Standard Deviation

The Conceptual Foundation — What Standard Deviation Measures

Standard deviation measures the dispersion of a set of values around their mean — it answers the question of how spread out the observations are relative to the average observation. In an investment context, the values being measured are periodic returns — daily, monthly, quarterly, or annual returns over a historical period — and the mean is the average of those periodic returns. Standard deviation expresses how far the typical observation departs from the average.

An investment with low standard deviation has returns that cluster tightly around the mean — the actual periodic returns are close to the average return in most periods, indicating stable and predictable performance. An investment with high standard deviation has returns that are widely dispersed around the mean — actual periodic returns deviate substantially from the average in most periods, indicating volatile and unpredictable performance.

The conceptual insight is intuitive — an investment whose annual returns have been one, two, three, four, and five percent over five years has a mean of three percent and very little dispersion — the worst year was only two percentage points below average. An investment whose annual returns have been negative twenty, negative five, zero, fifteen, and forty percent over five years also has a mean of six percent but enormous dispersion — the worst year was twenty-six percentage points below average. Both might appear attractive on average, but the second investment's high standard deviation signals that the investor must be prepared for and financially capable of sustaining very large deviations from the average in either direction.

The Calculation — Variance and Its Square Root

Standard deviation is the square root of variance — a mathematical relationship that is important to understand both for the examination and for appreciating why standard deviation rather than variance is the preferred risk measure in practice.

Variance is the average of the squared deviations of each observation from the mean. For a population of N observations, variance equals the sum of each squared deviation divided by N. For a sample of N observations drawn from a larger population — the typical investment finance context — the sample variance equals the sum of each squared deviation divided by N minus one, with the subtraction of one reflecting the degrees of freedom correction that produces an unbiased estimate of the population variance.

The calculation proceeds in five steps.

First, calculate the mean return — add all periodic returns and divide by the number of periods. If an investment returned five, eight, two, twelve, and three percent over five years, the mean is thirty divided by five, equalling six percent.

Second, calculate the deviation from the mean for each period — subtract the mean from each observation. The deviations are negative one percent, positive two percent, negative four percent, positive six percent, and negative three percent.

Third, square each deviation — the squared deviations are zero point zero one, zero point zero four, zero point one six, zero point three six, and zero point zero nine — squaring eliminates the algebraic sign and places extra weight on larger deviations.

Fourth, calculate the variance — sum the squared deviations and divide by N minus one for a sample. The sum is zero point six six, divided by four for the sample variance, equalling zero point one six five percent squared.

Fifth, take the square root of the variance — the square root of zero point one six five equals approximately zero point four zero six — approximately four point zero six percent.

This four point zero six percent is the investment's standard deviation of annual returns — it indicates that the typical annual return deviated by approximately four percentage points from the six percent average over the five-year sample period.

The reason standard deviation rather than variance is used as the primary risk measure in practice is dimensional consistency — variance is expressed in squared units, which are difficult to interpret. If returns are measured in percent, variance is measured in percent squared — a number that has no direct intuitive meaning. Standard deviation, being the square root of variance, is expressed in the same unit as the original returns — percent — making it directly interpretable and comparable to the return figures it is meant to contextualise.

The Normal Distribution — The Statistical Framework

The interpretation of standard deviation in investment analysis relies on the assumption that investment returns are approximately normally distributed — the bell-shaped probability distribution that is the foundation of classical statistics. Under the normal distribution, observations cluster symmetrically around the mean with the frequency of each observation declining smoothly as the deviation from the mean increases.

The normal distribution has a precise and directly examination-tested property regarding standard deviations. Approximately sixty-eight percent of all observations fall within one standard deviation of the mean — between mean minus one standard deviation and mean plus one standard deviation. Approximately ninety-five percent of all observations fall within two standard deviations of the mean. Approximately ninety-nine point seven percent of all observations fall within three standard deviations of the mean — the three-sigma rule that is sometimes called the empirical rule.

Applied to investment returns, these properties have direct analytical implications. An investment with a mean annual return of ten percent and a standard deviation of fifteen percent has a one-standard-deviation range of negative five percent to positive twenty-five percent — approximately sixty-eight percent of annual returns should fall within this range under normality. The two-standard-deviation range — approximately ninety-five percent coverage — runs from negative twenty percent to positive forty percent. A return below negative forty-five percent — three standard deviations below the mean — should occur in approximately zero point three percent of years under normality, or roughly once in three hundred years.

These probability estimates assume the normal distribution is a good description of actual return behavior — an assumption that is approximately but not perfectly true for investment returns. The critical departures from normality that are empirically documented in investment returns are fat tails — more frequent extreme outcomes than the normal distribution predicts — and negative skewness for equity returns — more frequent large negative returns than large positive returns of the same magnitude. These departures mean that standard deviation systematically understates the true probability of extreme negative outcomes — the actual frequency of returns three or more standard deviations below the mean is higher than the three-sigma rule suggests. Value-at-Risk and Conditional Value-at-Risk measures were developed specifically to address this limitation by directly modelling the tail of the return distribution without assuming normality.

Portfolio Standard Deviation — The Power of Diversification

One of the most important properties of standard deviation in the portfolio context is that the standard deviation of a portfolio of assets is generally less than the weighted average of the standard deviations of the individual assets — the mathematical expression of diversification's risk reduction benefit.

The portfolio standard deviation of a two-asset portfolio equals the square root of the sum of three terms — the squared weight of asset one multiplied by the variance of asset one, plus the squared weight of asset two multiplied by the variance of asset two, plus two times the weight of asset one times the weight of asset two times the standard deviation of asset one times the standard deviation of asset two times the correlation between the two assets.

Portfolio standard deviation equals the square root of the quantity weight one squared times variance one plus weight two squared times variance two plus two times weight one times weight two times standard deviation one times standard deviation two times correlation.

The critical insight in this formula is the role of the correlation coefficient — the term that determines how much of the diversification benefit is realised.

When correlation equals one — assets move perfectly in tandem — the portfolio standard deviation equals the weighted average of the individual standard deviations. No diversification benefit is achieved because the assets always move together and offsetting movements never occur.

When correlation is between zero and one — assets are positively correlated but not perfectly — the portfolio standard deviation is less than the weighted average of individual standard deviations. Some diversification benefit is achieved because the assets occasionally move in opposite directions, partially offsetting each other's volatility.

When correlation equals zero — assets are completely uncorrelated — the diversification benefit is meaningful. The portfolio standard deviation falls substantially below the weighted average of individual standard deviations.

When correlation is negative — assets tend to move in opposite directions — the diversification benefit is maximised. Assets that fall when others rise and rise when others fall provide the strongest volatility reduction when combined in a portfolio. A perfectly negatively correlated pair of assets — correlation of negative one — can theoretically be combined in proportions that produce zero portfolio standard deviation — a perfectly hedged portfolio with no return variability.

This mathematical relationship is the quantitative foundation of Harry Markowitz's 1952 mean-variance optimisation framework — the insight that combining imperfectly correlated assets reduces portfolio standard deviation without proportionately reducing expected return, enabling investors to achieve superior risk-adjusted returns through diversification. The efficient frontier — the set of portfolios offering the highest expected return for each level of standard deviation — is the graphical representation of this optimisation.

Standard Deviation Versus Beta — The Critical Comparison

Standard deviation and beta are both measures of risk — but they measure different dimensions of risk and are appropriate in different analytical contexts. The distinction between them is one of the most consistently tested concepts on the Series 65 examination.

Standard deviation measures total risk — the complete variability of an investment's returns, encompassing both the systematic risk that comes from broad market movements and the unsystematic risk that comes from factors specific to the individual security or portfolio. It makes no distinction between these two sources of volatility — it captures all return variability regardless of its origin.

Beta measures only systematic risk — the sensitivity of an investment's returns to movements in the market portfolio. A beta of one means the investment moves in perfect tandem with the market. A beta of two means the investment moves twice as dramatically as the market. A beta of zero means no systematic relationship to market movements. Beta entirely ignores unsystematic risk — the company-specific or sector-specific volatility that can be eliminated through diversification.

The choice between standard deviation and beta depends on the investor's portfolio context. Standard deviation is the appropriate risk measure when the investment being evaluated represents the investor's entire wealth — when there is no diversification, all volatility matters regardless of its source, and standard deviation captures the complete risk the investor bears. This is why the Sharpe ratio — appropriate for evaluating standalone portfolios — uses standard deviation in the denominator.

Beta is the appropriate risk measure when the investment is one component of a larger diversified portfolio — when the investor holds many assets across which unsystematic risk is substantially eliminated, only systematic risk remains relevant to the investor's total portfolio volatility, and beta captures exactly that systematic component. This is why the Treynor ratio — appropriate for evaluating one component of a multi-portfolio investment programme — uses beta in the denominator, and why the CAPM rewards only systematic risk with higher expected return.

Annualisation — Converting Periodic Standard Deviations

Standard deviation is period-dependent — a standard deviation calculated from monthly returns is a monthly standard deviation that is not directly comparable to an annual standard deviation. For comparison purposes, periodic standard deviations must be converted to a common annual basis through annualisation.

Annualised standard deviation equals the periodic standard deviation multiplied by the square root of the number of periods per year.

For daily returns with approximately two hundred and fifty-two trading days per year — the United States equity market convention — the annualised standard deviation equals the daily standard deviation multiplied by the square root of two hundred and fifty-two, approximately fifteen point eight seven.

For monthly returns with twelve months per year — the convention used by most mutual fund performance reporting systems — the annualised standard deviation equals the monthly standard deviation multiplied by the square root of twelve, approximately three point four six.

For weekly returns with fifty-two weeks per year — annualised standard deviation equals the weekly standard deviation multiplied by the square root of fifty-two, approximately seven point two one.

This annualisation formula assumes that returns are independently and identically distributed across periods — that each period's return is statistically independent of prior periods. This assumption holds approximately but not exactly for actual investment returns — return autocorrelation during trending market periods and volatility clustering during stressed periods produce modest departures from the independent-and-identically-distributed assumption that can cause the square-root-of-time scaling to slightly overstate or understate the true annualised volatility.

Standard Deviation in the Suitability and Fiduciary Framework

Under the fiduciary duty of the Investment Advisers Act of 1940 and the care obligation of Regulation Best Interest at 17 CFR 240.15l-1, investment advisers and broker-dealers must understand and communicate the risk characteristics of recommended investments to clients — and standard deviation is the primary quantitative risk measure applicable to portfolio-level risk communication.

FINRA Rule 2111's suitability requirements and Regulation Best Interest's care obligation both require that recommendations be based on a reasonable understanding of the potential risks of the investment — and standard deviation provides the most comprehensive single-number summary of total return variability that advisers can use to assess whether a portfolio's risk level is consistent with the client's assessed risk tolerance. A client with low risk tolerance should generally hold portfolios with low standard deviation — portfolios whose historical return variability demonstrates stable, predictable performance. A client with high risk tolerance may appropriately hold portfolios with higher standard deviation in exchange for higher expected returns.

The communication of standard deviation to clients requires care — most retail investors do not have an intuitive understanding of what a seventeen percent annual standard deviation means for their lived investment experience. Converting standard deviation into practical terms — using the normal distribution properties to describe likely ranges of annual outcomes — helps clients understand the volatility they should expect. An adviser who explains that a portfolio with a ten percent mean return and seventeen percent standard deviation would produce annual returns between negative seven percent and positive twenty-seven percent in approximately two years out of three — the one-standard-deviation range — provides the client with actionable risk context that a bare statistical number cannot.

The Limitations of Standard Deviation — What It Cannot Capture

Several well-documented limitations of standard deviation constrain its usefulness as a complete risk measure and require supplementation with other risk metrics in rigorous investment analysis.

Symmetry assumption — standard deviation treats upside and downside deviations from the mean identically. A return of positive thirty percent when the mean is ten percent produces the same twenty-percentage-point deviation as a return of negative ten percent when the mean is ten percent — and contributes equally to the standard deviation calculation. For most investors, the positive deviation is beneficial rather than risky — they do not object to returns that exceed the average. Standard deviation penalises upside volatility as heavily as downside volatility despite the asymmetric preferences of risk-averse investors. The Sortino ratio addresses this by substituting downside deviation — the standard deviation of returns below the minimum acceptable return — for total standard deviation in the Sharpe ratio denominator.

Fat tails and non-normality — the assumption that investment returns follow a normal distribution systematically underestimates the frequency of extreme outcomes. Equity markets exhibit negative skewness — large drawdowns occur more frequently than large rallies of equivalent magnitude — and excess kurtosis — fat tails with more frequent extreme outcomes than normality predicts. The 2008 financial crisis produced daily equity market moves that should have occurred perhaps once in thousands of years under the normal distribution assumption but occurred repeatedly within weeks. Value-at-Risk measures attempt to directly quantify the loss that will be exceeded with a specified probability — typically one percent or five percent — without relying on the normality assumption.

Historical backward-looking nature — standard deviation is calculated from historical returns that may not be representative of future return variability. A stock whose business model has undergone a fundamental transformation may have very different future volatility than its historical standard deviation suggests. A low-volatility stock that has never experienced the specific risk scenario it now faces — a pharmaceutical company that has never had a major drug recall, for example — will have a historical standard deviation that systematically understates its true prospective risk. This limitation requires analysts to supplement historical standard deviation with scenario analysis, stress testing, and qualitative risk assessment.

The Two Divisors — The Sample Correction Worked

The choice between dividing by N and dividing by N minus one changes the figure that results from identical data, and the size of the change can be worked out exactly. Take five annual returns of five, eight, two, twelve and three percent, whose mean is six percent. The deviations from that mean are negative one, positive two, negative four, positive six and negative three percentage points, and their squares are one, four, sixteen, thirty-six and nine — a sum of sixty-six in percentage points squared. Dividing that sum by N minus one, which is four, gives a sample variance of sixteen point five and a sample standard deviation of approximately four point zero six percent. Dividing the same sum by N, which is five, gives a population variance of thirteen point two and a population standard deviation of approximately three point six three percent. The population divisor always produces the smaller of the two figures, because the same numerator is divided by the larger number.

The gap between the two figures closes as the number of observations grows, and it closes in a way that can be stated as a ratio. The sample variance exceeds the population variance by the factor N over N minus one, so the sample standard deviation exceeds the population standard deviation by the square root of that factor. At five observations the factor is five over four, whose square root is approximately one point one one eight, and the sample figure of four point zero six percent accordingly stands approximately eleven point eight percent above the population figure of three point six three percent. At sixty-one observations, approximately five years of monthly returns, the factor is sixty-one over sixty and its square root is approximately one point zero zero eight — a difference of approximately eight tenths of one percent, small enough that the choice of divisor rarely changes a conclusion. The correction therefore makes a material difference only in short samples.

The reason the smaller divisor is the right one for a sample can be seen without any statistical theory, by measuring the squared deviations from a reference point other than the sample's own mean. For the same five returns the sum of squared deviations measured from six percent is sixty-six. Measured from five percent the squares are zero, nine, nine, forty-nine and four, a sum of seventy-one; measured from seven percent they are four, one, twenty-five, twenty-five and sixteen, a sum of seventy-one again. The sum of squared deviations is at its smallest when it is measured from the mean of the observations themselves and grows for any other reference point. A sample's own mean will not in general coincide with the mean of the wider population the sample was drawn from, so the squared deviations measured from the sample mean are systematically smaller than deviations measured from the population mean would have been. Dividing by N minus one rather than by N raises the result to compensate for that systematic shortfall.

Standard Deviation as Holdings Are Added — Where the Reduction Stops

The two-asset formula generalises to any number of holdings, and the generalised form shows where the reduction in standard deviation stops. Consider n holdings in equal weights of one over n each, every holding having the same variance V and every pair of holdings the same correlation. Each of the n variance terms contributes one over n squared multiplied by V, and each of the n multiplied by n minus one cross terms contributes one over n squared multiplied by V multiplied by the correlation. The variance terms therefore contribute V over n in total, the cross terms contribute V multiplied by the correlation multiplied by n minus one over n, and the portfolio variance is V multiplied by the quantity one over n plus the correlation multiplied by one minus one over n. At two holdings this reduces to one half of V plus one half of the correlation multiplied by V, which is what the two-asset formula gives for two equally weighted assets of equal standard deviation.

The two terms behave in ways that can be read directly off the expression. The first term, V over n, falls toward zero as holdings are added. The second term does not contain n at all in the limit: as n grows, one minus one over n approaches one, and the portfolio variance approaches V multiplied by the correlation. The portfolio standard deviation therefore approaches the standard deviation of a single holding multiplied by the square root of the common correlation, and no number of further holdings takes it below that level. The floor is set by the correlation among the holdings rather than by how many holdings are held, which is why a correlation assumption matters more to a large portfolio's standard deviation than a holding count does.

The arithmetic shows how quickly that floor is approached. Take holdings each with a standard deviation of thirty percent — a variance of nine hundred in percentage points squared — and a correlation of zero point three between every pair. One holding alone gives thirty percent. Ten holdings give nine hundred divided by ten, which is ninety, plus nine hundred multiplied by zero point three multiplied by nine tenths, which is two hundred and forty-three: a variance of three hundred and thirty-three and a standard deviation of approximately eighteen point two percent. Thirty holdings give thirty plus two hundred and sixty-one, a variance of two hundred and ninety-one and a standard deviation of approximately seventeen point one percent. The floor is nine hundred multiplied by zero point three, a variance of two hundred and seventy and a standard deviation of approximately sixteen point four percent. The first ten holdings remove approximately eleven point eight percentage points of standard deviation and the next twenty remove approximately one point two further.

The Coefficient of Variation — Dispersion Measured Against the Mean

A standard deviation taken on its own says nothing about the return that accompanies it, and the coefficient of variation is the measure that puts the two together. The coefficient of variation is the standard deviation divided by the mean return, and it states the units of dispersion borne for each unit of mean return. Five annual returns of five, eight, two, twelve and three percent, with their mean of six percent and their sample standard deviation of approximately four point zero six percent, give a coefficient of variation of four point zero six divided by six, approximately zero point six eight. Because numerator and denominator are both expressed in percentage points, the ratio carries no unit at all, and it is that absence of unit which allows the dispersion of two series with quite different mean returns to be compared on a single scale.

The measure can reverse the ordering that standard deviation alone produces. A portfolio with a mean annual return of four percent and a standard deviation of five percent has a coefficient of variation of five divided by four, which is one point two five. A portfolio with a mean annual return of ten percent and a standard deviation of nine percent has a coefficient of variation of nine divided by ten, which is zero point nine. The second portfolio carries much the larger standard deviation — nine percent against five — and the smaller coefficient of variation, because its dispersion is accompanied by more than twice the mean return. Ranking the two portfolios by standard deviation and ranking them by coefficient of variation therefore place them in opposite orders, the first ranking answering how widely returns have varied and the second how much variation each unit of mean return has cost.

The measure fails where the mean return approaches zero, and the arithmetic shows exactly how. A series with a standard deviation of four point zero six percent and a mean return of zero point five percent has a coefficient of variation of approximately eight point one two; the same dispersion against a mean of negative two percent gives negative two point zero three. The ratio grows without limit as the mean approaches zero, is undefined at a mean of exactly zero, and turns negative once the mean is negative — which would rank a loss-making series as the least variable of a group whose dispersions are identical. The coefficient of variation is therefore usable for ordering a set of series only where every mean return in the comparison is positive and clear of zero.

Chebyshev's Inequality — What Holds When Returns Are Not Normal

The sixty-eight, ninety-five and ninety-nine point seven percent figures are properties of the normal distribution and hold only to the extent that returns are normally distributed. A second and much weaker statement holds whatever the shape of the distribution. Chebyshev's inequality states that for any distribution with a finite variance the proportion of observations lying within k standard deviations of the mean is at least one minus one over k squared. At k equal to two, one minus one over four gives at least seventy-five percent. At k equal to three, one minus one over nine gives at least approximately eighty-eight point nine percent. The inequality assumes nothing about symmetry, about skewness or about the weight of the tails, and it applies to a finite set of observations as readily as to a theoretical distribution.

Setting the two sets of figures side by side measures how much of the usual interpretation rests on the normality assumption rather than on the data. Within two standard deviations the normal distribution gives approximately ninety-five percent and the distribution-free bound gives at least seventy-five percent, a gap of twenty percentage points. Within three standard deviations the normal distribution gives approximately ninety-nine point seven percent and the bound gives at least approximately eighty-eight point nine percent, a gap of approximately ten point eight percentage points. Read from the other side, the normal distribution places approximately five percent of observations outside two standard deviations where the bound permits as much as twenty-five percent, and approximately zero point three percent outside three standard deviations where the bound permits as much as approximately eleven point one percent.

Applied to a portfolio, the difference between the two statements is the difference between an estimate and a floor. A portfolio with a mean annual return of eight percent and a standard deviation of twelve percent has a two-standard-deviation range running from negative sixteen percent to positive thirty-two percent. Under the normal distribution approximately one year in twenty falls outside that range. Under Chebyshev's inequality alone as many as one year in four may fall outside it, because up to twenty-five percent of observations are permitted beyond two standard deviations whatever the distribution's shape. No set of observations can place more than one quarter of its members beyond two standard deviations of its own mean, so a series that respects the bound has demonstrated nothing about whether it is normally distributed.

Why Dispersion Grows With the Square Root of Time

The square root in the annualisation formula is not a convention but a consequence of how means and variances behave when periodic returns are added together. Over n periods whose returns are independent of one another, the expected sum of the returns is n multiplied by the expected return of one period, so the mean scales in direct proportion to the number of periods. The variance of that sum behaves differently. The two-asset formula shows that combining two quantities carries a cross term equal to twice the product of the two standard deviations and the correlation between them, each quantity here entering at a weight of one rather than at a fractional portfolio weight, and independence sets that correlation at zero. With every cross term at zero the variance of the sum is simply the sum of the n period variances, which is n multiplied by the variance of one period. The standard deviation, being the square root of the variance, is therefore the square root of n multiplied by the standard deviation of one period.

The two scalings pull apart as the horizon lengthens, and the arithmetic makes the divergence concrete. Take a strategy whose monthly returns have a mean of one percent and a standard deviation of four percent, and consider the sum of its monthly returns rather than their compounded product. Over twelve months the expected sum is twelve percent and the standard deviation is four multiplied by the square root of twelve, approximately thirteen point eight six percent. Over forty-eight months the expected sum is forty-eight percent and the standard deviation is four multiplied by the square root of forty-eight, approximately twenty-seven point seven one percent. The expected sum has quadrupled while the standard deviation has only doubled, which is the whole content of the square root in the formula.

The same algebra applied to the average of the periodic returns rather than to their sum runs the other way. The average is the sum divided by n, so its variance is the variance of the sum divided by n squared — that is, n multiplied by the period variance, divided by n squared, which is the period variance divided by n. The standard deviation of the average periodic return is therefore the periodic standard deviation divided by the square root of n. For monthly returns with a standard deviation of four percent, twelve months give four divided by the square root of twelve, approximately one point one five percentage points, and forty-eight months give four divided by the square root of forty-eight, approximately zero point five eight percentage points. A longer sample pins the mean return down more closely while saying nothing at all about the dispersion of the individual periodic returns, which does not shrink as observations accumulate.

When Periodic Returns Are Not Independent

Where consecutive returns are related to one another the cross terms do not vanish, and the square-root-of-time figure is no longer the right one. Take two consecutive months whose returns each have a standard deviation of four percent — a variance of sixteen in percentage points squared — and a correlation between the two months of positive zero point three. The variance of the two-month sum is sixteen plus sixteen plus twice zero point three multiplied by four multiplied by four, which is thirty-two plus nine point six, or forty-one point six, giving a standard deviation of approximately six point four five percent. The square-root-of-time figure for the same two months is four multiplied by the square root of two, approximately five point six six percent. Where consecutive returns are positively correlated the square-root-of-time rule therefore understates the true dispersion of the multi-period return.

A negative correlation between consecutive periods reverses the direction of the error. With the same two months at a correlation of negative zero point three, the variance of the sum is thirty-two minus nine point six, or twenty-two point four, and the standard deviation is approximately four point seven three percent — below the five point six six percent the square-root-of-time rule gives. A series whose strong periods tend to be followed by weak ones accumulates less dispersion over a long horizon than the rule predicts, and a series whose periods run in the same direction accumulates more. The sign of the error follows the sign of the correlation between the periods, so the rule carries no built-in direction of bias and a figure produced by it cannot be treated as conservative.

The error compounds over a long horizon because the number of cross terms grows faster than the number of variance terms. Over n periods there are n variance terms and n multiplied by n minus one ordered cross terms, so at twelve periods there are twelve variance terms and one hundred and thirty-two cross terms. With a monthly standard deviation of four percent and a correlation of positive zero point three between every pair of months, the variance of the twelve-month sum is twelve multiplied by sixteen, which is one hundred and ninety-two, plus one hundred and thirty-two multiplied by zero point three multiplied by sixteen, which is six hundred and thirty-three point six: a total of eight hundred and twenty-five point six and a standard deviation of approximately twenty-eight point seven three percent, against the thirteen point eight six percent the square-root-of-time rule gives for the same inputs. A correlation small enough to be difficult to detect in a short series can therefore more than double the true annual figure.

Common Errors in Calculating and Reading a Standard Deviation

The most common arithmetic error in annualisation is multiplying by the number of periods instead of by its square root. A monthly standard deviation of four percent multiplied by twelve gives forty-eight percent, where multiplying by the square root of twelve gives approximately thirteen point eight six percent. The erroneous figure is larger by a factor of twelve divided by the square root of twelve, which is itself the square root of twelve, approximately three point four six. The same mistake made on daily returns inflates the figure by two hundred and fifty-two divided by the square root of two hundred and fifty-two, which is the square root of two hundred and fifty-two, approximately fifteen point eight seven. The error is therefore not a small distortion but a figure between three and sixteen times too large, depending on the frequency of the underlying observations.

A one-standard-deviation range is frequently read as a worst case, which under the normal distribution it is not. If approximately sixty-eight percent of annual outcomes fall inside the band running from the mean less one standard deviation to the mean plus one standard deviation, approximately thirty-two percent fall outside it; and because the normal distribution is symmetrical about its mean, approximately sixteen percent of all outcomes fall below the lower edge of that band. A return worse than the mean less one standard deviation is therefore expected in approximately one year in six rather than as a rare event. The band describes the ordinary range of outcomes and is not a floor beneath them.

A third error substitutes the square root of the average of two variances for the average of the two standard deviations, and the two quantities are not equal. With standard deviations of twenty percent and ten percent the variances are four hundred and one hundred, whose average is two hundred and fifty, and the square root of two hundred and fifty is approximately fifteen point eight one percent. The average of the two standard deviations is fifteen percent, whose square is two hundred and twenty-five. The two squared figures differ by twenty-five, which is exactly one quarter of the square of the difference between the two standard deviations — ten squared is one hundred, and one hundred divided by four is twenty-five. The square root of the average variance therefore equals or exceeds the average of the standard deviations in every case, and the two coincide only where the two standard deviations are equal.

Examination Relevance and Key Takeaways

Standard deviation is tested on the Series 65 examination in the context of investment risk measurement, the Sharpe ratio, the efficient frontier, Modern Portfolio Theory, portfolio diversification, the distinction from beta, and the normal distribution properties.

The key points to retain are these.

Standard deviation is the primary measure of total investment risk — quantifying how widely an investment's returns have varied around their mean return over a specified period. It is the square root of variance — variance being the average of the squared deviations of each periodic return from the mean return. Higher standard deviation indicates greater return variability and greater investment risk — lower standard deviation indicates more stable and predictable returns. Standard deviation is expressed in the same unit as the original returns — typically percent — making it directly interpretable as the typical deviation of a periodic return from the mean.

Under the normal distribution assumption, approximately sixty-eight percent of observations fall within one standard deviation of the mean, approximately ninety-five percent within two standard deviations, and approximately ninety-nine point seven percent within three standard deviations — the empirical rule. Investment returns exhibit fat tails and negative skewness that cause standard deviation to understate the true probability of extreme negative outcomes relative to the normal distribution prediction.

Portfolio standard deviation is less than the weighted average of individual asset standard deviations whenever the correlation between assets is less than one — the mathematical basis of diversification's risk reduction benefit. The portfolio standard deviation formula equals the square root of the quantity weighted squared variances plus two times the product of weights times standard deviations times the correlation — with the correlation term determining how much diversification benefit is achieved.

Zero correlation produces substantial risk reduction. Negative correlation maximises risk reduction. Perfect positive correlation produces no diversification benefit. Standard deviation measures total risk — both systematic and unsystematic. Beta measures only systematic risk. Standard deviation is appropriate for evaluating standalone portfolios where all volatility matters. Beta is appropriate for evaluating individual components of a diversified programme where only systematic risk matters — which is why the Sharpe ratio uses standard deviation and the Treynor ratio uses beta. Annualisation converts periodic standard deviations to annual terms by multiplying by the square root of the number of periods per year — square root of two hundred and fifty-two for daily, square root of twelve for monthly, square root of fifty-two for weekly.