Understanding the Empirical Rule: Definition and Application

Search Fintrev

Search our guides, reviews and glossary.

The empirical rule is a statistical rule of thumb for normal distributions: about 68% of observations fall within one standard deviation of the mean, about 95% within two standard deviations, and about 99.7% within three. It is also called the 68-95-99.7 rule or the three-sigma rule, because the Greek letter sigma (σ) is the usual symbol for standard deviation. If you know only the mean and standard deviation of a normally distributed variable, the rule tells you roughly how the data spread around the center and how rare an extreme value is.

Key takeaways

• Because the normal curve is symmetric, the share of data outside a band splits evenly between the two tails.
• Three-sigma limits set control limits in statistical quality control and support risk tools such as value-at-risk (VaR).
• Investment returns are not reliably normal, so in investing the rule is a rough volatility guide, not a forecast.

What the Empirical Rule Says

Here µ (mu) is the mean and σ (sigma) is the standard deviation. The rule can be written as three intervals:

• One standard deviation: µ − σ to µ + σ contains about 68% of the data
• Two standard deviations: µ − 2σ to µ + 2σ contains about 95% of the data
• Three standard deviations: µ − 3σ to µ + 3σ contains about 99.7% of the data

The numbers are rounded. The exact areas under a normal curve are about 68.27%, 95.45% and 99.73%. The rule describes proportions of a data set, or equivalently probabilities for a single random observation. “95%” means that 95 out of every 100 observations, on average, should land within two standard deviations of the mean.

Despite the name, the percentages do not come from experiments or observation. They follow mathematically from the shape of the normal curve. The rule is “empirical” in the sense that it works as a practical shortcut when real-world data are close to normal.

How the Empirical Rule Works

Standard deviation measures how far data typically sit from the mean. A narrower standard deviation gives tighter bands and a wider one gives more spread-out bands, but the percentages stay the same because the rule expresses distance in units of σ rather than in raw units.

Each band splits into the share inside it, the share outside it and the share in each tail (rounded rule values):

Reading the Space Between Bands

Since the bell curve is symmetric, you can split each band in half. Each side of the mean holds:

• Mean to 1σ: about 34%
• 1σ to 2σ: about 13.5% (half of 95% minus half of 68%)
• 2σ to 3σ: about 2.35% (half of 99.7% minus half of 95%)
• Beyond 3σ: about 0.15%

These pieces add up to 50% on each side, and they let you answer questions about any range that starts or ends on a whole number of standard deviations.

Worked Example

Hypothetical example. A factory machine cuts bolts whose length is normally distributed with a mean of 50.0 mm and a standard deviation of 0.2 mm.

First, build the bands:

• One standard deviation: 50.0 ± 0.2, or 49.8 mm to 50.2 mm (about 68% of bolts)
• Two standard deviations: 50.0 ± (2 × 0.2), or 49.6 mm to 50.4 mm (about 95%)
• Three standard deviations: 50.0 ± (3 × 0.2), or 49.4 mm to 50.6 mm (about 99.7%)

Now answer some questions.

What share of bolts is longer than 50.4 mm? 50.4 mm is 2σ above the mean. About 5% of bolts fall outside the two-sigma band, and half of those are on the long side, so roughly 2.5% are longer than 50.4 mm.

What share is between 49.8 mm and 50.4 mm? That range runs from 1σ below the mean to 2σ above it. The left piece (mean to −1σ) is about 34%, and the right piece (mean to +2σ) is about 47.5%. Together, roughly 81.5% of bolts fall in that range.

How many bolts per 10,000 would fall outside 49.4 mm to 50.6 mm? About 0.3% lie beyond three sigma, or roughly 30 bolts in 10,000.

How the Empirical Rule Is Used

Quick estimates before full analysis. Once a mean and standard deviation are known, the rule gives a fast estimate of outcomes without detailed calculation. It is useful when collecting complete data would take too long or is not possible.

Statistical quality control. Control charts commonly set upper and lower control limits at three standard deviations from the process mean. If a process is stable and normal, only about 0.3% of measurements should land outside those limits by chance, so a point beyond them is treated as a signal that something in the process may have changed.

Risk analysis. Value-at-risk models in their simplest (parametric) form assume that gains and losses follow a normal distribution. Under that assumption, the standard deviation of returns translates directly into the probability of a loss of a given size, which is the empirical rule’s logic applied to risk.

A rough normality check. If far more than about 0.3% of a data set falls beyond three standard deviations, or the share within one standard deviation is far from 68%, the data are probably not normal. They may be skewed or have heavier tails than the bell curve allows.

Flagging outliers. Analysts often express a value’s distance from the mean as a z-score: z = (x − µ) / σ. Under the empirical rule, a value with |z| above 2 occurs about 5% of the time and one with |z| above 3 about 0.3% of the time, so these thresholds serve as common outlier flags.

Applying the Empirical Rule to Investing

In investing, the rule’s main building block, standard deviation, is the standard measure of volatility. You can calculate it for a single stock, a fund, a portfolio or an index such as the S&P 500.

The steps in a spreadsheet:

1. List closing prices in date order.
2. Calculate each period’s percentage change from the previous close.
3. Apply the standard deviation function to the column of returns, for example =STDEV(C2:C253).
4. To annualize a daily standard deviation, multiply it by the square root of the number of trading days in a year, usually 252.

Several years of returns give a more stable, representative estimate than a few weeks, which can swing widely depending on the period.

Hypothetical example. A fund’s daily returns have a standard deviation of 1.0%. Annualized, that is 1.0% × √252, or about 15.9%. Suppose the fund’s expected annual return is 8%. If returns were normal, about two years in three would fall between −7.9% and 23.9% (8% ± 15.9%), and about 19 years in 20 would fall between −23.8% and 39.8% (8% ± 2 × 15.9%).

That “if” carries most of the weight. Market returns are generally not normally distributed, so the 68-95-99.7 percentages should not be read as real probabilities for an investment. Analysts still compare standard deviations across investments: a higher figure means returns have historically been more spread out, which is generally read as higher risk.

Limitations of the Empirical Rule

It requires a normal distribution. The percentages hold only for data that are normal or close to it. For skewed data, such as income, house prices or the time until a loan defaults, the bands can be badly wrong.

Financial returns have fat tails. Large moves in markets happen far more often than a normal curve predicts. Under a true normal distribution, a move beyond five or six standard deviations is so improbable it should essentially never occur, yet market crashes have produced single-day moves of that size. A model built on the empirical rule will understate the chance of extreme losses, which is a known weakness of normal-based VaR.

Inputs are estimates. The true mean and standard deviation are rarely known. Sample values from a small or unrepresentative data set make every band inaccurate.

Standard deviation treats both directions alike. A sharp rally and a sharp drop raise volatility equally, even though investors care far more about the downside.

Empirical Rule vs. Related Concepts

Chebyshev’s Theorem

Chebyshev’s theorem applies to any distribution with a finite mean and standard deviation, not only normal ones. It guarantees that at least 1 − 1/k² of the data lie within k standard deviations: at least 75% within two standard deviations and at least about 88.9% within three. These minimums are much weaker than the empirical rule’s 95% and 99.7%, which is the price of working for every distribution. When you cannot assume normality, Chebyshev’s bound is the safe one.

Z-Scores and Normal Tables

The empirical rule covers only whole numbers of standard deviations. For any other distance, convert the value to a z-score and look up the probability in a standard normal table. For example, the exact cutoff for the central 95% of a normal distribution is ±1.96σ; the rule rounds it to 2.

Standard Deviation vs. Standard Error

Standard deviation describes how spread out individual observations are. Standard error describes how precisely a sample statistic, usually the mean, estimates the population value, and it shrinks as the sample grows. The empirical rule is applied to standard errors when building approximate confidence intervals: a sample mean plus or minus two standard errors gives a rough 95% interval.

The Bottom Line

The rule helps anyone who needs a quick sense of spread from just a mean and standard deviation, from quality engineers to risk analysts. Investors can use it to compare volatility across funds or stocks, but market returns have fat tails and are not reliably normal. Treat its percentages as a rough guide to risk, and rely on Chebyshev's bound when normality cannot be assumed.