Kurtosis Explained — Leptokurtic, Platykurtic & Mesokurtic

The Bridge Engineer & The Freak Storm

In 1998, a bridge engineer was reviewing 50 years of wind-speed data for a major crossing. The average wind speed looked perfectly safe. The standard deviation looked acceptable too. Every standard model gave the green light.

But one senior analyst noticed something the summary statistics had hidden: the wind data had an unusually heavy tail. Extreme gusts — the kind that happen once every 30 years — were far more likely than a normal distribution would predict. The bridge was redesigned. A near-disaster was avoided.

In finance, ignoring kurtosis caused multi-billion dollar losses in 2008 — risk models assumed normal distributions while the real data had fat tails. In manufacturing, kurtosis reveals whether defects cluster near the mean or explode at the extremes. In machine learning, it helps you understand feature distributions before modelling.

The Three Shapes — Mesokurtic, Leptokurtic, Platykurtic

Every distribution's tail behaviour falls into one of three categories. Think of them as three different exam score distributions in a school — same average, same spread, but very different risk of extreme outcomes.

Visual Diagram — Tail Behaviour at a Glance

The diagram below compares the three kurtosis types overlaid on the same axis. Notice that all three curves share the same mean and approximate spread — kurtosis only changes the peak height and tail weight.

Notice how the leptokurtic curve has a taller, sharper spike at the centre but its tails extend further and drop more slowly — meaning extreme values occur more often than you would expect from a normal distribution. The platykurtic curve is flatter overall, with very little probability mass in the extreme tails.

The Formula — Kurtosis & Excess Kurtosis

Kurtosis is the standardised fourth central moment of a distribution. The fourth power makes it extremely sensitive to values far from the mean — which is exactly what we want when measuring tail heaviness.

Step-by-Step Calculation

Let's manually calculate kurtosis for a small dataset of daily temperature readings (°C): [18, 20, 20, 21, 21, 21, 22, 22, 23, 32]. The last value (32) is an outlier — we want to see how it affects kurtosis.

Real-World Stories for Each Type

Leptokurtic — Stock Market Returns

Daily stock market returns are the classic example of a leptokurtic distribution. On most days, returns cluster tightly near zero — forming a sharp central peak. But market crashes and rallies (Black Monday 1987, the 2008 crisis, COVID March 2020) produce extreme returns that a normal distribution assigns near-zero probability. Traders who assumed normality were wiped out. This is the infamous "fat tails" problem in finance.

Platykurtic — Student Exam Scores (Uniform Marking)

Imagine a school that caps grades between 40 and 80, and grades are spread fairly evenly across that range. There are no students scoring 5 or 98 — the extremes are clipped. This distribution is platykurtic: a broad, flat peak with very thin tails. In manufacturing, a tightly controlled production process (e.g. a machine stamping parts to ±0.1 mm tolerance) produces platykurtic output — extreme defects are nearly impossible.

Mesokurtic — Human Heights

Adult heights in a large population follow approximately a normal distribution, which is the textbook mesokurtic case. There are very tall and very short people, but they occur at the rate the normal distribution predicts — no surprises in the tails. Excess kurtosis ≈ 0.

Python Implementation

Using SciPy

from scipy import stats
import numpy as np

# Temperature data with one outlier
data = [18, 20, 20, 21, 21, 21, 22, 22, 23, 32]

# scipy returns EXCESS kurtosis by default (normal = 0)
excess_kurt = stats.kurtosis(data)
print(f"Excess Kurtosis: {excess_kurt:.4f}")
# Output: Excess Kurtosis: 3.2842

# Raw kurtosis (fisher=False disables the -3 correction)
raw_kurt = stats.kurtosis(data, fisher=False)
print(f"Raw Kurtosis:    {raw_kurt:.4f}")
# Output: Raw Kurtosis:    6.2842

# Interpret the result
if excess_kurt > 0:
    print("Leptokurtic — fat tails, sharp peak")
elif excess_kurt < 0:
    print("Platykurtic — thin tails, flat peak")
else:
    print("Mesokurtic — normal-like tails")

Using Pandas

import pandas as pd

data = [18, 20, 20, 21, 21, 21, 22, 22, 23, 32]
s = pd.Series(data)

# .kurt() returns excess kurtosis
print(f"Excess Kurtosis: {s.kurt():.4f}")
# Output: Excess Kurtosis: 4.5765  (Pandas uses sample correction by default)

# On a DataFrame — kurtosis of each column
df = pd.DataFrame({
    'temperature': [18, 20, 20, 21, 21, 21, 22, 22, 23, 32],
    'humidity':    [55, 60, 62, 60, 58, 61, 63, 64, 62, 59]
})

print(df.kurt())
# temperature    4.576...
# humidity      -0.123...

Comparing All Three Types Numerically

from scipy import stats
import numpy as np

np.random.seed(42)

# Mesokurtic — standard normal
mesokurtic   = np.random.normal(loc=0, scale=1, size=10000)

# Leptokurtic — Student's t-distribution (low df = fatter tails)
leptokurtic  = np.random.standard_t(df=3, size=10000)

# Platykurtic — uniform distribution
platykurtic  = np.random.uniform(low=-3, high=3, size=10000)

for name, dist in [("Mesokurtic  (normal)", mesokurtic),
                   ("Leptokurtic (t, df=3)", leptokurtic),
                   ("Platykurtic (uniform)", platykurtic)]:
    k = stats.kurtosis(dist)
    print(f"{name}  →  Excess Kurtosis = {k:.3f}")

# Output (approximate):
# Mesokurtic  (normal)   →  Excess Kurtosis =  0.021
# Leptokurtic (t, df=3)  →  Excess Kurtosis =  5.847
# Platykurtic (uniform)  →  Excess Kurtosis = -1.193

Kurtosis Comparison Table

Golden Rules