Fat tails: Mediocristan vs. Extremistan
About this app
Two 3D cities from the same random numbers: uniform body heights against a Pareto skyline whose tallest tower pierces the clouds. α slider, running mean, log-log tail, Lorenz curve, the one-in-a-million outlier.
Mediocristan and Extremistan
With normally distributed quantities such as body height, no single value matters much and the mean settles quickly. With fat-tailed distributions, one value can make up a large part of the total. A standard example is the Pareto distribution with P(X > x) = (xₘ/x)^α. The tail index α decides: for α ≤ 2 the variance is infinite, for α ≤ 1 the mean is infinite as well. The terms Mediocristan and Extremistan come from Nassim Nicholas Taleb.
What you see
Two simulated cities are built from the same random numbers. In Mediocristan the towers are normally distributed like body heights; in Extremistan they follow a Pareto distribution. Tower heights are multiples of the median, both cities on the same scale. Figures show mean ÷ median, the share held by the top 1 % and 20 %, the Gini coefficient and the share of the tallest tower. Further panels show the running mean up to 100,000 values, the survival function on log-log axes with a Hill estimate and the Lorenz curve.
Things to try
Set α to 1.16: the top 20 % then hold about 80 % of the total. Push α below 1 and watch the running mean jump with every giant while the median stays put. Insert an outlier of the one-in-a-million kind and compare how far the mean moves in each city. On the log-log plot, Extremistan lies on a straight line with slope −α.
Frequently asked questions
- What are fat tails in statistics?
- Fat tails are distribution tails that decay much more slowly than those of the normal distribution. Extreme values are far more common, and single observations can dominate the sum and the mean. Examples include wealth, insurance losses and follower counts.
- What does the tail index α of a Pareto distribution mean?
- The tail index α sets how fast P(X > x) = (xₘ/x)^α falls off. The smaller α, the heavier the tail. For α ≤ 2 the variance is infinite, for α ≤ 1 the mean as well. The 80/20 rule corresponds to α = log 5 / log 4 ≈ 1.16.
- What is the difference between Mediocristan and Extremistan?
- The terms come from Nassim Nicholas Taleb. In Mediocristan, for example body height, a single value barely changes the average. In Extremistan, for example wealth, a single value can dominate the total, which makes the average an unstable figure.
- What can I adjust in the app?
- You set the tail index α between 0.8 and 4, choose 100, 400 or 1000 values per city and draw new samples. One button adds a one-in-a-million outlier to both cities, another flies the camera to the tallest tower. All data are simulated.
Subject: Data & decisions
More from Data & decisions
- Anchoring effect
- Benford's law
- Berkson's paradox
- Cherry-picking: base period
- Cherry-picking: base period 2.0
- The three-cowboy duel (truel)
- The three-cowboy duel 2.0
- The three means
A VisuApp by heyprof: interactive, free in your browser, no sign-up. What is a VisuApp? · All apps