Skewness and Kurtosis

ID: E-007

Introduction

Mean and standard deviation describe the center and dispersion of a distribution, but they do not fully describe its shape. Two return distributions can have the same mean and standard deviation while behaving very differently in their tails.

Skewness and kurtosis add two important dimensions to that picture. Skewness describes asymmetry; kurtosis is commonly used to describe the relative weight or propensity of extreme observations in a distribution. For traders, this matters because trading results are often dominated not by ordinary observations but by asymmetric outcomes and rare, large events.

The Expert objective is not to memorize threshold values for these statistics. It is to understand what shape information they add, how estimation becomes unstable in small samples, and how distribution shape changes the interpretation of expectancy, drawdown, risk, and strategy performance.

1. Why Distribution Shape Matters

Consider two strategies with identical average R-Multiple and identical standard deviation. One may generate frequent small wins and occasional very large losses. Another may produce frequent small losses and occasional very large wins.

Their mean and dispersion can be similar while their risk profile is completely different.

Shape statistics help reveal this difference. A trader who evaluates only average return and standard deviation can miss where the probability mass is concentrated and where the extreme outcomes sit.

This is particularly important for strategies whose edge depends on a small number of outsized winners or whose risk is concentrated in rare adverse events.

2. Skewness: Asymmetry Around the Center

Skewness describes the asymmetry of a distribution around its mean.

A positively skewed distribution has a relatively long or influential right tail. Large positive observations occur less frequently but can have substantial impact on the average.

A negatively skewed distribution has a relatively long or influential left tail. Large negative observations can dominate the risk profile.

A distribution with little asymmetry has skewness closer to zero, although zero skewness does not mean the distribution is normal.

For a trader, skewness answers an important question: are extreme outcomes more influential on the winning side or on the losing side?

3. Positive Skew in Trading

Many trend-following and breakout approaches can exhibit positive skew: numerous small losses or modest outcomes may be followed by a smaller number of very large winners.

This structure can make the median trade look unimpressive while the mean remains positive because a limited number of large winners contribute disproportionately to total expectancy.

Positive skew can therefore be compatible with a strategy that experiences long periods of frustration before a few outsized trades materially improve performance.

The existence of positive skew does not prove that such a strategy has an edge. The frequency, magnitude, stability, and transaction costs of the tail winners still have to be measured.

4. Negative Skew in Trading

Short-volatility, mean-reversion, or tightly profit-targeted approaches can sometimes display negative skew: many small wins are accompanied by occasional large losses.

Such a distribution can produce a high win rate while retaining significant tail risk.

This is one reason win rate alone is an incomplete description of a trading system. A strategy can win frequently and still have an unfavorable payoff structure if rare losses are sufficiently large.

The trader should therefore inspect the full distribution of R-Multiples, not only the percentage of winning trades.

5. Skewness Is Not a 'Good' or 'Bad' Number

Skewness has no universal trading preference.

Positive skew may be desirable for one strategy structure, while another strategy may intentionally accept negative skew in exchange for other characteristics. The relevant question is how the distribution interacts with the trader's risk limits, capital, holding period, and objectives.

Similarly, a skewness estimate near zero does not automatically imply a balanced or safe distribution.

Skewness is descriptive. Its trading meaning emerges only when combined with expectancy, dispersion, drawdown, position sizing, and the strategy's mechanism.

6. Kurtosis: A Careful Definition

Kurtosis is frequently described informally as a measure of 'tailedness' or the tendency of a distribution to produce extreme observations. More precisely, classical kurtosis is based on the fourth standardized moment and captures aspects of tail weight and concentration relative to a reference distribution.

Excess kurtosis subtracts 3 from classical kurtosis, so a normal distribution has excess kurtosis of zero.

Positive excess kurtosis is commonly associated with heavier tails and a greater propensity for extreme observations than a normal distribution. Negative excess kurtosis is associated with lighter tails in the relevant moment-based sense.

However, kurtosis is not simply a direct count of how many outliers exist. It is a moment-based statistic that can be strongly influenced by a small number of extreme observations.

7. Why Kurtosis Matters to Traders

A strategy with heavy tails can produce more extreme outcomes than a normal approximation suggests.

This matters for stop-outs, gap risk, portfolio shocks, and drawdown planning. If the distribution has more extreme observations than expected under a normal model, risk estimates based solely on normal assumptions can be too optimistic.

High kurtosis can also make sample estimates unstable. One extraordinary market event can materially change the measured value.

The practical lesson is not 'high kurtosis means danger.' The lesson is that tail behavior deserves explicit attention rather than being hidden inside a single volatility number.

8. Skewness, Kurtosis and the Normal Distribution

The normal distribution has zero skewness and zero excess kurtosis.

This makes the normal distribution a useful reference point, but financial returns frequently deviate from it.

Returns can be asymmetric, heavy-tailed, clustered in volatility, and dependent on market regime. A normal model may therefore underestimate the frequency or impact of extreme observations.

An Expert trader uses normality as a benchmark for comparison, not as an assumption that every market series must satisfy.

9. Why Mean and Standard Deviation Are Not Enough

Mean and standard deviation summarize location and dispersion. They do not tell us whether the distribution is symmetric or whether the tails are unusually influential.

Two strategies can share the same mean and standard deviation while having opposite skewness.

They can also have similar skewness while differing substantially in kurtosis.

This means a serious distribution review should examine at least four dimensions: center, dispersion, asymmetry, and tail behavior.

In trading terms: expectancy tells us about average outcome, standard deviation about dispersion, skewness about asymmetry, and kurtosis about the behavior of extreme observations.

10. R-Multiple Distributions

R-Multiples are especially suitable for studying distribution shape because they normalize trade outcomes by initial risk.

Imagine a strategy with many outcomes between −0.5R and +1R, plus a few +6R and +10R winners. The large winners may strongly influence mean expectancy and produce positive skew.

Another strategy may generate many +0.5R winners and occasional −6R losses. Its mean and standard deviation can be similar under some samples, but the negative tail creates a very different experience of risk.

Examining the histogram, median, mean, skewness, and tail quantiles together gives a much more complete picture than win rate alone.

11. Skewness and Expectancy

Expectancy is an average outcome. Skewness describes asymmetry around the distribution's center. They are related through the distribution but answer different questions.

A positive-expectancy strategy can have negative skew, and a positive-skew strategy can still have negative expectancy.

Therefore the trader should not infer edge from skewness alone.

The useful analysis is to ask how much of total expectancy comes from the tail. If removing the top few winners destroys expectancy, the strategy may be highly dependent on rare outcomes. That does not automatically invalidate the strategy, but it changes how its sample size, patience requirements, and drawdown expectations should be evaluated.

12. Tail Dependence and Drawdown

Skewness and kurtosis can help describe individual strategy distributions, but they do not fully capture portfolio tail dependence.

Two strategies may each look positively skewed in isolation while suffering large losses during the same market shock.

Conversely, one strategy's negative tail may be partially offset by another strategy under the same conditions.

This is why distribution shape should eventually be connected with the correlation and covariance analysis from E-004 and E-005.

The trader's real concern is not merely whether each strategy has an attractive distribution, but how those distributions interact when capital is deployed simultaneously.

13. Sample Size and Estimation Error

Higher moments are harder to estimate reliably than the mean. Skewness and especially kurtosis can be highly sensitive to sample size and extreme observations.

A few hundred trades may provide useful descriptive information, but the estimated tail shape can still change materially when additional extreme observations arrive.

Small samples create a dangerous illusion of precision. A platform may display skewness to several decimal places even though the underlying estimate is statistically unstable.

An Expert trader therefore records Sample Size and treats higher-moment estimates as uncertain measurements rather than immutable properties of the strategy.

14. Outliers: Signal or Contamination?

An extreme observation can be one of the most important observations in the dataset or it can be a data-quality error.

A genuine market shock should generally not be deleted merely because it makes kurtosis or skewness inconvenient. It may represent exactly the event the strategy must survive.

An incorrect price print, duplicate trade, execution error, or corrupted data point is different. Such observations should be investigated and handled according to a documented data-quality rule.

The objective is not to create a prettier distribution. It is to separate real tail behavior from measurement contamination.

15. Robust Alternatives and Quantiles

Because skewness and kurtosis can be sensitive to extreme observations, traders should often complement them with robust statistics and quantiles.

Median, interquartile range, percentile outcomes, and explicit tail quantiles can describe the distribution without relying exclusively on high-order moments.

For example, the 95th or 99th percentile of adverse R-Multiple outcomes may be more intuitive for a risk discussion than a single kurtosis coefficient.

These measures do not replace skewness and kurtosis. They provide additional perspectives that are often easier to connect to actual trading decisions.

16. Expert Workflow — Distribution Shape Audit

1. Build the trade or return series using a clearly defined observation unit.

2. Calculate mean, median, standard deviation, and relevant quantiles.

3. Calculate skewness and excess kurtosis using a documented convention.

4. Inspect the histogram or empirical distribution rather than trusting the summary statistics alone.

5. Identify the largest positive and negative observations and verify their data quality.

6. Measure how much expectancy is contributed by the extreme winners and losers.

7. Compare the distribution across market regimes and rolling samples.

8. Connect the shape to Drawdown and Volatility behavior.

9. Compare strategy distributions with one another and then examine their correlation or covariance.

10. Document Sample Size and the uncertainty of the higher-moment estimates.

17. Expert Case Study — Two Strategies, Two Very Different Tails

Strategy A produces 70% winning trades, usually between +0.2R and +0.8R, but occasionally loses −5R.

Strategy B produces only 42% winning trades, often loses between −0.3R and −0.8R, but occasionally produces +7R or +10R.

Looking only at win rate, the two systems tell very different stories. Looking at distribution shape, the difference becomes clearer: Strategy A may carry negative skew and concentrated downside tails, while Strategy B may exhibit positive skew and dependence on occasional large winners.

Neither description determines whether either strategy is profitable. The complete analysis requires expectancy, dispersion, sample size, drawdown, execution costs, and stability.

The expert lesson is that a strategy's personality is contained in its entire outcome distribution, not in one headline statistic.

18. Common Trading Mistakes

Treating positive skew as automatically good or negative skew as automatically bad.

Assuming high kurtosis is simply a synonym for 'many outliers.'

Using normal-distribution probabilities without checking the actual distribution.

Reporting skewness and kurtosis to excessive precision from a small sample.

Deleting genuine market shocks because they make the distribution look unattractive.

Judging a strategy from win rate while ignoring the shape and magnitude of losing trades.

Assuming a positive expectancy guarantees positive skew.

Assuming positive skew guarantees positive expectancy.

Ignoring how individual strategy tails interact at the portfolio level.

19. Knowledge Check

1. What does skewness describe?

2. What is the practical difference between positive and negative skew?

3. Why can a high win-rate strategy still have substantial tail risk?

4. What is classical kurtosis, and what does excess kurtosis change?

5. Why is the normal distribution useful as a reference but dangerous as an unquestioned assumption?

6. Why are higher moments more sensitive to sample size than the mean?

7. Why should genuine market shocks generally be retained in a strategy dataset?

8. How can quantiles complement skewness and kurtosis?

9. Why can two strategies with similar mean and standard deviation still have very different risk profiles?

10. How can skewness affect the interpretation of Expectancy?

11. Why does positive skew not prove a Strategy Edge?

12. Why should distribution shape eventually be connected with correlation and covariance?

13. What does a large positive tail contribution to expectancy tell you about the strategy?

14. Why can a platform's many-decimal skewness value create false confidence?

15. What should an Expert trader include in a distribution-shape audit?

Expert answer principle: describe the shape before judging the strategy, distinguish tail structure from profitability, and treat higher-moment estimates as uncertain sample-based measurements.

Conclusion

Skewness and kurtosis extend statistical analysis beyond center and dispersion. They help reveal whether outcomes are asymmetric and how strongly extreme observations influence the distribution.

For trading, this matters because many strategies are not well represented by a normal distribution. A small number of large winners or losses can dominate expectancy, drawdown, and the trader's actual experience of risk.

The Expert approach is therefore to examine the whole distribution: mean, median, dispersion, skewness, tail behavior, quantiles, and sample size. These statistics do not tell the trader what to trade. They tell the trader what kind of outcome distribution the trading process has historically produced.

Once that distribution is understood, the next step is to connect it to position sizing, portfolio interaction, and risk management. That is where descriptive statistics become practical trading intelligence.

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.