What are fat tails in stock returns?

Fat tails means extreme returns, both large falls and large rises, happen far more often than a normal bell curve predicts. Research going back to Mandelbrot in 1963 has found this pattern across many markets.

What is the difference between skew and kurtosis?

Skew measures how lopsided a distribution is, for example a long tail of large losses. Kurtosis measures how heavy the tails are overall. A normal distribution has a skew of 0 and a kurtosis of 3.

Distributions and fat tails in stock returns

Why stock returns have fat tails: mean, median, skew and kurtosis explained, and why extreme days happen far more often than a bell curve predicts.

Part of the Statistics for Investors and Traders track on Agenticks. About 11 minutes, written for a intermediate reader.

Read first: Probability basics for trading.

Last reviewed 2026-11-08.

Key takeaways

A distribution shows how often each result happens. The normal bell curve is one possible shape, not the shape of real returns.
The mean is the average and the median is the middle value. A mean below the median often signals a tail of large losses.
Skew measures lopsidedness and kurtosis measures tail heaviness. A normal curve has kurtosis 3, or excess kurtosis 0.
Real returns have fat tails: extreme days happen far more often than a bell curve predicts.
Read risk from the worst days and worst stretches in the data, not from the average spread alone.

A distribution is a picture of how often each possible result happens. Take a thousand daily returns of a stock, sort them into buckets (down 2% to 1%, down 1% to 0%, up 0% to 1%, and so on) and count each bucket. The shape you get, tall in the middle with tails running out to each side, is the return distribution. Many textbook tools assume that shape is the normal distribution, the symmetric bell curve. Real returns look fairly similar near the middle and very different at the edges. That gap, between the bell curve and the extreme days markets actually produce, is what people mean by fat tails. Four numbers describe a distribution's shape, and each is worth knowing.

The mean is the ordinary average: add the values and divide by how many there are. The median is the middle value once they are sorted. Half the results sit above it and half below. One extreme value can drag the mean a long way but barely moves the median. Skew measures lopsidedness. A long tail of big losses on the left is negative skew; a long tail of big gains on the right is positive skew. A quick clue: when the mean sits below the median, a few large losses are often pulling it down. Kurtosis measures how heavy the tails are. A normal distribution has a kurtosis of 3, so analysts often quote excess kurtosis, the value minus 3. Positive excess kurtosis means more extreme results than a bell curve with the same spread would produce.

Worked example: mean versus median

Five illustrative daily returns: +1%, +1%, +2%, -1% and -8%.

Sum of the returns (1 + 1 + 2 - 1 - 8)	-5%
Mean (-5% / 5)	-1%
Sorted, lowest first	-8%, -1%, +1%, +1%, +2%
Median (the middle value) (3rd of 5 sorted values)	+1%
Days above zero	3 of 5
Mean minus median	-2 percentage points: one large loss pulls the average below the typical day

Illustrative numbers chosen to keep the arithmetic easy. Real samples are far longer.

Strategies have shapes too, not just markets. Picture ten trades: nine wins of +1% and one loss of -12%. The win rate is 90% and the median trade is +1%, so it looks excellent. The mean is (9 - 12) / 10 = -0.3% per trade, so on average it loses money. That is negative skew: many small wins and an occasional large loss. Rules that sell insurance-like options, or buy every dip with a very wide stop, tend to have this shape. Rules that buy breakouts with tight stops tend to have the opposite shape, many small losses and an occasional large win, which is positive skew. Neither shape is good or bad on its own. What matters is that the win rate and the median can hide a negative mean, which is why the expected value lesson weights every outcome by its size, not just its frequency.

Mean: The ordinary average of all the results
Median: The middle result once they are sorted
Skew: Whether one tail stretches further than the other
Kurtosis: How heavy the tails are overall

The normal distribution has neat rules. About 68% of results fall within one standard deviation of the mean, about 95% within two and about 99.7% within three. Further out, the chances shrink extremely fast. If daily returns really were normal, with about 252 trading days a year, a fall of more than 4 standard deviations would happen about once every 125 years, and a fall of more than 5 standard deviations about once every 13,800 years. Markets do not respect those odds. On 19 October 1987 the Dow Jones Industrial Average fell 22.6% in a single session. If you assume a daily standard deviation of 1% (a round, illustrative figure), that day was more than 22 standard deviations from an ordinary day. A normal curve gives such a day a chance smaller than 1 in 10 to the power of 112, so small that it should not happen once in the whole life of the universe. It happened anyway.

This is not a new discovery. In 1963 the mathematician Benoit Mandelbrot studied cotton prices and showed that their changes had far more extreme values than a normal distribution allows. Decades of later research landed in the same place. A widely cited 2001 review by Rama Cont lists the regular patterns found across many markets, including these: Heavy tails. Extreme returns are much more common than a bell curve predicts. Gain and loss asymmetry. Large falls in stock prices and indexes are not matched by equally large rises. Volatility clustering. Wild days tend to arrive in groups, and so do calm ones. Longer horizons look more normal. Monthly returns look more like a bell curve than daily or intraday returns do.

The middle lies about the edges

A bell curve fitted to ordinary days badly understates how often extraordinary days happen. Measure the tails directly, by looking at the worst days and the worst stretches in the data, rather than inferring them from the average spread.

What a normal bell curve assumes: Big up days and big down days are mirror images, A 5 standard deviation fall comes about once in 13,800 years, Excess kurtosis of exactly zero
What real return data shows: Wild days tend to come in clusters, Extreme days happen far more often than expected, Excess kurtosis well above zero

Fat tails change how you read almost every risk number. Volatility describes the typical day well and the worst day badly, as the volatility lesson shows. A backtest that never saw a crash has not been tested against one. A Monte Carlo reshuffle of past trades can only replay the extremes those trades contained. And a position sized so that a "once in a century" move would wipe out the account is exposed to moves that markets deliver far more often than that. None of this says a crash is coming. It says the edges of the distribution deserve more respect than the middle, because the edges are where accounts are lost. The risk and return lesson covers the trade-off this sits inside.

You can see past the bell curve

You can now compute a mean and median, explain skew and kurtosis, and say why real returns produce extreme days far more often than a normal curve allows.

Check your understanding

Five daily returns are +2%, +1%, +1%, -1% and -6%. Which statement is true?: The mean is -0.6% and the median is +1%. The sum is -3%, so the mean is -3% / 5 = -0.6%. Sorted, the middle value is +1%.
What does positive excess kurtosis tell you about a set of returns?: It has more extreme results in its tails than a normal distribution with the same spread. Excess kurtosis is kurtosis minus 3. Above zero means heavier tails than a bell curve.
Under a normal distribution, about what share of results fall within two standard deviations of the mean?: About 95%. Roughly 68% fall within one standard deviation, 95% within two and 99.7% within three.
Why is the 22.6% one-day fall in the Dow on 19 October 1987 a problem for the normal distribution?: It is so many standard deviations from the average day that a normal curve treats it as effectively impossible. With a daily standard deviation around 1%, the fall was more than 22 standard deviations, which a bell curve gives a chance smaller than one in 10 to the power of 112.

Common questions

What are fat tails in stock returns?: Fat tails means extreme returns, both large falls and large rises, happen far more often than a normal bell curve predicts. Research going back to Mandelbrot in 1963 has found this pattern across many markets.
What is the difference between skew and kurtosis?: Skew measures how lopsided a distribution is, for example a long tail of large losses. Kurtosis measures how heavy the tails are overall. A normal distribution has a skew of 0 and a kurtosis of 3.

Terms defined in this lesson

Continue

Sources

Written and reviewed by the Agenticks Learn editorial team. How we write and check lessons.

Agenticks Learn is general education. It is not financial, investment, tax or legal advice, it does not take your situation into account, and nothing on it is a recommendation to buy, sell or hold any security or to open any account. Examples use illustrative numbers. Investing involves risk, including losing money you put in, and past performance does not predict future results. Tax rules and limits change; the official sources linked in each lesson are the final word.