Standard Deviation - The Foundation of Volatility
By EC Assets Research Team, Quantitative Foundations · Published · Updated
Standard Deviation: Standard deviation measures the dispersion of a set of returns around their mean. It is the foundational quantification of volatility and underlies VIX, implied volatility, Sharpe ratio, and most risk metrics.
What Standard Deviation Measures
Standard deviation measures how widely a set of values is spread around its mean. Applied to returns, it is the standard measure of volatility: take each return, subtract the mean, square the difference, average the squares and take the square root. The result is in the same units as the returns, so a monthly standard deviation of 3 percent means returns typically land within a few percentage points of their average.
Karl Pearson introduced the term in the 1890s. Its central place in finance dates from Markowitz (1952), who defined portfolio risk as the variance of portfolio return. Today it underlies the Sharpe ratio, parametric value at risk, risk budgets, tracking error and every option price.
How It Works
For n returns with mean r̄, the sample standard deviation is
$$\sigma = \sqrt{\frac{1}{n-1}\sum_{i=1}^{n}(r_i - \bar r)^2}$$
Dividing by n - 1 rather than n corrects the bias that comes from estimating the mean from the same data.
Institutional reporting quotes it annualised. Because variances add across independent periods, the scaling factor is the square root of the number of periods per year, the square root of time rule:
| Return frequency | Periods per year | Annualisation factor |
|---|---|---|
| Daily | 252 | 15.87 |
| Weekly | 52 | 7.21 |
| Monthly | 12 | 3.46 |
| Quarterly | 4 | 2.00 |
A daily standard deviation of 1 percent annualises to about 16 percent, the basis of the Rule of 16.
Worked Example
A portfolio returns 2, -1, 3, -2, 1 and 4 percent over six months. The mean is 7 / 6 ≈ 1.17 percent.
| Return | Deviation from mean | Squared deviation |
|---|---|---|
| 2.0% | 0.83 | 0.69 |
| -1.0% | -2.17 | 4.69 |
| 3.0% | 1.83 | 3.36 |
| -2.0% | -3.17 | 10.03 |
| 1.0% | -0.17 | 0.03 |
| 4.0% | 2.83 | 8.03 |
Deviations in percentage points, squared deviations in percent squared.
The squared deviations sum to 26.83. Dividing by n - 1 = 5 gives a variance of 5.37, and the square root is a monthly standard deviation of 2.32 percent. Annualised, 2.32 × √12 ≈ 8.02 percent.
With only six observations this figure is very imprecise. The standard error of a standard deviation is roughly σ / √(2(n - 1)), about 32 percent of the estimate here and still about 12 percent with 36 months of data.
When It Applies (and Limitations)
Returns are not normal. Standard deviation is defined for any distribution, but the familiar bands, 68 percent of outcomes within one standard deviation and 95 percent within two, assume normality. Equity returns have fatter tails, as Mandelbrot (1963) showed and Cont (2001) catalogued. Under a normal distribution a daily fall of four standard deviations should occur about once every 125 years and a fall of five about once every 13,800 years; equity markets have produced several of each. Skewness and kurtosis describe what the standard deviation leaves out.
It treats gains and losses alike. Returns of +6, -2, -2 and -2 percent and returns of -6, +2, +2 and +2 percent have the same standard deviation of 4 percent, yet the first series has one large gain and the second one large loss. The Sortino ratio uses downside deviation for this reason.
Estimates are noisy. From 36 monthly returns the 95 percent confidence band is roughly plus or minus 23 percent of the estimate, from 60 months about 18 percent and from 120 months about 13 percent. A three-year track record cannot separate a 12 percent volatility strategy from a 15 percent one.
Serial correlation breaks the scaling. The square-root rule assumes independent periods. Lo (2002) shows how autocorrelation distorts annualised statistics: with a monthly autocorrelation of 0.2, the true annual standard deviation is about 1.20 times the √12-scaled figure; with -0.2 it is about 0.83 times. Trend-following and smoothed, illiquid return series are the usual suspects.
It describes the past. Volatility clusters, and a standard deviation measured over a calm period understates the risk of the next regime.
Why It Matters for Institutional Investors
It is the unit of risk budgets. Volatility targets, tracking-error limits and leverage decisions are all expressed in standard deviations, so an error in the estimate becomes an error in position size.
It drives other metrics. The denominator of the Sharpe ratio, the scale of parametric value at risk and the volatility input of option models are all standard deviations; the VaR z-score tool shows how quickly a normal assumption understates tail losses.
It aggregates through correlation. Portfolio variance depends on the standard deviations of the holdings and on their correlations. An equal mix of two assets with 20 percent volatility each has a volatility of about 14.1 percent if they are uncorrelated and the full 20 percent if they move together.
It needs checking against the tails. Pairing standard deviation with maximum drawdown, stress losses and higher moments gives a fuller view than any single figure.
It rewards scepticism in due diligence. A smooth return series with low reported volatility and high positive autocorrelation often reflects stale or model-based pricing rather than low risk. Unsmoothing the returns, as Lo's adjustment does, frequently reveals a materially higher volatility.
References
- Pearson, K. (1894). Contributions to the Mathematical Theory of Evolution. Philosophical Transactions of the Royal Society of London A, 185, 71-110.
- Markowitz, H. (1952). Portfolio Selection. Journal of Finance, 7(1), 77-91.
- Lo, A. W. (2002). The Statistics of Sharpe Ratios. Financial Analysts Journal, 58(4), 36-52.
- Mandelbrot, B. (1963). The Variation of Certain Speculative Prices. Journal of Business, 36(4), 394-419.
- Cont, R. (2001). Empirical Properties of Asset Returns: Stylized Facts and Statistical Issues. Quantitative Finance, 1(2), 223-236.
Frequently asked questions
Why is standard deviation annualised with a square root rather than multiplied?
Because variances, not standard deviations, add across independent periods. Twelve independent months have twelve times the monthly variance, so the annual standard deviation is the monthly one times the square root of twelve. The rule fails when returns are serially correlated.
What is a typical standard deviation for equities?
Over long periods the S&P 500's annualised standard deviation has been in the mid-teens, with calm years such as 2017 below 10 percent and crisis years such as 2008 above 40 percent. Single stocks are usually considerably more volatile than the index, because diversification removes much of their individual risk.
How is standard deviation different from variance?
Variance is the average squared deviation from the mean; standard deviation is its square root. They carry the same information, but standard deviation is in the units of the returns themselves, percent rather than percent squared, which makes it easier to read. Variance is what adds up across assets and periods.
Why do quants use log returns when computing standard deviation?
Log returns add up across periods, which keeps multi-period calculations simple and consistent. For small daily or monthly moves log and simple returns are almost identical; for large moves they diverge, and the choice should be stated.
Is a lower standard deviation always better?
No. Lower volatility often comes with lower expected return, and a smooth return series can hide rare large losses. The relevant comparison is return per unit of risk, as in the Sharpe ratio, together with measures of the tail.
Stay informed
Market commentary, firm news and research from EC Assets - direct to your inbox.