Kurtosis is a statistical measure that describes how much of a dataset’s variation comes from extreme values, or outliers, rather than from moderate deviations around the average. It tells you whether the “tails” of your data distribution are heavy (lots of outliers) or light (few outliers) compared to a normal distribution. If you’re looking at a bell curve, kurtosis isn’t about how tall or short the peak is. It’s about how thick or thin the ends are.
What Kurtosis Actually Measures
One of the most persistent misunderstandings about kurtosis is that it measures “peakedness,” or how pointy a distribution looks. This is misleading. A distribution can have a sharp peak and still have low kurtosis, and a flatter-looking distribution can have high kurtosis. What kurtosis really captures is “tailedness”: how likely the distribution is to produce values far from the center.
Think of it this way. If you’re tracking the daily returns of a stock, kurtosis tells you how often you should expect surprisingly large gains or losses compared to what a standard bell curve would predict. High kurtosis means those extreme days happen more than you’d expect. Low kurtosis means the data stays closer to the average, with fewer dramatic swings.
The normal distribution, the classic bell curve, has a kurtosis value of 3. This serves as the baseline. Everything else is measured relative to it.
Excess Kurtosis and the Baseline of Zero
Because the normal distribution sits at a kurtosis of 3, many statistical tools subtract 3 from the raw value to make comparisons simpler. This adjusted number is called “excess kurtosis,” and it resets the baseline to zero. A normal distribution has an excess kurtosis of 0. A positive excess kurtosis means heavier tails than normal. A negative value means lighter tails.
This distinction matters more than it might seem, because different software handles it differently. Excel’s KURT function reports excess kurtosis (already subtracting 3), while some other tools like EViews report the raw value. If you’re comparing results across platforms, check which version you’re looking at. A kurtosis of 4.7 in Excel means something very different from a kurtosis of 4.7 in software that reports the raw number, since the Excel figure already has 3 removed.
The Three Types of Kurtosis
Distributions are grouped into three categories based on their kurtosis relative to the normal curve.
- Mesokurtic: Kurtosis near 3 (or excess kurtosis near 0). These distributions behave roughly like a normal distribution in terms of outlier frequency. The bell curve itself is the classic example.
- Leptokurtic: Kurtosis greater than 3 (positive excess kurtosis). These distributions have fat tails, meaning extreme values show up more often than a normal distribution would predict. Financial return data is frequently leptokurtic.
- Platykurtic: Kurtosis less than 3 (negative excess kurtosis). These distributions have thin tails, with data points clustering more tightly around the center and fewer outliers than expected.
A uniform distribution, where every outcome is equally likely within a range and nothing falls outside it, is a good example of a platykurtic shape. There are no outliers at all, so the tails are as light as they can be. On the other end, the distribution of insurance claims is often leptokurtic: most claims are small, but a meaningful number are extremely large.
How Kurtosis Differs From Skewness
Kurtosis and skewness are often discussed together because they both describe the shape of a distribution beyond just the average and spread. But they measure completely different things.
Skewness measures symmetry. It tells you whether the data leans to one side of the average. A skewness of zero means the distribution is balanced. Negative skewness means a longer tail on the left side (more extreme low values), and positive skewness means a longer tail on the right (more extreme high values). Kurtosis, by contrast, ignores which direction the extremes fall. It only cares how much total weight sits in both tails combined. You can have a perfectly symmetric distribution with very high kurtosis if outliers appear equally on both sides.
Why Kurtosis Matters in Practice
Kurtosis has real consequences whenever decisions depend on how often extreme events occur. In finance, this is particularly important. Standard risk models often assume that asset returns follow a normal distribution, but stock market data consistently shows positive excess kurtosis. This means large losses (and large gains) happen more frequently than those models predict. The 2008 financial crisis and similar market shocks are sometimes called “fat-tail events” precisely because they fall in the heavy tails that high kurtosis describes.
Risk analysts use kurtosis when calculating metrics like Value-at-Risk, which estimates the worst loss a portfolio might experience over a given period. If you ignore high kurtosis and assume a normal distribution, you’ll underestimate how often severe losses can occur. Research on S&P 500 data has shown that incorporating skewness and kurtosis into risk models meaningfully improves accuracy, especially for estimating losses in the tails of the distribution.
Outside of finance, kurtosis shows up in quality control (detecting unusual variation in manufacturing), signal processing, and any field where understanding the frequency of extreme observations matters. In medical research, for example, a dataset of patient blood pressure readings with high kurtosis would signal that an unusual number of patients have readings far from the group average, which could flag a subpopulation worth investigating.
How to Interpret Your Kurtosis Value
If you’ve calculated kurtosis and are staring at a number, here’s a practical framework. First, confirm whether your tool reported raw kurtosis or excess kurtosis. If the baseline is 3, you’re looking at raw kurtosis. If the baseline is 0, it’s excess kurtosis.
An excess kurtosis near zero means your data’s tails look roughly normal. There’s nothing unusual about the frequency of outliers. Positive values mean your data has more outliers than a normal distribution would suggest. The higher the number, the heavier the tails. An excess kurtosis of 1 or 2 indicates moderately heavy tails. Values above 5 or 6 suggest your data has a substantial number of extreme observations, which may warrant extra attention depending on your analysis.
Negative excess kurtosis means your data has fewer outliers than expected. The values are bunched closer to the center. This is less commonly a source of concern but can matter if your analysis assumes a normal distribution, since assuming normality would actually overestimate the chance of extreme values.
One important caveat: kurtosis is sensitive to sample size. With a small dataset, a single extreme value can dramatically inflate the number. If your sample is small and your kurtosis looks surprisingly high, it’s worth checking whether one or two data points are driving the result before drawing conclusions about the overall distribution.