What Is Kurtosis? Types and Excess Kurtosis Explained

Search Fintrev

Search our guides, reviews and glossary.

Kurtosis is a statistical measure of how much of a distribution’s variance comes from its tails, meaning observations far from the mean, compared with a normal distribution. A normal distribution has a kurtosis of 3. Higher values mean more of the spread is driven by rare, extreme observations. Lower values mean the data rarely strays far from its average. In finance, kurtosis is applied to return distributions to judge tail risk: the chance that an investment produces moves much larger than its standard deviation alone would suggest.

What Kurtosis Measures

Kurtosis answers a specific question: once you account for how spread out a dataset is, how much of that spread comes from a few observations far from the center? Two return series can have identical averages and identical standard deviations and still behave very differently. One gets its variability from many moderate moves. The other is calm most of the time and occasionally produces a huge move. Standard deviation cannot tell them apart. Kurtosis can.

The name comes from the Greek kyrtos, meaning curved or arched. Karl Pearson introduced the term and the labels mesokurtic, leptokurtic and platykurtic in the early 1900s. Because of those origins, many textbooks long described kurtosis as a measure of how “peaked” a distribution is. That description is misleading. A distribution can have a sharp peak and low kurtosis, or a flat-looking center and high kurtosis. The statistic is dominated by observations in the tails, because each deviation from the mean is raised to the fourth power. A point three standard deviations from the mean contributes 81 units to the calculation, while a point one standard deviation away contributes 1. Observations near the center contribute almost nothing.

The “peaked” appearance of a leptokurtic histogram is a side effect. When a few outliers stretch the horizontal axis, the bulk of the data looks squeezed into a narrow spike in the middle. The outliers are what the number is measuring.

A useful way to read kurtosis: it is the average of the fourth power of standardized values, so it summarizes how often values land far out and how far out they land, combined into one figure. It is not a probability. A kurtosis of 6 does not say extreme events happen twice as often as under a normal distribution. It says the tails contribute much more to the distribution’s spread than normal tails would.

Kurtosis Formula

For a full population (or when you treat a sample as the whole dataset), kurtosis is the fourth central moment divided by the square of the second central moment:

Kurtosis = m₄ / m₂²

where:

• m₂ = (1/n) × Σ(xᵢ − x̄)², the average squared deviation from the mean (the population variance)
• m₄ = (1/n) × Σ(xᵢ − x̄)⁴, the average deviation raised to the fourth power
• n = number of observations, x̄ = the mean

Dividing by m₂² makes the result unit-free. Doubling every value, or switching from percent to decimal returns, leaves kurtosis unchanged. That is why it describes shape rather than size.

Excess kurtosis = m₄ / m₂² − 3

Subtracting 3 recenters the scale so a normal distribution equals 0.

The sample-adjusted formula

The simple moment ratio is biased in small samples. Spreadsheets apply a correction that estimates the excess kurtosis of the population the sample came from:

Sample excess kurtosis = [n(n+1) / ((n−1)(n−2)(n−3))] × Σ((xᵢ − x̄) / s)⁴ − 3(n−1)² / ((n−2)(n−3))

Here s is the sample standard deviation (calculated with n − 1 in the denominator). The formula requires at least four observations. For large samples, the correction terms shrink and the result converges toward the simple moment version.

Calculating kurtosis in spreadsheets and software

In Excel and Google Sheets, =KURT(A1:A10) returns the sample-adjusted excess kurtosis shown above, so a normal-looking dataset gives a value near 0, not near 3. For the values 4, 5, 6, 3, 4, 5, 6, 7, 5 and 8, the function returns about −0.15, a slightly platykurtic result.

Other tools use different defaults. In Python, the pandas .kurt() method returns sample-adjusted excess kurtosis, while scipy.stats.kurtosis returns excess kurtosis without the small-sample correction unless you set bias=False, and returns raw kurtosis if you set fisher=False. Before comparing figures from two sources, confirm which version each one reports. The gap can be large, as the next example shows.

Worked Example: Same Volatility, Different Tails

The following are hypothetical monthly returns for two portfolios, used only to show the mechanics. Both average 0% per month.

Portfolio A: 1, −1, 1, −1, 1, −1, 1, −1, 6, −6 (percent)

Portfolio B: 3, −3, 3, −3, 2, −2, 3, −3, 2, −2 (percent)

Step 1: Deviations from the mean. Because both means are 0, each return is its own deviation.

Step 2: Second moment (m₂).

• Portfolio A: eight months contribute 1² = 1 each, and the two outlier months contribute 6² = 36 each. Sum = 8 + 72 = 80. m₂ = 80 / 10 = 8.
• Portfolio B: six months contribute 3² = 9 each, four contribute 2² = 4 each. Sum = 54 + 16 = 70. m₂ = 70 / 10 = 7.

The variances are close: a standard deviation of about 2.83% for A and 2.65% for B.

Step 3: Fourth moment (m₄).

• Portfolio A: eight months contribute 1⁴ = 1 each, the two outliers contribute 6⁴ = 1,296 each. Sum = 8 + 2,592 = 2,600. m₄ = 260.
• Portfolio B: six months contribute 3⁴ = 81 each, four contribute 2⁴ = 16 each. Sum = 486 + 64 = 550. m₄ = 55.

Step 4: Kurtosis.

• Portfolio A: 260 / 8² = 260 / 64 = 4.06, excess kurtosis +1.06 (leptokurtic)
• Portfolio B: 55 / 7² = 55 / 49 = 1.12, excess kurtosis −1.88 (platykurtic)

The two portfolios have nearly the same volatility, yet A’s risk sits in two large months while B’s sits in a steady band of moderate moves. In A, the two outlier months supply 2,592 of the 2,600 total in the fourth-moment sum, almost all of the result. That concentration is what kurtosis captures.

Step 5: The sample-adjusted figure. Entering Portfolio A into Excel’s KURT function gives about 2.84, not 1.06. With only 10 observations, the correction is large. For Portfolio B, KURT gives about −2.36. Neither figure is wrong; they answer different questions (describing these 10 numbers versus estimating the population they came from). Ten data points are also far too few for a reliable estimate in practice, a point covered under limitations.

Types of Kurtosis

All three categories are defined relative to the normal distribution.

Mesokurtic: Kurtosis About 3 (Excess About 0)

A mesokurtic distribution has tails similar in weight to a normal distribution. The normal distribution itself is the standard example, and the binomial distribution with a probability near 0.5 and many trials is close to mesokurtic. For returns, a mesokurtic shape means a normal model would describe the frequency and size of extreme moves reasonably well. It does not mean the investment has moderate risk: a mesokurtic asset can be extremely volatile, with its large moves simply following a bell-curve pattern.

Leptokurtic: Kurtosis Above 3 (Positive Excess)

A leptokurtic distribution has fat tails: extreme values contribute more to its spread than they would in a normal distribution. The prefix lepto means slender, a reference to the narrow-looking center that appears when outliers stretch the chart. Examples include the Laplace distribution (excess kurtosis of 3) and Student’s t-distribution, whose excess kurtosis is 6 / (ν − 4) for degrees of freedom ν above 4. With 4 or fewer degrees of freedom, the t-distribution’s kurtosis is infinite.

For an investor, positive excess kurtosis means outsized moves, in either direction, have appeared more often than a normal model would predict. The center may also hold many small moves, which can make the asset look calm most of the time.

Platykurtic: Kurtosis Below 3 (Negative Excess)

A platykurtic distribution has thin tails. Platy means broad or flat. Values cluster within a bounded range, and outliers are scarce. The uniform distribution, where every value in a range is equally likely, has an excess kurtosis of −1.2. The lowest possible kurtosis is 1 (excess −2), which occurs when a variable takes only two values with equal probability, such as a fair coin scored as +1 and −1. Portfolio B in the example sits close to that floor because its returns bounce between a narrow set of values.

Reference values

Raw Kurtosis vs. Excess Kurtosis

The most common source of confusion is not the concept but the convention. “Kurtosis” may refer to raw kurtosis (normal = 3), also called Pearson kurtosis, or to excess kurtosis (normal = 0), sometimes called Fisher kurtosis. Many software packages and data providers label excess kurtosis simply as “kurtosis.”

Some practical rules for reading a reported figure:

• A negative number is always excess kurtosis. Raw kurtosis cannot fall below 1, so a value such as −0.204 must be on the excess scale. It describes a distribution with slightly thinner tails than normal, close enough to 0 that, in a modest sample, it is consistent with roughly normal tails.
• A value between 0 and 1 must also be excess kurtosis, for the same reason.
• A value near 3 is ambiguous. It could be raw kurtosis for a normal-looking dataset or excess kurtosis for a strongly fat-tailed one. Check the source’s definition or the function used.
• To tell fat tails from thin tails, convert to excess kurtosis: positive means fat-tailed, negative means thin-tailed.

There is no single “standard measurement” that every source uses, which is why stating the convention next to the number matters.

Kurtosis vs. Skewness

Skewness and kurtosis are both shape statistics, but they answer different questions. Skewness (the third standardized moment) measures asymmetry: whether one tail is longer or heavier than the other. Kurtosis measures the combined weight of both tails, regardless of direction.

• Zero skewness: the left and right sides mirror each other.
• Positive skewness: the right tail is longer, so large gains are more pronounced than large losses.
• Negative skewness: the left tail is longer, so large losses dominate.

If one tail carries more weight than the other, that is a skewness finding, not a kurtosis finding. A distribution can be perfectly symmetric (zero skewness) with very high kurtosis, as Portfolio A in the example shows: its two outliers are +6% and −6%. A distribution can also be strongly skewed with low kurtosis.

The two measures are linked by a mathematical constraint: kurtosis is always at least skewness squared plus 1. A heavily skewed distribution must therefore have at least some tail weight. For investors, reading them together is more informative than either alone. Positive excess kurtosis with negative skewness points to rare but severe losses, a profile associated with strategies that collect small, steady gains and occasionally suffer a large drawdown, such as selling out-of-the-money options. Positive excess kurtosis with positive skewness points to rare large gains, the shape of a lottery-like payoff.

Kurtosis vs. Volatility and Other Risk Measures

Kurtosis is often described loosely as a gauge of “volatility risk,” but it measures something volatility does not.

Standard deviation measures the size of typical moves. Kurtosis is scale-free and measures the shape of the distribution after that size is accounted for. A low-volatility bond fund and a high-volatility stock could have the same kurtosis. A low-volatility asset can also have high kurtosis if it is usually quiet and occasionally jumps.

Beta measures how much an asset moves relative to a market benchmark. It is a regression slope and says nothing about the shape of the asset’s own return distribution.

Alpha measures return above what a benchmark or model would predict. It is a performance figure, not a distribution-shape figure.

R-squared measures how much of a fund’s movement a benchmark explains. It is a goodness-of-fit statistic from regression.

The Sharpe ratio divides excess return by standard deviation. Because it uses only the first two moments, two strategies with the same Sharpe ratio can carry very different tail risk. A strategy with high kurtosis and negative skewness can show an attractive Sharpe ratio for years before a tail event arrives. This is one of the main reasons analysts report skewness and kurtosis alongside Sharpe ratios.

How Kurtosis Is Used in Finance

Detecting fat tails in returns

The classic finding in empirical finance is that asset returns, particularly over short horizons such as daily data, tend to show positive excess kurtosis. Benoit Mandelbrot documented fat tails in cotton price changes in 1963, and Eugene Fama reported similar patterns in stock returns in the mid-1960s. The October 19, 1987, crash, when the Dow Jones Industrial Average fell 22.6% in a single session, is often cited as an event that a normal model calibrated to typical daily volatility would treat as virtually impossible. Measuring kurtosis on a return series is a direct way to check how far it departs from the normal assumption.

Risk models and value at risk

Many models assume normally distributed returns because the math is convenient. When actual returns are leptokurtic, a normal model understates the probability of large losses, so a value-at-risk (VaR) figure built on it can be too optimistic at high confidence levels such as 99%. Analysts respond in several ways: using fat-tailed distributions such as Student’s t, using historical simulation instead of a parametric formula, or applying the Cornish-Fisher expansion, which adjusts VaR using the measured skewness and excess kurtosis.

Testing for normality

The Jarque-Bera test checks whether a sample’s skewness and kurtosis are consistent with a normal distribution. Its statistic is JB = (n / 6) × [S² + (K − 3)² / 4], where S is skewness and K is raw kurtosis. Large values indicate a departure from normality. Because the statistic grows with sample size, even mild fat tails become statistically significant in long return histories.

Evaluating strategies and funds

Comparing the kurtosis of fund or strategy returns helps reveal risks that average returns and volatility hide. Strategies built on writing options, carry trades or leverage can produce smooth monthly results punctuated by sharp losses, a pattern that shows up as high kurtosis and negative skewness. Investors building a portfolio meant to limit large drawdowns may prefer holdings with lower kurtosis, while those willing to accept occasional large swings for upside potential may accept higher kurtosis.

Options pricing

The Black-Scholes model assumes lognormal prices, which implies thinner tails than markets often display. Option markets price this in: out-of-the-money options frequently trade at higher implied volatilities than at-the-money options, producing the volatility smile or skew. Part of that pattern reflects the market assigning more probability to extreme moves than a normal model would.

Is High Kurtosis Good or Bad?

High kurtosis is neither good nor bad on its own. It indicates that extreme outcomes contribute heavily to an investment’s spread of returns, and what that means depends on which tail is heavy and on the investor’s tolerance for large surprises.

Combined with positive skewness, high kurtosis can describe an asset whose rare big moves are mostly gains. Combined with negative skewness, it describes an asset exposed to rare, severe losses. Even with zero skewness, high kurtosis matters for anyone using leverage, holding concentrated positions or facing margin calls, because a single extreme move can force a sale before any recovery. Low kurtosis implies fewer surprises relative to the asset’s volatility, but it does not mean the asset is low-risk: a platykurtic investment with a high standard deviation can still lose a lot.

Limitations of Kurtosis

It is highly sensitive to outliers. Because deviations are raised to the fourth power, one or two observations can dominate the result, as Portfolio A shows. Adding or removing a single crash month can change a fund’s measured kurtosis dramatically.

It needs large samples. Kurtosis estimates are noisy, and small samples can produce very different figures for the same underlying process. The gap between 1.06 and 2.84 for the same 10 numbers shows how much the small-sample correction alone can move the result. Short histories also may not contain any tail event at all, producing a reassuringly low kurtosis that reflects luck rather than safety.

It depends on the period and frequency. A calm sample window and a crisis window give very different readings. Measured kurtosis also tends to fall as returns are aggregated over longer horizons, so daily, weekly and monthly kurtosis for the same asset are not directly comparable.

It does not separate frequency from magnitude. A high figure can come from a few very large outliers or from more numerous moderately large ones. The statistic blends both, so a histogram or a count of large moves is a useful companion.

It does not indicate direction. Kurtosis treats large gains and large losses identically. Skewness or downside-specific measures are needed to know which tail is the concern.

It can be undefined. Some fat-tailed distributions have infinite kurtosis. If returns follow such a distribution, a sample kurtosis will exist but will not settle on a stable value as more data arrive.

It describes the past. Measured kurtosis reflects the historical sample. Market structure, leverage and regime changes can produce tail events of a size never seen in the data used.

FAQs

The Bottom Line

Kurtosis matters to investors and analysts who rely on volatility, Sharpe ratios or value at risk, because those figures can hide how often extreme moves occur. Check whether a reported figure is raw or excess kurtosis, read it alongside skewness to see which tail is heavy, and treat estimates from short or calm histories with caution, since a single outlier can shift the result dramatically.