Kurtosis

What is kurtosis? Provide the definition and intuitive explanation.

Answer
\[ \text{Kurtosis}(X) \;=\; \frac{\mathbb{E}\big[(X - \mu)^4\big]}{\sigma^4} \]

Kurtosis measures the tailedness of a distribution, usually compared to a normal distribution.

Often one uses excess kurtosis (normal distribution has a kurtosis of 3.):

\[ \text{Excess Kurtosis}(X) = \text{Kurtosis}(X) - 3 \]

In general (not always) we have:

  • Excess kurtosis \(> 0\): heavy tails / more extreme outliers (leptokurtic).

  • Excess kurtosis \(= 0\): tails similar to normal (mesokurtic).

  • Excess kurtosis \(< 0\): lighter tails; fewer extreme outliers (platykurtic).

Notes and comments

Comment 1: High kurtosis does not necessarily mean a very “sharp peak”and it is not related to "peakdness".

Back to collection