CFA Level 1 · Quantitative Methods
Mesokurtic Distributions: The Baseline Everything Else Is Measured Against
A mesokurtic distribution is one whose tails carry the same weight as a normal distribution. Its kurtosis is 3, and its excess kurtosis is therefore zero. That is the whole definition, and on its own it explains nothing about why the term exists.
Mesokurtic earns its place by being the reference point. Leptokurtic and platykurtic are only meaningful relative to something, and mesokurtic is that something. A candidate who can compute kurtosis but cannot say what it is being compared against has learned the arithmetic and missed the concept.
Key Takeaways
- Mesokurtic means kurtosis of 3, or excess kurtosis of zero, matching the tails of a normal distribution.
- Kurtosis is the fourth central moment divided by the square of the variance, which makes it unit free.
- Because deviations are raised to the fourth power, kurtosis describes the tails rather than the peak.
- Check whether a reported figure is kurtosis or excess kurtosis before interpreting it, because packages differ.
- Kurtosis ignores direction and ignores the size of the spread, and it is unstable in small samples.
- Normality-based risk models assume the mesokurtic case, so a leptokurtic series makes them understate severe losses.
What mesokurtic means
Kurtosis measures how much of a distribution’s variance comes from rare, large deviations rather than from frequent, small ones. A distribution is mesokurtic when that split matches the normal distribution.
The word is often explained as peakedness, and that description causes more trouble than it saves. Two distributions can have very similar peaks and very different tails, and it is the tails that kurtosis is actually responding to, because deviations are raised to the fourth power before being averaged. A deviation of 4 standard deviations contributes 256 times as much as a deviation of 1. The peak contributes almost nothing by comparison.
It helps to see where kurtosis sits among the moments. The mean is the first moment and locates the distribution. The variance is the second and measures how far observations sit from that location. Skewness is the third and asks whether the distance is distributed evenly on both sides. Kurtosis is the fourth, and it asks how much of the total distance is delivered by a handful of very large deviations rather than by many modest ones. Each step up the ladder puts more weight on the observations furthest from the mean, and by the fourth power that weighting has become extreme.
Read kurtosis as a statement about tails, not about the peak. The fourth power is what does it: large deviations dominate the calculation so completely that the shape near the mean barely registers.
The formula, and why 3 is the number to remember
Kurtosis is the fourth central moment divided by the square of the variance:
where μ is the mean, σ is the standard deviation, and the expectation is taken over the distribution. Dividing by σ4 makes the measure unit free, so a result can be compared across assets priced in different currencies or scales.
For a normal distribution this expression evaluates to exactly 3, whatever its mean and whatever its standard deviation. That constant is the entire reason the term mesokurtic exists. Because 3 is awkward to carry around as a baseline, most treatments subtract it and work with excess kurtosis instead:
Which gives the classification in its usable form. Excess kurtosis of zero is mesokurtic, above zero is leptokurtic, below zero is platykurtic.
The value of 3 is not a convention that could have been chosen differently. It falls out of the algebra of the normal distribution, and it does so for every normal distribution regardless of its mean or its standard deviation. That invariance is what makes it usable as a benchmark: any distribution can be compared against it without first being rescaled.
One practical detail separates a textbook answer from a spreadsheet answer. The formula above is the population version, dividing by n. Sample estimators apply a bias correction that depends on the number of observations, and for small samples the two can differ noticeably. The correction matters less as the sample grows, but for the short return windows common in practice it is worth knowing which version produced a number before comparing it against 3.
Reading a reported figure without checking which of the two measures it is. Software packages differ: some report kurtosis, others report excess kurtosis, and a few label excess kurtosis as kurtosis. A reported 3 means a normal-tailed distribution under one convention and a distinctly fat-tailed one under the other. Check whether a normal distribution would score 3 or 0 in that output before interpreting anything.
A worked calculation
The formula is easier to trust once it has been run on real numbers.
A fund reports eight monthly returns, in percent: 2, −1, 3, 0, −2, 1, 4, −3. Compute the kurtosis and classify the distribution.
- Find the mean. (2 − 1 + 3 + 0 − 2 + 1 + 4 − 3) / 8 = 4 / 8 = 0.50.
- Take deviations from the mean: 1.5, −1.5, 2.5, −0.5, −2.5, 0.5, 3.5, −3.5.
- Square them and average. The squares sum to 42.00, so the variance is 42.00 / 8 = 5.25.
- Raise the deviations to the fourth power and average. Those sum to 388.50, so the mean is 388.50 / 8 = 48.5625.
- Divide by the squared variance. 5.252 = 27.5625, and 48.5625 / 27.5625 = 1.76.
- Subtract 3 for excess kurtosis. 1.76 − 3 = −1.24.
Answer: kurtosis 1.76, excess kurtosis −1.24. The distribution is platykurtic, not mesokurtic. Its returns are spread more evenly than a normal distribution would be, with no observation far enough out to dominate the fourth moment.
That result is worth sitting with, because it is the outcome a candidate expects least. A short series of unremarkable returns did not come out at 3. It came out well below it, and it would take an observation far out in a tail to pull it up.
A second fund reports ten monthly returns, in percent: 0.4, −0.3, 0.5, 0.2, −0.4, 0.3, −0.2, 0.1, 9.0, −8.6. Eight quiet months and two violent ones. What does the kurtosis do?
- The mean is 0.10, so the eight quiet months sit within half a point of it and the two violent months sit roughly nine points away.
- The variance comes to 15.57, almost all of it produced by those two observations.
- The mean fourth power comes to 1,200.6, of which the two extreme months supply more than 99%.
- Dividing gives 1,200.6 / 15.572 = 1,200.6 / 242.4 = 4.95.
Answer: kurtosis 4.95, excess kurtosis 1.95. Clearly leptokurtic. Two observations out of ten moved the measure by more than three points against the first fund, and removing either one would move it back most of the way.
Put the two funds side by side and the property to remember becomes obvious. Kurtosis is close to being a measure of the largest few observations. That is what makes it useful as a warning, and it is also what makes it fragile.
The three shapes, side by side
The figure is drawn with all three curves sharing a mean and a variance, which is the comparison that matters. If the variances differed, the curves would simply be wider or narrower and nothing would be learned. Holding the variance fixed forces the question of where that variance comes from, and the answer separates the three shapes: the leptokurtic curve buys its variance from a few distant observations, so it must be thinner through the shoulders and taller at the centre to compensate, while the platykurtic curve spreads the same variance across many moderate ones.
| Shape | Kurtosis | Excess | Tails carry | Typical of |
|---|---|---|---|---|
| Leptokurtic | > 3 | > 0 | More weight than normal | Most financial return series |
| Mesokurtic | = 3 | = 0 | The same weight as normal | The normal distribution itself |
| Platykurtic | < 3 | < 0 | Less weight than normal | Bounded or capped outcomes |
What kurtosis does not measure
Three limits are worth carrying, because each one is a question a candidate will meet.
Kurtosis says nothing about direction. It is built on deviations raised to an even power, so a large loss and a large gain of the same size contribute identically. Take the second fund above and flip the sign of both extreme months, so that the 9.0 becomes a crash and the −8.6 becomes a rally. Every fourth power is unchanged, the variance is unchanged, and the kurtosis is unchanged, yet the two funds would have felt nothing alike to hold. Skewness is the measure that separates them, and the two are read together rather than one instead of the other.
Kurtosis says nothing about the size of the spread. Dividing by the fourth power of the standard deviation removes scale deliberately, so a violently volatile series and a placid one can both be mesokurtic. Kurtosis describes the shape of the spread, and standard deviation describes its size.
Kurtosis also rarely reports platykurtic in finance, and when it does the reason is usually structural rather than statistical. Returns that are capped, floored or otherwise bounded, such as a strategy with a hard stop loss or an instrument with a contractual maximum payout, cannot produce the distant observations that lift the fourth moment. A platykurtic reading on an unconstrained asset is more often a sign that the sample window missed a crisis than that the asset is genuinely well behaved.
Kurtosis is unstable in small samples, which the worked calculation above demonstrates. With eight observations, one more extreme month would have moved the answer substantially. Any kurtosis estimated from a short window should be treated as an indication rather than a measurement, and the window length should be stated whenever the figure is quoted.
Why the baseline matters in risk work
The practical importance of mesokurtic comes from what is built on top of it. A large amount of standard risk machinery assumes normally distributed returns, and a normal distribution is mesokurtic by construction. Every such model is therefore assuming excess kurtosis of zero, whether or not it says so.
Actual market returns are usually leptokurtic. Extreme days arrive more often than a normal distribution allows, which means a model resting on the mesokurtic baseline understates how often severe losses occur. The normal distribution is not wrong in itself. The problem begins with the assumption that market returns follow it.
The scale of the gap is easier to feel with frequencies than with a kurtosis number. Assuming daily returns, roughly 250 trading days a year and counting both tails, a normal distribution puts a four standard deviation move at roughly one occurrence in 60 years, a five standard deviation move at roughly one in 7,000 years, and a six standard deviation move at roughly one in two million years. Moves of that size have been observed on major indices several times within a single career. That mismatch is what excess kurtosis is reporting, and it is why the baseline is worth checking rather than assuming.
That is why the classification is a risk question rather than a vocabulary question. Comparing a return series against the mesokurtic benchmark is how you find out whether a model built on normality is safe to use on it, and a positive excess kurtosis is a direct warning that a value at risk figure calculated under normality will be too comfortable.
Where excess kurtosis is material, the response is to stop relying on the normal assumption rather than to adjust it slightly. Historical simulation, a Student’s t distribution with fatter tails, or extreme value theory applied to the tail are the usual replacements, each of which drops the mesokurtic assumption instead of patching it.
How the exam frames it
Direct definition questions are rare. What appears instead is a figure and a classification, or a comparison in which mesokurtic is the reference the other two are measured against.
Run four checks whenever a kurtosis figure appears. First, is the number kurtosis or excess kurtosis, since 3 and 0 mean the same shape under the two conventions. Second, which way does it sit relative to the baseline, because above means fatter tails and more frequent extremes. Third, what is being assumed downstream, since a normality-based model has already assumed the mesokurtic case. Fourth, how many observations produced the figure, because a short window makes it unreliable.
The classification itself is worth almost no marks. What carries the marks is the consequence: a leptokurtic series breaks a normality-based risk number, and saying so is the answer the question is looking for.
The same applies in an interview, where the question usually arrives as a figure on a screen rather than as a definition. Being able to say that a reported excess kurtosis of 2 means the model sitting behind that number has been assuming zero, and is therefore too comfortable about severe losses, is the answer that lands. Reciting that leptokurtic means fat tails is not.
The short version
- Mesokurtic means kurtosis of 3, or excess kurtosis of zero, matching the normal distribution.
- Kurtosis is the fourth central moment divided by the square of the variance, so it is unit free.
- Because deviations are raised to the fourth power, kurtosis is a statement about tails rather than about the peak.
- Check whether a reported figure is kurtosis or excess kurtosis before interpreting it.
- Kurtosis ignores direction, ignores the size of the spread, and is unstable in small samples.
- Normality-based risk models assume the mesokurtic case, so a leptokurtic series makes them understate severe losses.
Frequently Asked Questions
What does mesokurtic mean?
A distribution is mesokurtic when its tails carry the same weight as a normal distribution. Its kurtosis equals 3 and its excess kurtosis equals zero. The normal distribution is the standard example, and mesokurtic exists mainly as the reference point that leptokurtic and platykurtic are measured against.
Is kurtosis about the peak or the tails?
The tails. Kurtosis raises deviations from the mean to the fourth power before averaging them, so a deviation of 4 standard deviations contributes 256 times as much as a deviation of 1. The observations far from the mean dominate the calculation, and the shape near the peak has very little influence on the result.
What is the difference between kurtosis and excess kurtosis?
Excess kurtosis is kurtosis minus 3. Because a normal distribution always has a kurtosis of exactly 3, subtracting it puts the benchmark at zero, which is easier to read. Always check which one a reported figure is, because software packages differ and a value of 3 means normal tails under one convention and distinctly fat tails under the other.
Are financial returns mesokurtic?
Usually not. Most financial return series are leptokurtic, meaning extreme days arrive more often than a normal distribution allows. That matters because a large amount of standard risk machinery assumes normality, and normality is mesokurtic by construction, so those models understate how often severe losses occur.
Can two distributions have the same kurtosis but look completely different?
Yes. Kurtosis ignores direction, so a series with one huge loss and a series with one huge gain can report the same figure, and skewness is what separates them. It also ignores scale, because dividing by the fourth power of the standard deviation removes it, so a volatile series and a calm one can both be mesokurtic.
How many observations do I need for a reliable kurtosis estimate?
More than a short window provides. Kurtosis is dominated by the largest few observations, so a single additional extreme value can move the estimate substantially, as the worked example in this article shows with only eight returns. Treat any kurtosis from a short sample as an indication rather than a measurement, and state the window length whenever the figure is quoted.
Loading comments...
Add your Thoughts: