Fat-Tail Risk

Fat-tail risk is the reality that financial market returns exhibit fat tails — extreme price moves happen much more frequently and intensely than a normal (Gaussian) distribution would predict. A fat-tailed distribution has a higher probability of extreme outcomes, captured mathematically by excess kurtosis.

Normal distribution is wrong for markets

Finance traditionally assumes market returns follow a normal distribution — the bell curve. A normal distribution is defined by two parameters: mean (average return) and standard deviation (volatility). From these two, you can calculate the probability of any outcome.

For example, if a stock has a mean return of 10% per year with a standard deviation of 15%, a normal distribution predicts:

• A 68% chance of returns between -5% and +25% (within 1 standard deviation).
• A 95% chance of returns between -20% and +40% (within 2 standard deviations).
• A 0.3% chance of a loss exceeding 45% (more than 3 standard deviations).

But markets do not follow normal distributions. Extreme moves happen far more often than normal distributions predict. This discrepancy is fat tails.

Historical evidence of fat tails

The US stock market crash of October 19, 1987, saw a 22% single-day decline. Using the normal distribution assumption, the probability of a move that extreme should be infinitesimal — roughly 1 in 100 million. Yet it happened. This is fat tails.

Studies of market returns over the past century show:

• Moves exceeding 3 standard deviations (supposed to be 0.3% of days) actually occur 1-2% of days.
• Moves exceeding 5 standard deviations happen roughly once every few years; normal distribution says once every 15 million years.
• The largest daily losses in history are much larger than normal distributions would predict.

This is not coincidence or measurement error. It is a fundamental characteristic of market returns: they have fat tails.

Why markets have fat tails

Several mechanisms:

1. Herding and panic. When fear spikes, everyone sells simultaneously, pushing prices down far faster than individual risk assessments would suggest. Panic is non-linear; small bad news can trigger selling waves.

2. Leverage unwinding. When leveraged investors face losses, margin calls force them to sell, driving prices down faster. This feedback loop amplifies moves beyond what fundamentals alone would justify.

3. Liquidity evaporation. In normal times, there are many buyers and sellers. In crises, liquidity evaporates; bids disappear; prices must fall sharply to clear trades. A small shock becomes a large price move due to thin order books.

4. Correlations breakdown. In calm times, assets move somewhat independently. In crises, correlations jump to 1 — everything falls together, amplifying the decline in a diversified portfolio.

5. Tail risk itself. Investors are aware of tail risks and try to hedge. But hedges (like options) become expensive as tail events approach. When they materialize, hedges fail or are circumvented, amplifying the move.

Measuring fat tails: Kurtosis

Fat tails are quantified by kurtosis, a statistical measure of the thickness of distribution tails. A normal distribution has kurtosis of 3 (this number is called excess kurtosis = 0 for normal).

• Kurtosis > 3 (excess kurtosis > 0) = fat tails. Real market returns.
• Kurtosis = 3 (excess kurtosis = 0) = normal distribution. Theoretical.
• Kurtosis < 3 (excess kurtosis < 0) = thin tails. Rare.

US stock market excess kurtosis is typically 3-10, depending on the period. Higher kurtosis = fatter tails = more extreme moves.

Consequences for risk models

Models assuming normal distributions systematically underestimate value-at-risk and tail risk. A value-at-risk model might say the portfolio’s 99% worst-case loss is 5%, but if returns have fat tails, the actual 99% loss is 8% or more.

This is why value-at-risk models have repeatedly failed to predict crashes. The 2008 financial crisis saw losses 5-10x larger than value-at-risk models predicted. This was partly model-risk, but a large part was simply the failure to account for fat tails.

Protecting against fat-tail risk

• Use non-normal distributions. Model returns using Student-t distributions or other fat-tailed distributions. This increases the estimated tail probability.

• Expected-shortfall. Instead of value-at-risk, use expected shortfall, which measures the average loss in the tail. This is less sensitive to distributional assumptions.

• Empirical tail analysis. Look at the historical distribution of returns and estimate tail probabilities from data, not assuming any particular distribution.

• Stress-testing. Explicitly model extreme scenarios and calculate losses, rather than relying on distributional models.

• Conservative position sizing. If your model says the 99% loss is 5%, assume it is really 10% and size positions accordingly.

• Diversification with fat-tail awareness. Recognize that correlations spike in tail events, so diversification does not work as well in crises.

See also