Fat Tails

Fat tails, also called heavy tails, describe probability distributions in which very large or very small outcomes occur much more often than under the normal distribution. On a chart, the ends of the curve, the "tails" that represent rare events, stay thicker for longer instead of shrinking quickly towards zero. In finance this means that market crashes, currency collapses and sudden spikes in volatility are not once-in-a-millennium freaks but recurring features of markets.

The idea matters for anyone who measures risk: banks, insurers, fund managers, regulators and central banks. Models that assume a bell curve tend to understate the chance of large losses. Fat tails are closely linked to the Pareto distribution and other power laws, and to the popular notion of the black swan event. They are also connected to Knightian uncertainty, where the odds themselves are unknown.

Imagine a river that is usually calm. On most days its level barely changes; once in a while it rises a little after rain. A town that plans its flood walls using only typical days will build walls that look generous. But a river with fat tails occasionally produces floods far higher than anything those typical days suggested, and it does so more often than anyone expected.

Markets behave like that river. Most days are quiet, which makes the risk look small. Then, every few years, prices fall further in a day or a week than a calm-weather model says should ever happen. Planning only for ordinary days leaves you exposed to the days that matter most.

A common measure is kurtosis, which compares how much of a distribution's variation comes from extreme values. Excess kurtosis is:

• $X$ is the random variable, such as daily returns.
• $\mu$ is its mean and $\sigma$ its standard deviation.
• $E[\,\cdot\,]$ means the expected (average) value.
• The 3 is subtracted because a normal distribution has kurtosis of exactly 3, so a normal distribution scores zero.

A positive value signals fat tails. Daily stock index returns typically show excess kurtosis well above zero. As a simple illustration of what that means in practice: under a normal model, a daily move of five standard deviations or more should occur roughly once in several thousand years of trading, yet major stock indexes have produced several such days in recent decades. Analysts also use tail indexes from extreme value theory and the Student's t-distribution, which has fatter tails than the normal, to model returns.

In 1963 the mathematician Benoit Mandelbrot showed that changes in cotton prices had far more extreme moves than a normal model allows, and proposed heavy-tailed alternatives. Later evidence has repeatedly supported him:

• On 19 October 1987, Black Monday, the Dow Jones Industrial Average fell about 22.6% in one day.
• In 1998 the hedge fund Long-Term Capital Management, whose partners included two Nobel laureates, lost most of its capital after Russia's default, when correlations and price moves far exceeded its models' assumptions; the hedge fund needed a Federal Reserve-organised rescue by its creditors.
• During the 2008 financial crisis, losses on mortgage securities and credit default swaps were far larger than risk models had suggested.
• Currency pegs can break suddenly, producing jumps that ordinary volatility measures miss, as in the Swiss franc's 2015 surge after the Swiss National Bank removed its cap against the euro.

Economists point to several causes. Leverage forces investors to sell when prices fall, which pushes prices down further. Herding and speculation mean many traders act together. Liquidity can vanish in stress, so prices gap rather than move smoothly. Volatility itself changes over time, rising sharply in crises, so a mixture of calm and stormy periods produces fat tails overall. In economies more broadly, incomes, firm sizes and city populations follow heavy-tailed patterns, often described by power laws.

Fat tails weaken risk measures built on the bell curve, such as simple value at risk, which can say little about how bad losses are once a threshold is breached. After 2008, the Basel Committee's revised market risk framework replaced value at risk with expected shortfall for capital purposes, because it averages losses in the worst cases. Banks and central banks also run stress tests built on severe historical and hypothetical scenarios. Insurers and reinsurers use extreme value models for catastrophes such as hurricanes and earthquakes. Investors respond with diversification across uncorrelated assets, limits on leverage, and tail-hedging strategies using options, though such protection is costly in normal times and requires accepting lower average returns as part of the risk-return trade-off.

A practical lesson is humility about models: a long calm period does not prove that the tails are thin, and historical data may simply not yet contain the worst event.

• Basel Committee on Banking Supervision: Minimum capital requirements for market risk