Skewness and kurtosis

On this page

Learning Outcome Statement

Describe, interpret, and evaluate measures of skewness and kurtosis.

On the exam: The question gives periodic return observations or summary moments, states the sample or population convention and any null or tail convention, then asks for a calculation, interpretation, or evaluation in a table, graph, fill-in, or multiple-choice shape.

Skewness tells you whether unusual outcomes stretch farther on the gain side or loss side. Kurtosis tells you how strongly extreme deviations affect the return distribution, without telling you which side is longer. You will see periodic returns, so keep their frequency and the sample or population convention fixed before calculating.

The normal benchmarks are reference points rather than pass or fail cutoffs. Cubing preserves a centered return's sign, while the fourth power makes both tails contribute positively.

For a return observation, use $r_i$, where $i$ identifies the observation. The count is $n$. For a sample, the center is $\bar{r}$ and the standard deviation is $s$, both in percentage points per month. For a population, the center is $\mu$ and the standard deviation is $\sigma$ in the same units.

MeasureNormal benchmarkForm
Skewness$0$Third central moment over standard deviation cubed
Pearson kurtosis$3$Fourth standardized moment
Excess kurtosis$0$Pearson kurtosis minus $3$

Population skewness divides the average cubed distance from $\mu$ by $\sigma^3$. Call this measure $\gamma_1$.

$$ \gamma_1 = \frac{\frac{1}{n}\sum_{i=1}^{n}(r_i-\mu)^3}{\sigma^3} $$

For sample observations, use $\bar{r}$ and $s$ instead of $\mu$ and $\sigma$. The sample standard deviation $s$ uses the $n-1$ denominator, and sample skewness uses the correction shown below. Call it $G_1$.

$$ G_1 = \frac{n}{(n-1)(n-2)} \sum_{i=1}^{n} \left(\frac{r_i-\bar{r}}{s}\right)^3 $$

Population Pearson kurtosis uses the fourth standardized moment. Call it $\beta_2$.

$$ \beta_2 = \frac{1}{n} \sum_{i=1}^{n} \left(\frac{r_i-\mu}{\sigma}\right)^4 $$

For a sample, sample excess kurtosis uses a correction involving $n$, $n-1$, $n-2$, and $n-3$. Call the result $G_2$.

Applied powerPositive centered returnNegative centered return
Third powerPositive contributionNegative contribution
Fourth powerPositive contributionPositive contribution

$$ \begin{aligned} G_2 &= \frac{n(n+1)}{(n-1)(n-2)(n-3)} \sum_{i=1}^{n} \left(\frac{r_i-\bar{r}}{s}\right)^4 \ &\quad - \frac{3(n-1)^2}{(n-2)(n-3)} \end{aligned} $$

Excess kurtosis shifts Pearson kurtosis so a normal distribution has benchmark $0$ rather than $3$.

$$ \text{Excess kurtosis} = \text{Pearson kurtosis} - 3 $$

The cube means a large positive centered return pushes skewness upward, while a large negative centered return pushes it downward. Positive values indicate a longer right tail, and negative values indicate a longer left tail. The sample correction changes the magnitude, especially when $n$ is small. Because the numerator and denominator use matching powers, the final measure is dimensionless.

Kurtosis gives no tail direction because fourth powers are positive for either sign. Pearson kurtosis above $3$, or excess kurtosis above $0$, indicates a higher fourth standardized moment than the normal benchmark. Values below those benchmarks indicate a lower one. Keep one periodicity throughout and do not silently convert monthly percentage points into annualized decimal returns. Outliers can dominate the third and fourth moments, so a descriptive magnitude is not statistical or economic significance.

When raw observations arrive, move through the same checks each time.

Loading diagram…

The sign is easier to retain as a shape. A symmetric distribution has balanced tails and skewness at the normal benchmark of zero. When the right tail stretches farther, positive skewness records that direction; when the left tail stretches farther, the value is negative. Read the plotted shape as a description of observed returns, then use the moment calculation and its stated sample or population convention to quantify it.

Worked example

You are handed $5$ monthly hedge-fund returns from the investor perspective. Positive returns are gains and negative returns are losses. The observations are a sample, and the exam asks for sample skewness, sample excess kurtosis, and a brief interpretation.

GivenValue
Monthly returns$-10.00$, $-5.00$, $0.00$, $5.00$, $20.00$ percentage points
Observations$5$ monthly observations
FrequencyMonthly
ConventionSample, investor perspective
Sample standard deviation$11.5109$ percentage points per month

The line follows the return observations from left to right. Read each point as one monthly return, with gains above zero and losses below it. The final observation at 20.00 percentage points extends farther on the gain side than the other returns extend on the loss side. That shape suggests positive skewness, but the calculation uses all five centered returns and the sample correction. The chart helps you anticipate the sign; it does not replace the numerical work or determine kurtosis by itself.

Loading diagram…
  1. First find the sample center because every central moment is measured from it. Use the following calculation.

$$ \begin{aligned} \bar{r} &= \frac{-10.00+(-5.00)+0.00+5.00+20.00}{5} \ &= 2.0000 \end{aligned} $$

  1. Next subtract the sample center from every return because skewness uses centered observations. The cubed deviation total is calculated below.

$$ \begin{aligned} \sum_{i=1}^{5}(r_i-\bar{r})^3 &= (-12.0000)^3+(-7.0000)^3+(-2.0000)^3 \ &\quad +(3.0000)^3+(18.0000)^3 \ &= 3,780.0000 \end{aligned} $$

  1. Use the same centered distances for kurtosis because the fourth moment raises each distance to the fourth power. The fourth-deviation total is calculated below.

$$ \sum_{i=1}^{5}(r_i-\bar{r})^4 = 81,904.0000 $$

  1. Apply the sample skewness correction because the observations are a sample. Substitute the cubed total and the displayed sample standard deviation.

$$ \begin{aligned} G_1 &= \frac{5}{(5-1)(5-2)} \times \frac{3,780.0000}{(11.5109)^3} \ &= 1.0327 \end{aligned} $$

  1. Apply the corrected sample excess kurtosis formula. This keeps the answer on the excess scale, where the normal benchmark is $0$.

$$ \begin{aligned} G_2 &= \frac{5(5+1)}{(5-1)(5-2)(5-3)} \times \frac{81,904.0000}{(11.5109)^4} \ &\quad - \frac{3(5-1)^2}{(5-2)(5-3)} \ &= -2.1685 \end{aligned} $$

  1. Convert to Pearson kurtosis only if the question asks for that convention.

$$ \begin{aligned} \text{Pearson kurtosis} &= -2.1685+3 \ &= 0.8315 \end{aligned} $$

The positive sample skewness points to a longer gain-side tail, while the below-normal kurtosis readings describe shape only and do not establish statistical or economic significance.

Changed condition

Keep the same monthly setup, but replace the final return $20.00$ with $-25.00$. This changes the right-tail observation into a left-tail observation. Before calculating, predict a lower mean and negative skewness, then check how the fourth-moment measure moves under the sample convention.

MeasureOriginalChanged final return
Mean return$2.0000$$-7.0000$
Sample skewness$1.0327$$-1.0327$
Sample excess kurtosis$-2.1685$$-13.8315$

The mean moves down, and the odd third moment changes sign, so the skewness direction reverses. The fourth-moment reading moves sharply under this changed return, which is why you check the convention, periodicity, sample size, and outlier sensitivity before drawing a conclusion.

Before reading on, calculate the sample skewness for the changed return set.

Common error

When the question asks for sample excess kurtosis, you might use the population denominator or report Pearson kurtosis. That choice feels natural because the uncorrected moment is the direct calculation and Pearson kurtosis is the familiar fourth-moment label. If the question states a sample convention or gives the normal benchmark of $0$ for excess kurtosis, use the small-sample correction and subtract $3$ instead of reporting Pearson kurtosis.

Calculator

For the baseline sample skewness, key the count, cubed-deviation sum, and sample standard deviation first. Then apply the correction and check the display.

KeysDisplay
2ND CLR TVM0.0000
5 N5.0000
37803,780.0000
11.510864433211.5109
CPT G11.0327

Go deeper

Descriptive tail language tells you what the observed distribution looks like. It does not ask you to make an inferential rejection decision. Describe the observed shape, then evaluate whether the description fits the convention and consider periodicity, sample size, and sensitivity to outliers.

Check yourself

Practice

The changed return set has sample skewness -1.0327. Which interpretation follows?

Put the evaluation steps in order.

  • Identify the monthly periodicity
  • Identify the sample convention
  • Calculate centered moments
  • Apply sample correction
  • Compare with normal benchmarks
  • Interpret tails and outliers

Revision snapshot

Skewness equals the third central moment divided by standard deviation cubed. Positive values indicate a longer right tail and negative values indicate a longer left tail. Sample skewness uses the small-sample correction $\frac{n}{(n-1)(n-2)}$ multiplied by the sum of cubed standardized deviations. Pearson kurtosis is the fourth standardized moment, and a normal distribution has Pearson kurtosis $3$. Excess kurtosis equals Pearson kurtosis minus $3$, and a normal distribution has excess kurtosis $0$. Evaluate the convention, periodicity, sample size, and outlier sensitivity. Descriptive magnitude is not statistical or economic significance. The trap is using the population denominator or reporting Pearson kurtosis when sample excess kurtosis is requested.

Unlock the full lesson and continue reading

Signup to continue reading this lesson and unlimited access questions, flashcards, AI notes, and more

Scan to download app

Scan to download
UNLOCK FREE ACCESS

Practice Questions: Skewness and kurtosis

  1. 1. For H0: μ ≤ 8% versus Ha: μ > 8%, a sample of 64 returns has mean 9.4% and standard deviation 4.0%. If the 5% critical t-value is 1.669, the analyst should:

    • A. Reject the null; the mean return exceeds 8%
    • B. Fail to reject the null; the mean return equals 8%
    • C. Reject the null; the mean return is below 8%
    Show answer

    Answer: A. Reject the null; the mean return exceeds 8%

    Reject the null; the mean return exceeds 8% - The standard error is 4.0%/√64 = 0.5%, so the test statistic is (9.4% − 8.0%)/0.5% = 2.80. - Because 2.80 exceeds the upper-tail critical value of 1.669, reject H0 in favor of μ > 8%. Fail to reject the null; the mean return equals 8% - A test statistic of 2.80 exceeds 1.669, so failing to reject H0 is incorrect. - Even if the test had failed to reject H0, that would not establish that μ equals 8%. Reject the null; the mean return is below 8% - The sample mean of 9.4% is above 8%, giving a positive test statistic of 2.80. - The alternative is μ > 8%, so rejection supports a mean above 8%, not below it.

  2. 2. An analyst models continuously compounded returns as normally distributed, implying terminal asset prices are lognormally distributed. Which conclusion is most appropriate?

    • A. Zero probability of a negative terminal price
    • B. Equal probabilities of gains and losses
    • C. Constant variance across all horizons
    Show answer

    Answer: A. Zero probability of a negative terminal price

    Zero probability of a negative terminal price - A terminal price is the positive initial price multiplied by the exponential of a continuously compounded return. - Because that exponential is strictly positive, a lognormally distributed terminal price cannot be negative. Equal probabilities of gains and losses - A gain occurs when the continuously compounded return is above zero; a loss occurs when it is below zero. - Normality alone does not make those probabilities equal: that would require the return distribution to be centered at zero. Constant variance across all horizons - A normal model for continuously compounded returns does not imply that return variance stays constant as the horizon changes. - Under standard independent-increment assumptions, return variance scales with the length of the horizon.

  3. 3. USD 50,000 is invested for three years at a 6% nominal annual rate compounded monthly. The calculated value of the investment is closest to:

    • A. USD 59,000
    • B. USD 59,551
    • C. USD 59,834
    Show answer

    Answer: C. USD 59,834

    USD 59,834 - Monthly compounding gives a periodic rate of 0.06 ÷ 12 = 0.005 over 12 × 3 = 36 periods. - The future value is USD 50,000(1.005)^36 ≈ USD 59,834. USD 59,000 - This applies simple interest: USD 50,000(1 + 0.06 × 3) = USD 59,000. - It omits the interest earned on interest during the three years. USD 59,551 - This compounds annually: USD 50,000(1.06)^3 = USD 59,550.80, or approximately USD 59,551. - The stated 6% rate is compounded monthly, so the calculation must use 36 monthly periods rather than three annual periods.

  4. 4. A portfolio's 64 monthly active returns have a mean of 1.20% and standard deviation of 4.80%. For H0: μ ≤ 0 versus H1: μ > 0, the 5% critical t-value is 1.669. The analyst should most likely:

    • A. Fail to reject the null because 1.20% is economically small
    • B. Fail to reject the null because the sample mean is below 1.669
    • C. Reject the null and conclude the mean active return is positive
    Show answer

    Answer: C. Reject the null and conclude the mean active return is positive

    Reject the null and conclude the mean active return is positive - The standard error is 4.80%/√64 = 0.60%, giving a t-statistic of 1.20%/0.60% = 2.00. - Because 2.00 exceeds the one-sided 5% critical value of 1.669, reject H0 in favor of a positive mean active return. Fail to reject the null because 1.20% is economically small - Whether 1.20% per month is economically meaningful is separate from the hypothesis test. - The decision is to reject H0 because the t-statistic of 2.00 exceeds 1.669. Fail to reject the null because the sample mean is below 1.669 - The 1.669 critical value applies to a t-statistic, not to a sample mean expressed as a percentage. - Dividing 1.20% by its 0.60% standard error gives 2.00, which exceeds 1.669.

  5. 5. A bank quotes a 6.00% nominal annual rate compounded monthly. The calculated value of the effective annual rate is closest to:

    • A. 6.00%
    • B. 6.17%
    • C. 6.18%
    Show answer

    Answer: B. 6.17%

    6.17% - Monthly compounding makes the monthly rate 0.06/12 = 0.005. - The effective annual rate is (1.005)^12 − 1 = 6.1678%, which rounds to 6.17%. 6.00% - 6.00% is the quoted nominal annual rate, not the effective annual rate. - Reporting it without applying the 12 monthly compounding periods omits the interest earned on interest. 6.18% - Using continuous compounding instead gives e^0.06 − 1 ≈ 6.1837%, which rounds to 6.18%. - The bank specifies monthly, not continuous, compounding.

  6. 6. An investor deposits USD 10,000 at the beginning of each year for four years into an account earning 6% annually. The calculated value at the end of Year 4 is closest to:

    • A. USD 43,746
    • B. USD 46,371
    • C. USD 49,153
    Show answer

    Answer: B. USD 46,371

    USD 46,371 - Deposits at the beginning of each year form a four-payment annuity due. - Its end-of-Year-4 value is USD 10,000 × [(1.06⁴ − 1) / 0.06] × 1.06 = USD 46,370.93, closest to USD 46,371. USD 43,746 - USD 10,000 × [(1.06⁴ − 1) / 0.06] = USD 43,746.16. - This treats the deposits as an ordinary annuity made at each year-end, rather than at each year’s beginning. USD 49,153 - Compounding the annuity-due value for one extra year gives USD 46,370.9296 × 1.06 = USD 49,153.19. - That places the value at the end of Year 5, not the requested end of Year 4.

  7. 7. A stock follows geometric Brownian motion with annual drift 8%, volatility 20%, and simulated standard-normal shock 0.50. The one-year continuously compounded return is closest to:

    • A. 6.00%
    • B. 11.00%
    • C. 16.00%
    Show answer

    Answer: C. 16.00%

    16.00% - Under geometric Brownian motion, the one-year continuously compounded return is \((\mu-\tfrac12\sigma^2)t+\sigma\sqrt{t}Z\). - Substituting the given values yields \((0.08-\tfrac12(0.20)^2)(1)+0.20(0.50)=\mathbf{0.16}\), or 16.00%. 6.00% - The drift after the volatility adjustment is \(0.08-\tfrac12(0.20)^2=\mathbf{0.06}\). - This answer omits the simulated shock of \(0.20(0.50)=0.10\). 11.00% - This answer starts with the adjusted drift of 0.06 but adds only half the shock contribution: \(0.06+0.05=0.11\). - The full shock contribution is \(0.20(0.50)=0.10\), giving 0.16 instead.

Flashcards: Skewness and kurtosis

  1. 1. Define Quartile, Quintile, Decile, and Percentile.

    Show answer

    Divisions of a distribution into quarters (4), fifths (5), tenths (10), and hundredths (100), respectively.

  2. 2. What does kurtosis measure?

    Show answer

    The degree to which a distribution is more or less peaked (and has fatter or thinner tails) than a normal distribution.

  3. 3. What does skewness measure in a distribution?

    Show answer

    The extent to which a distribution is not symmetrical about its mean.

  4. 4. What formula do you use to measure relative dispersion (risk per unit of expected return)?

    Show answer

    The Coefficient of Variation (CV): \[CV = \frac{s}{\bar{X}}\]

  5. 5. What is the median of a dataset?

    Show answer

    The midpoint of a dataset when the data are arranged in ascending or descending order.

  6. 6. What is the mode of a dataset?

    Show answer

    The value that occurs most frequently. (For continuous data, it is the modal interval).

  7. 7. What is the formula for the correlation coefficient \(\rho{XY}\)?

    Show answer

    \[\rho{XY} = \frac{s{XY}}{sX sY}\] where \(s{XY}\) is covariance, and \(sX, sY\) are standard deviations.

  8. 8. What is the arithmetic mean return?

    Show answer

    The simple average of a series of periodic returns.

  9. 9. What is the possible range of values for a correlation coefficient?

    Show answer

    From \(-1.0\) to \(+1.0\).

  10. 10. What does target downside deviation (semideviation) measure?

    Show answer

    Downside risk, by measuring the dispersion of outcomes that fall only below a specific target value.

Everything you need for CFA prep

Get full Oncourse access with lessons, practice questions, flashcards and AI study tools.

GET STARTED FOR FREE
75% OFFLimited time offer
GET 75% OFF