Quant Analytics: from scratch to pro · Quant foundations: returns, risk and clean data · Beginner
Volatility: from daily moves to an annual figure
Volatility is the standard deviation of returns. Quoted volatilities are annual, so daily figures are scaled by the square root of the number of periods per year. This lesson shows how to compute it, why 252 or 365 is used, and when the square-root rule fails.
In this lesson
- Compute a sample standard deviation of returns by hand
- Annualise a daily volatility with the right number of periods per year
- Convert between daily, weekly, monthly and annual volatility
- Say when the square-root-of-time rule is unreliable
Read first: Returns: simple, log and compounded, Volatility explained
A price chart shows that a market moves. Volatility puts a number on how much. It is the standard deviation of returns, and it is the basic input to risk limits, position sizes, option prices and most of the portfolio tools in this curriculum. The catch is that volatility depends on the period you measure it over, so figures are annualised to make them comparable.
Computing it
Take returns r_1 ... r_n (log returns are the usual choice, see simple and log returns). The sample standard deviation is:
sigma = sqrt( sum( (r_i - mean)^2 ) / (n - 1) )
Dividing by n - 1 instead of n corrects for the fact that the mean was estimated from the same data. The squared version, sigma^2, is the variance.
Why the square root of time
If daily returns are independent with the same variance, the variance of a return over T days is T x sigma_daily^2. Variances add; volatilities do not. Taking the square root gives:
sigma_T = sigma_daily x sqrt(T)
So volatility grows with the square root of the horizon, as the chart shows. The annual figure uses T equal to the number of return periods in a year.
Which number of days
- Equities and most futures: about 252 trading days, a convention. The exact count varies by year and market.
- Crypto and other markets that trade every day: 365.
- Weekly or monthly returns: 52 or 12 periods.
The choice matters. Using 252 for a market that trades 365 days understates its annual volatility by about 17%. State the convention whenever you quote a figure.
Worked example
Five daily returns: +1.0%, -0.5%, +0.8%, -1.2%, +0.4%.
- Mean: 0.10%.
- Sum of squared deviations from the mean: 0.000344.
- Sample variance: 0.000344 / 4 = 0.000086.
- Daily standard deviation: 0.927%. (Dividing by 5 instead gives 0.829%, a smaller figure that ignores the estimated mean.)
- Annualise: 0.927% x 15.87, where 15.87 is
sqrt(252), gives 14.72%.
A steady 1% a day annualises to 15.87%. Over a week of 5 days it is 2.24%, and over a month of 21 days it is 4.58%. Going the other way, 16% a year is 1.008% a day and 4.62% a month.
For a crypto asset at 3% a day, the annual figure is 3% x sqrt(365) = 57.31%, not the 47.62% that 252 would give.
When the square-root rule fails
The rule needs returns that are uncorrelated from one period to the next. If they are not, the variance of a two-day return is 2 x sigma^2 x (1 + rho), where rho is the correlation between consecutive days.
With a daily volatility of 1%, two-day volatility is 1.414% if rho is zero, 1.549% if rho is 0.2, and 1.265% if rho is -0.2. Trending returns make long-horizon risk larger than the rule says; reversing returns make it smaller. Lo and MacKinlay's variance-ratio test, cited above, compares the two. See time series basics.
Volatility also changes over time: calm weeks and turbulent weeks cluster. A single annual figure averages over both. Finally, a volatility from few returns is imprecise. If returns were normal, one computed from 21 daily returns has a relative standard error of about 15.8%.
Common mistakes
- Dividing by
non small samples, or forgetting to annualise at all. - Using 252 for a market that trades every day.
- Scaling volatility in proportion to time, so a month looks like 21 days of daily volatility.
- Mixing units: a daily figure compared with an annual one.
- Reading volatility as a ceiling on loss. It describes typical moves, and rare moves are larger. See return distributions and fat tails.
In practice
Say how the figure was built: return type, window length, divisor and periods per year. Two volatilities are comparable only if all four match. Analytics on FinUties applies measures like these to the markets it covers, so they are at hand when you read a chart.
Volatility over longer horizons: square root against straight line
- square-root rule (1% a day)
- wrong: straight-line scaling
Worked numbers
Each result is recomputed from the formula shown.
| What | Formula | Result |
|---|---|---|
| day 1 return | 0.010 | +1.0% |
| day 2 return | -0.005 | -0.5% |
| day 3 return | 0.008 | +0.8% |
| day 4 return | -0.012 | -1.2% |
| day 5 return | 0.004 | +0.4% |
| mean daily return | (r1 + r2 + r3 + r4 + r5) / 5 | 0.10% |
| sum of squared deviations | (r1-mean_r)**2 + (r2-mean_r)**2 + (r3-mean_r)**2 + (r4-mean_r)**2 + (r5-mean_r)**2 | 0.000344 |
| sample variance | ss / 4 | 0.000086 |
| sample daily standard deviation | sqrt(var_d) | 0.927% |
| same data, dividing by n | sqrt(ss / 5) | 0.829% |
| square root of 252 | sqrt(252) | 15.87 |
| annualised volatility from the sample | sd_d * sqrt(252) | 14.72% |
| 1% a day annualised | 0.01 * sqrt(252) | 15.87% |
| 1% a day over 5 days | 0.01 * sqrt(5) | 2.24% |
| 1% a day over 21 days | 0.01 * sqrt(21) | 4.58% |
| 3% a day with 365 days | 0.03 * sqrt(365) | 57.31% |
| 3% a day with 252 days | 0.03 * sqrt(252) | 47.62% |
| relative shortfall using 252 | 1 - crypto252 / crypto365 | 17% |
| daily from 16% a year | 0.16 / sqrt(252) | 1.008% |
| monthly from 16% a year | daily_from_16 * sqrt(21) | 4.62% |
| two-day volatility, rho = 0 | 0.01 * sqrt(2) | 1.414% |
| two-day volatility, rho = 0.2 | 0.01 * sqrt(2 * (1 + 0.2)) | 1.549% |
| two-day volatility, rho = -0.2 | 0.01 * sqrt(2 * (1 - 0.2)) | 1.265% |
| relative standard error of a volatility from 21 returns, if returns were normal | 1 / sqrt(2 * (21 - 1)) | 15.8% |
Check yourself
Daily returns are +1.0%, -0.5%, +0.8%, -1.2% and +0.4%. What is the sample standard deviation, and the annualised figure with 252 days?
The mean is 0.10%. The squared deviations from it sum to 0.000344. Dividing by n - 1 = 4 gives a variance of 0.000086, so the standard deviation is 0.927% a day. Times sqrt(252) = 15.87 gives 14.72% a year.
A crypto asset has a daily volatility of 3% and trades every day. What is its annual volatility, and what error does using 252 cause?
Use 365 days: 3% x sqrt(365) = 57.31%. With 252 the result is 47.62%, about 17% too low in relative terms.
Annual volatility is 16%. What is the approximate one-month (21 trading days) volatility?
Daily is 16% / sqrt(252) = 1.008%. Monthly is 1.008% x sqrt(21) = 4.62%. Equivalently 16% x sqrt(21/252).
Why does the square-root rule give too low a figure when returns trend from day to day?
The variance of a two-day return is 2 x sigma^2 x (1 + rho). With positive rho the true two-day volatility exceeds sigma x sqrt(2). With sigma = 1% and rho = 0.2 it is 1.549% rather than 1.414%.
Key terms
Sources
- The Econometrics of Financial Markets. Princeton University Press (John Y. Campbell, Andrew W. Lo, A. Craig MacKinlay), 1997. Accessed 9 Nov 2026
- Stock Market Prices Do Not Follow Random Walks: Evidence from a Simple Specification Test. The Review of Financial Studies 1(1), 41-66 (Andrew W. Lo, A. Craig MacKinlay), 1988. Accessed 9 Nov 2026