What Is Skewness? Right-Skewed vs. Left-Skewed Distributions

Search Fintrev

Search our guides, reviews and glossary.

Skewness is a statistical measure of how asymmetric a distribution is around its mean. A distribution with zero skewness is symmetric: values spread out the same way on both sides of the center. A right-skewed (positively skewed) distribution has a longer or heavier tail on the right, so a few unusually large values stretch it toward higher numbers. A left-skewed (negatively skewed) distribution has its longer or heavier tail on the left, where a few unusually small or very negative values stretch it toward lower numbers.

The direction of the skew is named after the tail, not after the bulk of the data. “Skewed left” means the tail points left while most observations cluster on the right. That naming catches many readers, and it matters in trading: a left-skewed return series is one where most outcomes are modest gains and the rare outcomes are large losses.

What skewness measures

Mean and standard deviation describe the center of a data set and how widely it spreads. Neither says anything about shape. Two return series can share the same average and the same volatility and still behave very differently: one might produce steady small gains punctuated by an occasional crash, the other steady small losses punctuated by an occasional windfall. Skewness is the statistic that tells those two apart.

Skewness reflects where activity is concentrated. When most observations bunch in one range and a thinner set of observations trails off in the other direction, the distribution is skewed toward that trailing side. The trailing values do not have to be numerous; a handful of extreme points can create strong skew because they sit far from everything else.

The benchmark for “no skew” is the normal distribution, or bell curve, where the mean, median and mode coincide and each side mirrors the other. Many statistical tools and financial models start from that benchmark, which is why departures from symmetry carry practical weight.

Right-skewed vs. left-skewed distributions

Right-skewed (positive skewness)

In a right-skewed distribution, the peak of the data sits toward the left side of the range and a long tail stretches to the right. Most values are relatively low, and a smaller number of values are much higher than the rest.

Those high values pull the average upward. The median, which only cares about the middle position in the sorted data, is barely affected by how far the top values sit. The result is the familiar pattern: mean greater than median. The mode, the most frequent value, usually sits at the peak, below both. So in a right-skewed distribution the order from left to right is typically mode, then median, then mean.

What it means in plain terms when data is skewed to the right: the typical observation is smaller than the average suggests. If someone quotes an average from right-skewed data, most observations fall below that figure.

Left-skewed (negative skewness)

A left-skewed distribution is the mirror image. The peak sits toward the right side of the range, and a long tail stretches to the left. Most values are relatively high, and a few are much lower.

The low outliers drag the mean down, so the mean falls below the median, and the typical order from left to right becomes mean, then median, then mode. The typical observation is larger than the average suggests, and the average is pulled down by a small number of poor outcomes.

A left-tailed distribution in this sense is a left-skewed one. That is different from a “left-tailed test” in hypothesis testing, which refers to where the rejection region sits, not to the shape of the data.

Zero skew and symmetry

A skewness of zero means the two sides balance. For a perfectly symmetric distribution, the mean and median are equal, and for a symmetric single-peaked distribution the mode sits in the same place.

Zero skew does not mean the data is normal. A distribution can be perfectly symmetric and still have much fatter tails than a bell curve, producing far more extreme values on both sides. How heavy the tails are is measured by kurtosis, a separate statistic. Skewness only says whether the extremes are balanced, not how extreme they are.

Reading a skewness graph

When looking at a histogram or density chart, three checks identify the direction of skew:

1. Find the tail. Whichever side has the longer, thinner stretch of observations names the skew. Tail on the right means right-skewed; tail on the left means left-skewed.
2. Find the peak. In a right-skewed chart the peak leans left; in a left-skewed chart it leans right.
3. Compare mean and median if they are marked. Mean to the right of the median points to positive skew; mean to the left points to negative skew.

If the chart shows two or more separate peaks, these checks become unreliable, because the “tail” may actually be a second cluster of data rather than a stretch of outliers.

How to measure skewness

Several formulas exist. They agree on the sign for ordinary single-peaked data, but they scale differently, so a skewness figure should always be read alongside the method that produced it.

Pearson’s first coefficient (mode skewness)

Karl Pearson proposed two simple measures that compare the mean with another measure of center.

Sk₁ = (Mean - Mode) / s

where s is the sample standard deviation. Subtracting the mode from the mean and dividing by the standard deviation produces a unit-free number. Positive values indicate right skew, negative values left skew.

This version works best when the data has one clear, strong mode. With continuous data such as returns, almost no value repeats exactly, so the mode may be poorly defined or depend on how the data is grouped into bins.

Pearson’s second coefficient (median skewness)

Sk₂ = 3 × (Mean - Median) / s

This version replaces the mode with the median and multiplies the difference by three. It is preferable when the mode is weak, ambiguous or when there are several modes, because the median is always defined. The factor of three comes from an approximate empirical relationship in moderately skewed distributions, where the gap between mean and mode tends to be about three times the gap between mean and median.

Moment coefficient of skewness

Statistical software and spreadsheets usually report the moment-based measure, often called the Fisher-Pearson coefficient. It is built from the third power of each observation’s deviation from the mean:

g₁ = m₃ / m₂^(3/2)

where:

• m₃ = the average of (x - mean)³ across all observations
• m₂ = the average of (x - mean)² across all observations (the population variance)

Cubing preserves the sign of each deviation, so values below the mean contribute negative amounts and values above contribute positive amounts. Cubing also magnifies large deviations far more than small ones. A point three standard deviations from the mean contributes 27 times as much as a point one standard deviation away. That is why skewness is driven by the tails.

Many programs apply a small-sample adjustment:

G₁ = g₁ × √[n(n - 1)] / (n - 2)

where n is the number of observations. The spreadsheet SKEW function in common tools reports this adjusted version. For large samples the two figures converge; for small samples the adjusted figure is noticeably larger in magnitude.

Interpreting the number

A frequently cited rule of thumb for the moment coefficient:

• Between -0.5 and +0.5: approximately symmetric
• Between 0.5 and 1 in either direction: moderately skewed
• Beyond 1 in either direction: highly skewed

These cutoffs are conventions, not tests. Whether a given skewness figure means anything also depends on sample size. For data that really is normally distributed, the standard error of the sample skewness is roughly √(6/n). With 10 observations that is about 0.77, so a reading of 0.6 tells you almost nothing. With 250 observations, roughly a year of daily trading data, it falls to about 0.15, and a reading of 0.6 starts to look meaningful.

Worked example: two hypothetical strategies

The following figures are invented to show the mechanics. They do not represent any real product or market.

Strategy A produces ten monthly returns: +2% in nine months and -8% in one month.

Step 1: mean. (9 × 2 + (-8)) / 10 = 10 / 10 = 1.0% per month.

Step 2: median and mode. Sorting the returns gives nine values of 2% and one of -8%. The middle values are both 2%, so the median is 2%, and the mode is also 2%.

The mean sits below the median, which already signals left skew.

Step 3: deviations from the mean. Nine months deviate by +1 point; one month deviates by -9 points.

Step 4: squared deviations. 9 × 1² + (-9)² = 9 + 81 = 90. Population variance m₂ = 90 / 10 = 9, so the population standard deviation is 3%. The sample standard deviation, dividing by n - 1 = 9, is √10 ≈ 3.16%.

Step 5: cubed deviations. 9 × 1³ + (-9)³ = 9 - 729 = -720. So m₃ = -720 / 10 = -72.

Step 6: moment coefficient. g₁ = -72 / 9^(3/2) = -72 / 27 ≈ -2.67. Applying the small-sample adjustment: √(10 × 9) / 8 ≈ 1.186, and -2.67 × 1.186 ≈ -3.16.

Step 7: Pearson’s coefficients (using the sample standard deviation of 3.16):

• Sk₁ = (1.0 - 2.0) / 3.16 ≈ -0.32
• Sk₂ = 3 × (1.0 - 2.0) / 3.16 ≈ -0.95

Every measure agrees the distribution is left-skewed, but the magnitude ranges from about -0.3 to about -3.2. The moment coefficient reacts to the single -9 point deviation being cubed; the Pearson measures only see how far the mean has moved from the center. That spread is why a skewness figure means little without knowing which formula produced it.

Strategy B is the mirror image: -2% in nine months and +8% in one. Its mean is -1.0%, its median -2%, and every skewness measure has the same size with the opposite sign, so the moment coefficient is about +2.67 (or +3.16 adjusted). Strategy B has a positive-skew profile: frequent small losses and a rare large gain.

Both strategies have the same standard deviation. A volatility-based risk measure would rate them as equally risky, yet Strategy A’s worst month is -8% while Strategy B’s worst month is -2%. Skewness is what captures that difference.

Example of skewness in a game of chance

Gambling offers a clean real-world case because the payoffs are fixed by the rules. Take a single-number bet in roulette. An American wheel has 38 pockets (1 to 36, 0 and 00); a European wheel has 37 (1 to 36 and a single 0). A straight-up bet on one number pays 35 to 1.

Suppose a player stakes $100 on one number on an American wheel:

• With probability 1/38, the ball lands on the number and the player gains $3,500 (the stake is also returned).
• With probability 37/38, the player loses the $100 stake.

The expected result per spin is (1/38 × $3,500) - (37/38 × $100) = ($3,500 - $3,700) / 38 ≈ -$5.26, an average loss of about 5.3% of the stake. On a European wheel the same calculation gives ($3,500 - $3,600) / 37 ≈ -$2.70, or about 2.7%.

Almost every spin loses the full $100. The average loss is only around $5 because a rare $3,500 win sits far out on the right. The distribution of outcomes is heavily right-skewed: using the skewness formula for a two-outcome bet, (1 - 2p) / √[p(1 - p)] with p = 1/38, the skewness is about +5.9.

The example shows something the mean alone hides: a player can face a modest average loss and still lose on the overwhelming majority of bets. Lotteries and deep out-of-the-money options purchased outright share this profile.

Why skewness matters to traders and investors

Skewness risk and the normal assumption

Standard deviation, the Sharpe ratio, many value-at-risk models and plenty of position-sizing rules treat returns as if they followed a bell curve. In a normal distribution, a move of three standard deviations or more on the downside is very rare, and upside and downside extremes are equally likely.

When the real distribution is skewed, a model built on the normal assumption misprices the extremes. If returns are negatively skewed, the model underestimates how likely and how large the big losses are. If returns are positively skewed, it underestimates the chance of big gains and may overstate the downside. The greater the skew, the larger the error. This gap is sometimes called skewness risk: the risk of meeting an extreme observation that a symmetric model treated as nearly impossible.

Skewness focuses attention on extremes rather than on the average. That matters most for anyone who cannot rely on the long run to even things out. A short-term trader, or anyone using leverage, may not survive the losing tail long enough to collect the average, and a deep drawdown in a leveraged account can end in a margin call before the expected profit arrives.

Strategy payoff profiles

Skewness is one of the clearest ways to describe what a trading approach feels like to run.

Negative-skew strategies win often and lose rarely but heavily. Common examples by design include selling options and collecting premium, mean-reversion systems that fade moves, grid systems, and strategies that add to losing positions. The Martingale strategy is an extreme case: doubling after each loss produces a long run of small wins and, eventually, one catastrophic loss. Carry trades in forex, which earn an interest differential while holding a higher-yielding currency, are often described in similar terms, since slow accumulation of interest can be erased by a sharp move in the exchange rate. A high win rate is a typical side effect of negative skew, not proof of an edge.

Positive-skew strategies lose often and win rarely but large. Trend following, breakout trading and buying options outright tend to show this profile: many small stopped-out losses and an occasional large winner that pays for them. These strategies are harder to stick with psychologically because the drawdowns are long, but the worst single loss is usually controlled.

Risk management tools change the shape of a trade’s outcome distribution. A stop-loss order is meant to cut off the left tail of an individual trade’s result. It does not remove that tail entirely, because a stop becomes a market order when triggered and can fill at a worse price during gaps or fast markets. Position sizing determines how much one trip into the left tail costs; a position size calculator converts a stop distance and risk percentage into a lot size, which limits damage per trade but does not change the skew of the strategy itself.

Skewness in markets

Broad stock indexes are widely described as negatively skewed: they tend to rise gradually and fall abruptly. The crash of October 19, 1987, when the Dow Jones Industrial Average lost about 22.6% in a single day, is the textbook example of a left-tail event that a normal model would have treated as practically impossible. After that crash, equity index options began to price out-of-the-money puts at higher implied volatility than equivalent calls, a pattern known as the volatility skew or smirk.

Individual stocks are a different case. Academic work has found evidence that single-stock returns can be positively skewed, partly because a share price cannot fall below zero but can rise many times over. Measured skewness also depends on the horizon studied: research on the skewness of stock market returns over long horizons shows the figure changes with the length of the return period. Any statement about whether “the market” is skewed is a statement about a particular asset, sample and horizon.

Investors with a preference for positive skew may accept a lower average return in exchange for a small chance of a very large payoff, which is one explanation offered for the popularity of lottery-like assets. The roulette example above shows the cost of that preference in its purest form.

Real-world examples of right and left skew

Skewness appears well beyond trading, and familiar examples help fix the direction of each type.

Right-skewed examples

• Household income. Income in the United States is a standard example. Most households earn within a moderate range, while a small number earn far more, stretching the tail to the right and pulling the mean income above the median. That is why median income is usually the preferred summary.
• Home prices. Most homes in a region sell within a band, while a few luxury properties sit far above it.
• Insurance claims. Most claims are small; a few are very large.
• Waiting times and durations. A process that usually takes a few minutes but occasionally takes hours produces a right tail.
• Lottery and roulette payoffs, as shown earlier.

Left-skewed examples

• Age at death in developed countries. Most people die at an older age, and fewer die young, so the long tail stretches toward younger ages.
• Scores on an easy exam. When most students score high, a few low scores form a left tail.
• Returns on short option positions and other premium-collecting strategies, which accumulate small gains and occasionally take large losses.
• Many broad equity index return series, as discussed above.

Roughly symmetric examples

Adult height within a single population is often cited as approximately normal. “Approximately” is the key word: real data rarely matches a bell curve exactly, and slight skew can appear in the tails even when the center of the distribution looks symmetric.

What causes skewness?

Skewness arises whenever the process that generates the data treats one direction differently from the other. Common mechanisms include:

• Natural boundaries. Values with a hard floor but no ceiling, such as prices, incomes and waiting times, cannot spread far below the floor but can spread far above it. The result is right skew. A ceiling, such as a maximum score on a test, does the opposite.
• Multiplicative growth. When values grow by percentages compounded over time, the outcomes spread out asymmetrically, producing right skew in levels even when the percentage changes are symmetric.
• Payoff design. Contracts and bets with capped gains and large potential losses (or the reverse) build skew into their outcomes. Selling insurance or options caps the gain at the premium received.
• Feedback in stress. In markets, falling prices can trigger forced selling, margin calls and stop orders, which accelerate declines. Rallies tend not to carry the same forced buying, which is one explanation for abrupt drops and gradual rises.
• Mixtures. Combining two groups with different centers can create apparent skew, or even a second peak.

Skewness is not an error. In many data sets it is a built-in feature of what is being measured, and the right response is to choose summaries and models that respect the shape rather than assuming it away.

Limitations of skewness

Sample size. Skewness estimates are unstable in small samples, and one extreme value can flip the sign. With a few dozen observations, the measured figure may say more about one unusual month than about the underlying process.

Sensitivity to outliers. The moment coefficient cubes deviations, so a single data error or one-off event can dominate it. Checking the data for errors and seeing whether the result survives removing the most extreme point is a reasonable robustness test.

Location, not quantity, of outliers. Skewness indicates on which side the extremes fall. It does not count them or measure tail thickness. Kurtosis, maximum drawdown and tail percentiles fill that gap.

Method dependence. As the worked example showed, the Pearson and moment coefficients can differ several-fold for the same data. Figures from different sources are only comparable if they use the same method and the same return frequency.

The mean-median rule has exceptions. The guideline that mean above median implies right skew works for smooth, single-peaked distributions. It can fail for discrete data, multimodal data or distributions where one tail is long and the other is heavy, so the sign of the moment coefficient and the mean-median gap can occasionally disagree.

Time variation. Skewness in financial returns is not fixed. A strategy can look symmetric through a calm period and reveal strong negative skew only when conditions change. Negative-skew strategies in particular can show a deceptively benign history until their first large loss, so past skewness is an estimate, not a guarantee.

Frequency matters. Daily, weekly and monthly returns from the same asset can show different skewness, because compounding and the clustering of large moves change the shape as the horizon lengthens.

Skewness vs. related concepts

Skewness vs. kurtosis. Skewness measures asymmetry using the third power of deviations. Kurtosis measures tail heaviness using the fourth power. A distribution can have zero skew and high kurtosis (symmetric but prone to extremes on both sides), or strong skew with ordinary kurtosis.

Skewness vs. standard deviation. Standard deviation measures spread regardless of direction. Two series with identical standard deviations can have opposite skewness, as Strategies A and B demonstrated.

Skewed data vs. biased data. In everyday language, “skewed results” can mean results distorted by a flawed sample or method. In statistics, skewness is a property of the shape of a distribution and says nothing about whether the data was collected fairly. A perfectly unbiased sample of incomes will still be right-skewed.

Return skewness vs. volatility skew. Options traders use “skew” for the pattern of implied volatility across strike prices. It is related to the market’s expectations of asymmetric moves, but it is a property of option prices, not the computed skewness of a return series.

Positive skew vs. positive returns. The sign of skewness is unrelated to whether an asset makes money. The roulette bet has strong positive skew and a negative expected return; a premium-selling strategy can have negative skew and a positive average return until its tail event arrives.

Skewness vs. hedging. Hedging changes the shape of a portfolio’s outcomes, often by trimming one tail at the cost of some return. Buying protective puts, for example, cuts the left tail and shifts a portfolio’s skew in the positive direction.

FAQs

The Bottom Line

Skewness matters to traders and investors who judge risk by averages and volatility alone, since two strategies with matching figures can carry very different worst cases. A high win rate often comes from negative skew rather than a real edge. Check which formula produced a skewness figure, make sure the sample is large enough to trust it, and pair it with kurtosis, drawdown and tail percentiles.