What it is
A fat-tailed distribution is one in which extreme outcomes — events far from the mean — occur with much greater frequency than a normal (Gaussian) distribution would predict. The "tails" of the distribution are "fat" in the sense that they have more probability mass at the extremes.
The normal distribution is ubiquitous in statistics precisely because the Central Limit Theorem guarantees that averages of independent, identically distributed random variables converge to it. But many real-world processes — financial returns, natural disasters, city sizes, wealth distributions — are not well described by the normal distribution and produce far more extreme events than it predicts.
The normal distribution's 10-sigma problem
In August 2007, Goldman Sachs CFO David Viniar described the market moves his firm was observing as "25-sigma events" that were occurring several days in a row. Under a normal distribution, a 25-sigma event should not occur once in the lifetime of the universe. It was occurring daily.
This is not a failure of measurement. It is a failure of model. Financial returns are not normally distributed. They exhibit excess kurtosis — heavier tails than the normal — and in many markets, negative skew. The 2008 financial crisis, the 1987 Black Monday crash (a 20% single-day decline), and the 1997-1998 Asian/Russian crises all produced returns in the range of 10-25 standard deviations under a normal model. Under a fat-tailed model, they are rare but not impossible.
Power laws and Pareto distributions
Many financial and economic variables are better described by power law distributions. A power law distribution has the property that the probability of an event of size x is proportional to x raised to a negative exponent. This produces a distribution where very large events, while rare, are far more likely than the normal distribution predicts.
Wealth distribution follows a power law (Pareto distribution): the top 1% hold far more than 1% of total wealth in every measured country. City sizes follow a power law: a country's largest city is typically 2-3× larger than its second-largest city, which is 2-3× larger than its third-largest, and so on. Financial losses in crises follow a power law.
Why this matters for risk management
Value at Risk (VaR) — the standard risk measure used by banks and mandated by regulation — typically assumes normally distributed returns. A 99% daily VaR means the model predicts losses will exceed this threshold 1% of days. Under a fat-tailed distribution, the 99th percentile loss is much larger, and the 100th percentile — the expected loss given that VaR is exceeded — is dramatically larger.
Banks that used normal-distribution VaR models in 2008 found their models had systematically underestimated the frequency and severity of extreme losses. The models said these events should not happen; the events happened anyway. The problem was not the data — the problem was using a distributional model that was fundamentally inconsistent with how financial returns actually behave.
One thing most people get wrong
Fat tails don't just mean larger extremes — they mean diversification works less well precisely when it is needed most. In a normal-distribution world, correlations between assets are assumed to be stable. In fat-tailed financial markets, correlations spike during crises: assets that were weakly correlated in normal times become highly correlated when markets fall sharply. The diversification benefit that a portfolio was constructed to provide largely disappears in exactly the scenario it was designed to protect against.
This correlation breakdown is well-documented empirically. In the 2008 crisis, nearly all risk asset classes fell together, including asset classes whose historical correlations to US equities were low or negative. The correct inference for portfolio construction is to stress-test correlation assumptions under crisis scenarios and to not rely on diversification for tail risk protection. Tail risk requires explicit tail hedging — options, variance swaps, or crisis-specific positioning — not just cross-asset diversification.