations:

x̄ = (x₁ + x₂ + … + xₙ) / n

In trading, the observations might be daily returns, trade results, R-multiples, holding times, drawdowns, or any other measurable variable.

Consider five trade outcomes:

+2R

+1R

-1R

-1R

+4R

The total is +5R, so the mean trade outcome is +1R.

The mean compresses five observations into one number. That makes it useful, but compression always removes information. The mean does not tell us whether the observations were tightly clustered, widely dispersed, heavily skewed, or dominated by a small number of extreme values.

For a trader, this distinction matters because the same mean can be produced by very different trading mechanisms.

2. The Median as a Robust Center

The median is the middle observation after the data have been ordered from smallest to largest. If the number of observations is even, the median is the average of the two central observations.

Consider:

-1R

-0.5R

0R

0.5R

1R

The median is 0R.

Now replace the final observation with +10R:

-1R

-0.5R

0R

0.5R

+10R

The median remains 0R, while the mean becomes +1.8R.

This illustrates the median's main practical advantage: it is less sensitive to extreme observations than the mean.

That does not make the median universally superior. It makes it informative in a different way. The mean tells us about the arithmetic center of the outcomes. The median tells us where the middle observation sits. The distance between them can itself reveal asymmetry.

3. Mean and Median Should Be Read Together

A trader who looks only at the mean can miss an important part of the outcome structure. A trader who looks only at the median can make the opposite mistake.

Suppose a strategy produces:

Median trade: -0.1R

Mean trade: +0.35R

At first glance, the result may seem contradictory. How can the average be positive if the middle trade is negative?

Because the distribution may be right-skewed. A relatively small number of large positive outcomes can pull the mean upward even though more than half of the observations are at or below the median.

This is not automatically a problem. Some trading strategies are intentionally designed to capture rare, large moves. Trend-following systems are a classic example of a mechanism in which a limited number of outsized winners can contribute disproportionately to total performance.

The correct question is therefore not: "Is it bad if the median is lower than the mean?"

The correct question is: "Is the observed relationship between the mean and median consistent with the strategy's intended edge and risk structure?"

4. Distribution Is the Bigger Picture

A distribution describes how observations are spread across possible values. In trading, it is the structure behind the individual results.

Imagine two strategies, each with a mean trade outcome of +0.30R.

Strategy A produces many outcomes close to +0.30R, with relatively few extremes.

Strategy B produces many small losses and small wins, but occasionally generates very large winners.

The mean is identical. The trading experience is not.

The distribution affects the frequency of winning and losing trades, the size of typical outcomes, the frequency of extreme outcomes, the shape of drawdowns, and the psychological experience of executing the system.

A professional performance review therefore asks not only where the center is, but also how the observations are arranged around it.

5. Symmetry and Skewness

A distribution is symmetric when observations are distributed in a broadly balanced way around its center. A skewed distribution has a longer or heavier tail on one side.

A positive or right-skewed trading distribution may contain many relatively small outcomes and a smaller number of very large positive outcomes. A negative or left-skewed distribution may contain many modest gains and occasional large losses.

These two structures can produce similar means while carrying very different risks.

This is why the phrase "average return" is incomplete unless the trader also understands the distribution that produced it.

Skewness should not be interpreted as a simple good-versus-bad classification. Its meaning depends on the strategy. A system designed around rare large winners may naturally have positive skew. A strategy that collects small premiums while accepting rare large losses may naturally have negative skew.

The important task is to identify the mechanism behind the shape and determine whether the risk model is designed for it.

6. Why the Most Common Outcome Is Not the Same as the Mean

Another useful distinction is between the mean, the median, and the mode. The mode is the most frequently occurring value.

Trading data often contain continuous or highly varied values, so an exact mode may be less informative than in a simple discrete dataset. Still, the idea is useful: the most common outcome is not necessarily the central arithmetic outcome.

For example, suppose a strategy produces many results around -1R and +1R, with occasional +5R outcomes. The most common region of the distribution may sit far from the mean.

This matters because a trader may experience the typical trade very differently from the average trade.

The mean is an expectation of the arithmetic values in the observed sample. It is not a description of what the "typical" trade must look like.

7. Outliers Can Move the Mean Without Changing the Median

An extreme observation can have a large influence on the mean because every observation contributes directly to its calculation.

The median is much more resistant to a single extreme value.

Suppose 99 trades are clustered between -1R and +1R and one trade is +20R. The +20R trade can materially increase the mean, while the median may barely move.

This creates an important diagnostic opportunity.

If mean and median are far apart, the trader should investigate why.

Possible explanations include:

A genuinely asymmetric strategy distribution.

A small number of very large winners or losers.

A special market regime.

An unusual event.

An execution or data-quality problem.

The gap between mean and median is therefore not a verdict. It is a question generator.

8. The Distribution of R-Multiples

For trading-system research, R-multiple is often a particularly useful unit because it normalizes outcomes relative to the initial risk.

A sequence such as:

-1R

-1R

+0.5R

+1R

+4R

can be compared across trades even when position sizes or account values differ.

Now consider two systems:

System A: mean +0.30R, median +0.25R.

System B: mean +0.30R, median -0.20R.

The equal mean does not imply equal experience or equal distribution.

System B may depend much more heavily on a small number of large winners. That can have major implications for losing streaks, drawdowns, Sample Size requirements, and the trader's ability to remain disciplined while waiting for the distribution to express itself.

9. Why Win Rate Alone Is Not a Distribution

A common mistake is to summarize a strategy with its win rate and stop there.

A 70% win rate does not tell us how large the winners and losers are.

A 40% win rate does not tell us that the strategy is necessarily poor.

Two strategies can both win 50% of their trades while having completely different distributions of outcomes.

For example:

Strategy A: winners average +0.8R; losers average -0.8R.

Strategy B: winners average +2.5R; losers average -1R.

The same win rate can coexist with very different mean outcomes.

The distribution therefore requires at least two dimensions of thought: how often outcomes occur and how large those outcomes are.

10. The Distribution Behind Expectancy

Expectancy is closely related to the distribution of trade outcomes. A simplified formulation is:

Expectancy = (Win Rate × Average Win) − (Loss Rate × Average Loss)

This shows why frequency and magnitude cannot be separated.

A strategy with a lower win rate can still have positive Expectancy if its winners are sufficiently larger than its losses. A high win rate can coexist with negative Expectancy if occasional losses are sufficiently large.

But even Expectancy is a summary statistic. It does not describe the full distribution.

Two strategies can have the same Expectancy while one relies on frequent moderate outcomes and the other relies on rare extreme outcomes.

This is why the Expert trader treats Expectancy as one component of the distributional picture, not as a complete description of strategy behavior.

11. Distribution and Drawdown

Drawdown is not independent of the distribution of trade outcomes.

If a strategy depends on rare large winners, it may experience long periods in which those winners do not occur. The absence of the positive tail can therefore create substantial temporary drawdowns even when the long-run sample remains profitable.

Conversely, a strategy with many small wins and occasional large losses may produce a smooth-looking equity curve for long periods before a tail event creates a severe drawdown.

The trader must therefore ask what kind of distribution is compatible with the drawdown profile.

A distribution with negative skew can feel deceptively stable before a large loss. A positively skewed distribution can feel frustrating because many ordinary trades may be small losses or small gains before a large winner appears.

The psychological experience of the strategy is partly a consequence of its distribution.

12. Sample Size Changes the Meaning of the Distribution

A distribution observed in 20 trades is not equivalent to a distribution observed in 2,000 trades.

With a small Sample Size, the observed mean and median can be heavily influenced by a few observations. The apparent shape of the distribution may also be unstable.

For example, a strategy may show a +1.0R mean after 20 trades because one +15R trade occurred. After 500 additional trades, that single observation may become much less dominant.

This does not mean the +15R trade was irrelevant. It means its relative influence on the summary statistics changes as more evidence accumulates.

Sample Size therefore belongs inside the interpretation of every descriptive statistic. A number without its sample context is incomplete information.

13. Aggregation Can Hide the Distribution

Suppose a strategy has a +0.25R average across 1,000 trades.

That single number may hide important conditional structures.

Breakout setups may have +0.60R expectancy.

Mean-reversion setups may have -0.05R.

High-volatility conditions may produce +0.70R.

Low-volatility conditions may produce +0.02R.

The aggregate mean is still mathematically correct. It is simply not the whole story.

A professional researcher therefore moves between aggregate and segmented distributions. The purpose is not to search endlessly for a favorable subset. The purpose is to understand whether the strategy's behavior changes systematically under identifiable conditions.

14. When Mean and Median Give a False Sense of Stability

Imagine two datasets with the same mean and median.

The first has a narrow distribution. Most observations sit close to the center.

The second has a very wide distribution. The same center is produced by large positive and negative observations that largely offset each other.

The two datasets have similar centers but very different dispersion and tail behavior.

This is why central tendency must be combined with measures of spread and distribution shape. E-003 will examine variance and standard deviation in detail, but the conceptual point belongs here:

A center without information about dispersion is incomplete.

15. Practical Interpretation Framework

When reviewing a trading distribution, work through the following sequence:

Identify the variable being measured. Do not mix P&L, percentage return, and R-multiple without a clear reason.

Calculate the mean.

Calculate the median.

Compare the two.

Inspect the minimum and maximum values.

Examine how concentrated or dispersed the observations are.

Look for asymmetry and extreme tails.

Investigate outliers rather than automatically deleting them.

Segment the data when a plausible trading condition could affect the distribution.

Interpret every summary statistic together with its Sample Size and measurement unit.

The key habit is to move from a number to a structure.

16. Case Study — Same Mean, Different Trading Reality

Two strategies each have 400 historical trades.

Both report a mean outcome of +0.25R per trade.

Strategy A has a median of +0.20R. Most outcomes are relatively close to the center, and extreme observations are uncommon.

Strategy B has a median of -0.10R. Most trades are small losses or small gains, while a limited number of large winners contribute a substantial part of the total result.

If the trader sees only the mean, the systems appear similar.

If the trader examines the distributions, the difference becomes obvious.

Strategy B may require the trader to tolerate more ordinary losing outcomes before the positive tail appears. Its drawdown and psychological profile may therefore differ substantially from Strategy A even though the mean is identical.

Neither distribution is automatically preferable. The relevant question is whether the trader understands the mechanism, has tested it over an appropriate Sample Size, and has a Risk Management framework capable of surviving the observed behavior.

This is the central lesson of E-002: the mean describes one property of the sample. The distribution explains the environment in which that property exists.

17. Practical Application — Distribution Audit

Select one strategy and export a clean sample of historical R-multiple results.

Calculate:

Mean

Median

Minimum

Maximum

Number of observations

Win rate

Average win

Average loss

Then create an ordered list of all R-multiple outcomes and inspect the distribution.

Answer the following:

Is the mean above, below, or close to the median?

What does that relationship suggest about asymmetry?

How much of the total result comes from the largest five winners?

How much changes if the largest winner is temporarily removed?

Are the extreme observations normal expressions of the strategy or unusual events?

Does the strategy appear to depend on a positive or negative tail?

What does the distribution imply about the trader's likely losing periods?

Would the same conclusion be reasonable if the Sample Size were only 20 trades?

Finally, write a three-sentence summary:

Sentence 1: describe the center of the distribution.

Sentence 2: describe the shape and major sources of asymmetry.

Sentence 3: state one conclusion that the data do not justify.

18. Expert Exercise — Mean-Median Stress Test

Construct two artificial trading datasets with exactly the same mean.

Make the first dataset relatively concentrated around its center.

Make the second dataset strongly dependent on a small number of extreme winners.

Keep the means equal.

Then compare:

Median

Minimum

Maximum

Win rate

Largest winner as a percentage of total profit

Largest loss

Now imagine that both systems are traded with identical Position Sizing.

Write down how their likely trading experience could differ even though the mean is identical.

The goal is to demonstrate that the mean does not define the distribution and that identical average outcomes do not imply identical risk, drawdown, or psychological requirements.

Knowledge Check

What does the arithmetic mean measure?

How is the median determined?

Why is the median less sensitive to extreme observations than the mean?

Why can a strategy have a positive mean while its median is negative?

What does a large difference between mean and median potentially tell a trader?

Why is distribution more informative than a single summary statistic?

What is the difference between a symmetric and a skewed distribution?

Why is skewness not automatically good or bad?

Why can two strategies with the same win rate have very different performance?

Why can two strategies with the same mean have very different trading experiences?

How does R-multiple help compare trade outcomes?

Why is Expectancy still only a summary of the underlying distribution?

How can the distribution of outcomes influence Drawdown?

Why does Sample Size change how much confidence we should place in an observed distribution?

Why can aggregation hide important conditional behavior?

Why should an outlier be investigated rather than automatically removed?

What can the relationship between mean and median reveal about a strategy?

Why is a center without information about dispersion incomplete?

What does it mean to say that a statistical number must be interpreted in context?

Key Principle: The mean tells you where the arithmetic center of the observed outcomes lies. The median tells you where the middle observation lies. The distribution tells you how the outcomes are actually arranged around those measures. A professional trader does not confuse any one of these summaries with the full behavior of the strategy.

Copying, reproducing, adapting, distributing, or publicly displaying any part or all of the Academy’s textual content without the author’s prior written consent is prohibited.