Volatility Metrics Beginner Typical distance from the mean

Standard Deviation σ

Every volatility figure on this site is a standard deviation wearing a different hat.

Quick answer: Standard deviation is the typical distance of a set of numbers from their average, and in finance it is the exact quantity that volatility measures — specifically the standard deviation of an asset's returns, not of its prices.

In simple words

Take a set of numbers and their average. Standard deviation asks: how far, typically, does a number sit from that average? If NIFTY's daily returns average roughly zero but swing about 0.8% on a normal day and 3% on a wild one, the standard deviation of those daily returns is the single number that summarises 'how spread out' they are. A small standard deviation means the numbers huddle near the average; a large one means they scatter. Volatility is this and nothing more: it is the standard deviation of returns, annualised so that a weekly figure and a monthly figure can be compared on one scale.

The famous shorthand is the 68–95–99.7 rule. If returns followed a perfect bell curve, about 68.27% of them would land within one standard deviation of the average, 95.45% within two, and 99.73% within three. That is why a '1σ move' means the ordinary case and a '3σ move' means the rare one. But those percentages are a property of the bell curve, not of the market — and the market, as we will see, does not read the textbook.

Not to be confused with: Variance, which is the standard deviation squared. Variance is the quantity that adds across independent periods and positions, which is why models are built on it; standard deviation is the quantity traders quote, because it is in the same units as the data. They carry identical information, but only one of them adds, and mixing them up is the source of half the errors in volatility arithmetic.

The normal distribution and its standard-deviation bands

68, 95, 99.7 — the rule, stated exactly

A normal distribution with the ±1σ, ±2σ and ±3σ regions shaded.

Exactly 68.27% of a normal distribution lies within one standard deviation of the mean, 95.45% within two and 99.73% within three. These are precise mathematical facts about this curve — and precisely the reason they mislead, because real index returns are not this curve, and their tails are far heavier than the shaded regions suggest.

Professional explanation

Why volatility is a standard deviation of returns, not of prices

It would be natural to take the standard deviation of NIFTY's price and call it volatility, but that number is almost meaningless — it depends entirely on the price level. A stock at ₹100 that moves ₹2 a day and a stock at ₹1,000 that moves ₹20 a day are equally volatile in any sense a trader cares about, yet the standard deviation of their prices differs tenfold. Returns fix this. By working with the percentage change from one day to the next — or more precisely the log return — the calculation becomes scale-free: a 1% day is a 1% day whether the index is at 8,000 or 24,000. Volatility is therefore the standard deviation of returns, and that single choice is what makes it comparable across assets, across time and across price levels.

The n−1 versus n denominator, and finance's zero-mean shortcut

When you compute a standard deviation from a sample rather than an entire population, you divide the sum of squared deviations by n−1, not n. This is Bessel's correction, and it exists because the sample's own mean is estimated from the same data, which uses up one degree of freedom and makes the naive estimate slightly too small. So statistical software and textbooks use n−1. Finance, however, often does something different again: for high-frequency return series the true mean is tiny compared with the daily swings, so practitioners set it to zero and divide by n, measuring dispersion around zero rather than around the sample mean. This is not sloppiness — over a short window the drift is genuinely negligible, and the zero-mean estimator is more stable. It is worth knowing that a volatility figure can differ slightly depending on which of these three conventions produced it.

The uncomfortable part: the percentages belong to the bell curve, not the market

The 68–95–99.7 rule is exact, and it is exactly the problem. Those numbers are properties of the normal distribution, and index returns are not normally distributed. They are fat-tailed: extreme moves happen far more often than a bell curve permits. Under the normal model a five-standard-deviation daily move should occur about once every seven thousand years. In real equity markets, moves of that magnitude arrive several times a decade — the crashes of 1987, 2008, 2020 and every flash event in between. The bell curve is a modelling convenience chosen because it is mathematically tractable, not because anyone checked that the market obeys it. Every time you read '2σ' as 'this happens 5% of the time', you are quoting a fact about a curve that the market has repeatedly and violently declined to follow. A measure you cannot criticise is a measure you cannot use, and this is the criticism that matters most.

Standard deviation scales with the square root of time

A daily standard deviation and an annual one are not related by a factor of 252 but by a factor of √252, because it is variance, not standard deviation, that adds across independent days. If daily returns are independent with standard deviation s, then n days of them have variance n × s² and therefore standard deviation s × √n. This is why a daily volatility is annualised by multiplying by √252 ≈ 15.87, and why a 30-day expected move is the annual figure times √(30 ÷ 365). The square-root-of-time law is not a separate rule to memorise; it is a direct consequence of the fact that standard deviation is the square root of a variance, and variance is what adds. Every √t you see in options pricing traces back to this one line.

The same σ, translated into a NIFTY price band

The ±1σ and ±2σ standard-deviation bands mapped onto NIFTY at 24,000 over a 30-day horizon.

A standard deviation of returns becomes a band of prices once you fix a horizon and a spot: this is the bridge from the abstract ruler to a tradeable range. The band is symmetric here by assumption, but the real distribution of NIFTY returns leans left, so the lower edge is reached more often than the picture implies.

Formula

Sample standard deviation

σ = √( Σ(x_i − x̄)² ÷ (n − 1) )

Divide by n − 1 for a sample (Bessel's correction) or by n for a full population. In finance, short-window volatility often uses a zero-mean estimator — replacing x̄ with 0 and dividing by n — because the true drift is negligible over the window and the estimate is more stable. Volatility annualises by multiplying a daily σ by √252.

σThe standard deviation — the typical distance of the data from its mean, in the same units as the data.
x_iThe i-th observation. For volatility, this is the log return of day i, not the price.
x̄The arithmetic mean of the observations. In the zero-mean finance estimator this is set to 0.
nThe number of observations in the sample.
ΣSummation over all observations from i = 1 to n.

Annualising a daily standard deviation of returns

σ_annual = σ_daily × √252

Because variance adds across independent trading days, the standard deviation grows with the square root of the number of days. There are about 252 trading days in a year, so √252 ≈ 15.87 converts a daily volatility to an annual one.

How to compute a standard deviation of returns

Convert prices to returns. For each day, take the natural log of today's close divided by yesterday's close — the log return. This is what makes the figure scale-free.
Find the mean of those returns, or set it to zero if the window is short and the drift is negligible, which is the common finance convention.
Subtract the mean from each return and square the result. Squaring removes the sign and penalises large deviations far more than small ones.
Sum the squared deviations and divide by n − 1 for a sample, or by n for a population or a zero-mean estimator. This is the variance.
Take the square root. That is the daily standard deviation of returns.
Annualise by multiplying by √252, and express as a percentage. That figure is the volatility everyone quotes.

Practical example

NIFTY worked example

Suppose NIFTY's daily log returns over a recent window have a standard deviation of 0.8% — that is 0.008 per day. Annualise it: 0.008 × √252 = 0.008 × 15.87 ≈ 0.127, or about 12.7% — right in the range India VIX usually occupies. Now turn that annual σ into a price band. Over 30 calendar days the one-standard-deviation move on NIFTY at 24,000 is 24,000 × 0.127 × √(30 ÷ 365) ≈ 872 points. Interpret it: if returns were normal, NIFTY would finish within 872 points of 24,000 about 68% of the time and outside that band about 32% of the time. The number is a ruler, not a rail — it tells you the typical size of a move, and it quietly assumes a bell curve that the next crash will ignore.

BANKNIFTY worked example

BANKNIFTY teaches the danger of the tail. Take a daily return standard deviation of about 1.0%, annualising to 1.0% × √252 ≈ 15.9%. Over 30 days on a 52,000 spot that is a 1σ move of 52,000 × 0.159 × √(30 ÷ 365) ≈ 2,368 points, and a 2σ move of about 4,736. The bell curve says a 3σ single-day drop — roughly 3% here, or about 1,560 points in a session — should be a once-in-a-century event. BANKNIFTY has delivered several in living memory: banking is a leveraged, correlated sector, and its return distribution has one of the fattest left tails on the exchange. The standard deviation is computed the same way as NIFTY's; what differs is how badly the normal assumption underneath the 99.7% figure fails when a banking shock arrives.

Lot sizes used above (NIFTY 75, BANKNIFTY 30) are those in force at the time of writing; NSE revises them periodically. Figures exclude brokerage, STT, exchange charges, stamp duty and GST. Examples are teaching scenarios built on round numbers — they are not historical quotes, not backtests and not trade calls.

Risk note. The standard deviation is used to justify position sizing and value-at-risk limits across the entire industry, and almost all of it rests on the normal-distribution percentages. A risk system that sets a limit at '2σ, breached 5% of the time' is quoting a fact about a bell curve, not about the market, and the market breaches those limits more often and more violently than the number admits. Sizing a position as though 3σ events are once-a-century when they arrive once a decade is how leverage that looks conservative becomes ruinous.

Advantages & limitations

What it is good for

It is scale-free when applied to returns, so it makes a stock at ₹100 and an index at 24,000 directly comparable on one volatility scale.
It is in the same units as the data, unlike variance, which makes it intuitive — a 12.7% volatility is a percentage a human can reason about, where 1.6%² is not.
It underpins the entire 68–95–99.7 framework, giving a fast, if imperfect, way to judge whether a move is ordinary or exceptional.
It scales cleanly through time via the √t law, so a daily figure converts to weekly, monthly or annual with a single square root.
It is the common language of every volatility measure on the site — implied, realised, historical and the VIX are all standard deviations of returns expressed in a common annualised form.

Where it breaks down

It assumes the percentages of a normal distribution, and real index returns are fat-tailed, so it systematically understates how often large moves occur.
It weights an upside surprise and a downside surprise identically, because it squares deviations symmetrically — yet on an equity index the two are not equally likely or equally feared.
It is a single summary of an entire distribution, so two very different return series — one calm with rare jumps, one steadily choppy — can share the same standard deviation and behave nothing alike.
It depends on the estimation window: a 20-day standard deviation and a 250-day one can differ substantially, and neither is more 'correct' — they measure different periods.
It is sensitive to the denominator convention. The same data gives slightly different figures under n, n − 1 and the zero-mean estimator, which matters when volatilities are compared across sources.
It says nothing about the order of the returns. A standard deviation is identical whether the big moves clustered together or scattered evenly, even though clustering is exactly what makes a market dangerous.

Common mistakes

Taking the standard deviation of prices instead of returns. The price-based figure scales with the price level and is not comparable across assets or across time; volatility is always the standard deviation of returns.
Reading '2σ, so 5% of the time' as a hard fact about the market. That 5% is a property of the normal distribution, and real markets breach 2σ noticeably more often because their tails are fatter.
Annualising by multiplying by 252 instead of √252. Variance scales with time, standard deviation with the square root of time, and mixing them inflates the annual figure enormously.
Treating a 5σ event as impossible. Under the bell curve it is a once-in-seven-thousand-years event; in real equity markets moves of that size arrive several times a decade, and betting they cannot is how books blow up.
Comparing standard deviations computed over different windows and concluding one asset is more volatile, when the difference is entirely down to the periods measured.
Ignoring which denominator produced a figure. A volatility computed with n, one with n − 1 and one with a zero mean differ, and quoting them against each other as though identical introduces a spurious gap.
Assuming a symmetric band because the standard deviation is symmetric. On an index the downside is reached more often than the upside, and the standard deviation alone cannot see that skew.

Professional usage

Risk desks live inside the standard deviation. Value-at-risk, expected shortfall, margin models and position limits are all built on an estimate of the standard deviation of returns, usually one that gives more weight to recent observations so it reacts to a regime change rather than averaging it away. The sophisticated ones know the normal assumption is wrong and patch it — fitting a fatter-tailed distribution, adding a jump component, or stress-testing beyond the σ bands entirely — precisely because a 3σ limit calibrated to a bell curve is breached far too often to be trusted alone. The standard deviation is the starting point of every risk calculation and almost never the whole of one.

On the trading side, the standard deviation is the unit in which a volatility trader thinks about everything. An implied volatility is an annualised standard deviation the market is charging; a realised volatility is the annualised standard deviation that actually arrived; the trade is a bet on the gap between the two. When a desk says an option is 'trading at 14 vol', it means the market's implied standard deviation of NIFTY's returns is 14% annualised, and every strike, spread and hedge is priced off that number. The standard deviation is not one tool among many for these traders — it is the coordinate system.

Key takeaways

Standard deviation is the typical distance of the data from its mean, and volatility is exactly the standard deviation of returns — not of prices, which is what makes it scale-free.
The 68–95–99.7 rule is precise: 68.27% within 1σ, 95.45% within 2σ, 99.73% within 3σ — but only for a normal distribution.
Real index returns are fat-tailed, so 'σ events' that the bell curve says are astronomically rare arrive several times a decade. The percentages describe the model, not the market.
Divide by n − 1 for a sample, n for a population, and note that short-window finance often uses a zero-mean estimator dividing by n.
Standard deviation scales with the square root of time because variance is what adds — the √t law behind every volatility and expected-move calculation.

Standard deviation is the ruler underneath every volatility number you will ever read, and once you see that, implied volatility, realised volatility, the VIX and the expected move all become the same measurement expressed in different tenors. But keep the criticism close. The clean percentages that make the standard deviation so useful — 68, 95, 99.7 — are the property of a curve the market has repeatedly torn up, and the moment you forget that the bell curve is a convenience rather than a description, the standard deviation stops being a ruler and becomes a false sense of safety.

Frequently asked questions

What is standard deviation in simple terms?

Standard deviation is the typical distance of a set of numbers from their average. A small standard deviation means the numbers huddle near the mean; a large one means they scatter widely. In finance it is applied to returns, and that figure is what everyone calls volatility.

How is standard deviation related to volatility?

Volatility is the standard deviation of an asset's returns, annualised. There is no difference between the two concepts — implied volatility, realised volatility and India VIX are all standard deviations of returns expressed on a common annual scale. When a desk quotes '14 vol', it means a 14% annualised standard deviation.

Why is volatility the standard deviation of returns and not prices?

Because the standard deviation of prices depends on the price level and is not comparable across assets — a ₹1,000 stock moving ₹20 looks ten times more volatile than a ₹100 stock moving ₹2, though they are identical in percentage terms. Returns make the figure scale-free, so a 1% day is a 1% day at any index level.

What is the 68–95–99.7 rule?

For a normal distribution, about 68.27% of observations fall within one standard deviation of the mean, 95.45% within two and 99.73% within three. It is a fast way to judge whether a move is ordinary or exceptional — but the percentages are exact only for the bell curve, which real markets do not follow.

Are the 68–95–99.7 percentages true for the stock market?

No, and this is the crucial caveat. Those percentages are properties of the normal distribution. Index returns are fat-tailed, so large moves happen far more often than the rule allows — a 5σ day that the bell curve says is a once-in-7,000-years event arrives several times a decade in reality.

What is the difference between the n and n−1 denominators?

Dividing the sum of squared deviations by n gives the population standard deviation; dividing by n−1 gives the sample standard deviation, using Bessel's correction because the mean was estimated from the same data. Finance sometimes uses a third convention — a zero mean divided by n — for short return windows.

What is Bessel's correction?

Bessel's correction is dividing by n−1 instead of n when computing a sample standard deviation. It compensates for the fact that the sample mean, estimated from the same data, makes the naive sum of squared deviations slightly too small. It matters most for small samples and becomes negligible as n grows large.

Why does finance sometimes use a zero mean?

Because over a short window — a few days or weeks of returns — the true average daily return is tiny compared with the daily swings, so setting it to zero barely changes the dispersion and produces a more stable estimate. The zero-mean estimator divides the sum of squared returns directly by n.

How do you annualise a daily standard deviation?

Multiply the daily standard deviation of returns by √252, because there are about 252 trading days in a year and variance — not standard deviation — adds across independent days. A 0.8% daily figure annualises to 0.8% × 15.87 ≈ 12.7%.

Why do you multiply by √252 and not 252?

Because variance scales linearly with time and standard deviation is the square root of variance, so standard deviation scales with the square root of time. Multiplying by 252 instead of √252 would inflate the annual figure roughly sixteenfold — a common and catastrophic beginner error.

What is a 1σ move on NIFTY?

It is a move of one standard deviation, which for NIFTY at 24,000 with a 12.7% annual volatility over 30 days is about 872 points. Under a normal distribution the market finishes within that band about 68% of the time and outside it about 32% of the time.

What is a 3σ event?

A move three standard deviations from the mean, which a normal distribution says should happen about 0.3% of the time — roughly once in 370 observations. In real markets, because the tails are fat, 3σ and larger moves occur considerably more often, which is why risk limits set at 3σ are breached more than the model predicts.

Does standard deviation tell you about direction?

No. It measures the typical size of a deviation from the mean, not its sign. A large standard deviation is equally consistent with sharp rises and sharp falls. Any directional information — such as the heavier downside on an equity index — has to come from the skew, not the standard deviation.

Can two different return series have the same standard deviation?

Yes, easily, and they can behave nothing alike. A calm series with rare violent jumps and a steadily choppy one can share a standard deviation, because the figure summarises spread without regard to the shape of the tails or the order of the moves.

Does standard deviation capture fat tails?

No. It is a single number that assumes, when combined with the 68–95–99.7 rule, a normal distribution. Fat tails — the excess of extreme moves over what the bell curve allows — are measured by higher moments like kurtosis, which the standard deviation cannot see.

How does the estimation window affect standard deviation?

Substantially. A 20-day standard deviation reacts quickly to a recent burst of movement, while a 250-day one smooths it into a long average. Neither is more correct; they simply measure different periods, and comparing them as if interchangeable is a common source of false conclusions.

Is variance or standard deviation better for calculations?

Variance for calculations, standard deviation for quoting. Variance adds across independent periods and positions, which makes the mathematics clean, while standard deviation is in the same units as the data, which makes it intuitive. Models compute in variance and report in standard deviation.

Why does the standard deviation weight up and down moves equally?

Because it squares the deviation from the mean, and squaring removes the sign, so a +2% return and a −2% return contribute identically. On an equity index this symmetry is a limitation, because downside moves are more frequent and more feared than upside moves of the same size.

How does standard deviation relate to India VIX?

India VIX is an annualised standard deviation of NIFTY returns over a 30-day horizon, implied from option prices. When it reads 13, the market is pricing a 13% annualised standard deviation, which translates into a 30-day expected move of roughly 24,000 × 0.13 × √(30 ÷ 365) ≈ 894 points.

What does it mean if a stock has a higher standard deviation than NIFTY?

It means its returns scatter more widely around their average, so it typically moves more per day in percentage terms. But a fair comparison requires the same estimation window and the same return convention — otherwise the difference may be an artefact of how each figure was computed, not a real gap in volatility.

Is standard deviation enough to measure risk?

No, and treating it as sufficient is dangerous. It captures typical dispersion under a normal assumption but misses fat tails, skew, jumps and the clustering of volatility in time — all of which are where real losses come from. It is the starting point of a risk calculation, never the whole of one.

Voice search & related questions

Natural-language questions people ask about standard deviation.

What is standard deviation?

Standard deviation is how far, on average, a set of numbers sits from their mean. In markets it is applied to returns, and that figure is exactly what we call volatility — a scale-free measure of how much an asset typically moves.

Why is a bell curve used for the market when it doesn't fit?

Because it is mathematically convenient — the 68–95–99.7 rule and most of statistics are built on it. The honest answer is that the market has fatter tails than the bell curve, so extreme moves happen far more often than it predicts, and the curve is a useful map that lies about the edges.

How rare is a 5-sigma day really?

The bell curve says roughly once every seven thousand years. Reality says several times a decade — 1987, 2008, 2020 and every flash crash between them. That gap between the model and the market is the single most important thing to remember about the standard deviation.

Do I divide by n or n minus 1?

Divide by n−1 for a sample, which is the statistical default, and by n for a full population. In finance, over short windows, people often set the mean to zero and divide by n instead, because the drift is negligible and the estimate is steadier.

How do I turn a daily volatility into an annual one?

Multiply the daily standard deviation by the square root of 252, which is about 15.87. It is √252 and not 252 because variance adds with time and standard deviation is the square root of variance, so it grows with the square root of the number of days.

Does a low standard deviation mean the market is safe?

Not really. A low standard deviation means recent moves have been small, but it says nothing about the tail risk building underneath. Calm periods are exactly when leverage accumulates, and the quietest standard deviations often precede the loudest moves. It is a measure of recent wobble, not of safety.

Is volatility just another word for standard deviation?

Essentially yes — volatility is the standard deviation of returns, annualised. Implied volatility, realised volatility and India VIX are all the same ruler applied to different periods and expressed on one annual scale, which is why learning the standard deviation unlocks all of them at once.

Sources & references

Last reviewed 10 July 2026. Educational content only — not investment advice.