Volatility and Tail Risk Glossary: VaR, Expected Shortfall and Stress

Volatility and tail-risk terms measure different features of possible financial losses. Standard deviation summarises typical dispersion; value at risk (VaR) estimates a chosen loss threshold over a horizon; expected shortfall (ES) describes the average loss beyond a specified tail cutoff under a defined methodology. Neither VaR nor ES is an unconditional limit on how much a portfolio can lose, and both depend on data, assumptions and the risk model.

This specialised glossary expands on the broad risk management terminology by explaining distributions, asymmetry, stress scenarios and the difference between frequent price movement and rare severe loss. It uses examples suitable for portfolio researchers and professional trading education without presenting a numerical risk measure as a safety guarantee.

Professional glossary definitions

Realised volatility

Realised or historical volatility summarises the dispersion of observed returns over a stated sample and frequency. Analysts often annualise daily standard deviation using a square-root-of-time convention, subject to return assumptions. A 20% annualised figure is not a statement that the market will move exactly 20% in the next year, and it is highly sensitive to the chosen window and frequency.

Implied volatility

Implied volatility is the volatility input that reconciles an option pricing model to the market price of that option, given other inputs. It may contain expected variability, jump exposure, risk premia and supply-demand effects. Different strikes and maturities can have different implied volatilities, so quoting a single implied volatility for an entire asset can conceal the volatility surface.

Variance

Variance is the average squared deviation of observations from their mean under a specified statistical convention. The units are squared returns, which is why analysts more commonly communicate volatility as the square root of variance. Variance is additive only under particular covariance assumptions; a portfolio’s variance depends on interactions among assets, not merely the average of individual variances.

Standard deviation

Standard deviation is the square root of variance and gives a measure of how dispersed returns are around their mean. It treats positive and negative deviations symmetrically, which can be useful for risk comparison but is imperfect for investors especially concerned about losses. Two return series can have identical standard deviation yet very different worst-case outcomes and return asymmetries.

Downside deviation

Downside deviation measures variation below a chosen target or threshold rather than all variation around the mean. The threshold may be zero or a required return and must be stated. It underlies measures such as the Sortino ratio, but can be unstable when few observations fall below the threshold or when the sample omits a serious future drawdown.

Skewness

Skewness measures asymmetry in a return distribution. Negative skew is associated with relatively extended left-tail outcomes under a statistical definition, while positive skew has a relatively extended right tail. A strategy collecting small gains but occasionally suffering severe losses may exhibit negative skew despite a strong percentage of winning periods. Sample skew estimates can be unstable.

Kurtosis

Kurtosis describes the influence of distribution tails and, in some contexts, extreme observations. Excess kurtosis compares the statistic with the normal-distribution benchmark. A high observed kurtosis can signal that a simple normal model underestimates extreme outcomes, but the estimate may be driven by a small number of exceptional observations. It does not identify when another extreme event will happen.

Value at risk (VaR)

VaR states a loss threshold corresponding to a selected confidence level and time horizon under a model. A one-day 99% VaR of £1 million roughly means a 1% modelled chance of losing more than £1 million over one day, conditional on its assumptions. VaR does not describe how large losses may be once that threshold is exceeded.

Expected shortfall (ES)

Expected shortfall, sometimes called conditional VaR under certain definitions, estimates the average loss in the tail beyond the chosen VaR quantile, using a consistent method. Basel market-risk terminology describes ES as average losses exceeding the VaR threshold. It provides tail severity information missing from VaR, but remains vulnerable to modelling and data limitations.

Confidence level

The confidence level determines what portion of the modelled loss distribution lies at or below the VaR threshold. Moving from 95% to 99% shifts the cutoff into a more extreme tail and generally increases the reported threshold for a given portfolio and method. A 99% confidence level does not mean the portfolio is ‘99% safe’ from loss; it refers only to a particular risk statistic.

Holding period / liquidity horizon

The holding period is the length of time over which the risk calculation applies. A one-day VaR cannot be compared directly with a ten-day measure without matching scaling assumptions and market dynamics. Basel defines liquidity horizon as the time assumed necessary to exit or hedge a position without materially affecting market prices under stress. Liquid and illiquid assets require different treatment.

Historical simulation VaR

Historical simulation revalues or approximates portfolio outcomes using past observed market changes. It can capture some non-normal features of returns without selecting a full parametric distribution. However, it is constrained by the selected sample; an unprecedented shock absent from the data may not be reflected, and old market regimes can be poorly representative of current conditions.

Parametric VaR

Parametric approaches assume a distribution or relationships between risk factors and then calculate the relevant loss quantile. A common basic method uses estimated mean and volatility under a normality assumption. This is tractable but may seriously understate tail risk when price distributions have jumps, skew, fat tails or changing correlations. Results are only as credible as the model.

Stress testing

Stress testing estimates the effects of a specified severe scenario, such as a large rate jump, currency devaluation or liquidity collapse. The scenario may be historical, hypothetical or reverse engineered. It complements VaR and ES by examining plausible breakdowns that a statistical model may miss. The analyst must state the shock, horizon and assumptions about liquidity and correlated markets.

Reverse stress test

Reverse stress testing begins with an unacceptable outcome, such as a loss that breaches a liquidity or capital threshold, and asks what scenarios could cause it. This directs attention toward vulnerabilities hidden by normal return conditions. It is not a claim that the specified disaster is the most probable; it is a tool for identifying failure conditions and mitigation decisions.

Correlation breakdown

Correlation breakdown describes the possibility that relationships between asset returns change in stress, often reducing the protection expected from diversification. Hedged positions can also retain basis risk when related instruments diverge. A portfolio designed using average historical correlations may suffer substantially greater losses if both correlation and volatility shift during a market shock.

Drawdown

Drawdown measures the fall in portfolio value from a previous high to a subsequent low, typically expressed as a percentage of the peak. Maximum drawdown is the largest such decline in the observation period. It is path-dependent and distinct from volatility or VaR. A backtest with a small historical maximum drawdown may simply not include a severe enough crisis to reveal true vulnerabilities.

Worked example

Suppose a hypothetical £10 million portfolio has normally distributed daily returns with zero expected mean and 1% daily standard deviation. A simplified one-day 99% normal VaR is roughly 2.326 × 1% × £10m ≈ £232,600. This is a model quantile, not the maximum loss. Under the same distribution, a corresponding normal expected shortfall is roughly 2.665 × 1% × £10m ≈ £266,500, representing average modelled loss beyond the cutoff.

Now imagine that historical returns include sudden gaps or a highly negative skew that the normal model does not capture. Actual worst-day loss might exceed £500,000 even though the calculated VaR is near £233,000. The arithmetic is still internally correct under its assumptions, but the model fails to capture relevant outcomes. An analyst should use historical, scenario and liquidity analysis rather than treat one quantile as a capital guarantee.

Annualised realised volatility also answers a different question. If the daily return standard deviation is 1%, the usual square-root-of-252 convention yields about 1% × √252 ≈ 15.9% annualised volatility, assuming roughly 252 trading days and appropriate scaling. This is not equivalent to a 15.9% maximum loss or a 15.9% probability of a crash.

How investors and analysts apply the concepts

Risk teams select a common horizon, confidence level, data frequency and modelling convention before comparing VaR across books. They then validate exposures, check whether option positions are appropriately revalued under nonlinear shocks and examine whether modelled loss distributions reflect current volatility. Reports should document which risks are intentionally excluded, such as legal, operational or illiquidity losses.

Expected shortfall is especially useful for understanding the severity of tail outcomes relative to a VaR cutoff, but numerical ES can be unstable because extreme observations are scarce. Scenario analysis can reveal concentrations and interconnected losses not evident in a single loss distribution. A risk committee may need to know what happens if equities fall while correlations rise and a hedge ceases to behave as expected.

Position limits, collateral and liquidity planning should consider whether a modelled stop loss can realistically be executed. A security may gap over an intended exit price. A derivatives position can face margin calls before its theoretical expiry loss limit is reached. Risk metrics describe possibilities under assumptions; they do not directly execute risk controls.

Common errors and limitations

Do not say that 99% VaR represents a 99% probability of no loss. It usually defines the boundary for the more extreme 1% of modelled outcomes over a specified interval. Expected shortfall is not guaranteed to equal the actual worst possible loss. Both can underestimate exposures when data, liquidity or modelling assumptions are inadequate.

A portfolio’s annualised standard deviation is not interchangeable with maximum drawdown, and positive skew does not guarantee low risk. Diversification measured in calm markets may be weaker in crisis. Historical stress tests are informative but cannot reproduce every future shock.

Term comparison

Frequently asked questions

Does 99% VaR mean losses cannot exceed that amount?

No. Losses can exceed the threshold in the modelled tail, and actual risks may be larger.

What is the main difference between VaR and expected shortfall?

VaR marks a cutoff; ES estimates average loss beyond that cutoff under its methodology.

Can low volatility hide tail risk?

Yes. Stable periods can precede rare discontinuous losses or correlation breakdown.

Is maximum drawdown the same as VaR?

No. Drawdown is a historical path measure while VaR is a modelled loss quantile for a specified horizon.

Does a stress test predict a crisis?

No. It estimates sensitivity to chosen severe conditions rather than the probability or exact timing of those conditions.

Sources and further reading

• BIS Basel Framework: Market Risk Terminology
• CFA Institute: Risk Management: An Introduction
• BIS Basel Framework overview

Related Traders MBA resources: trading performance statistics and trading risk terminology.

General education only, not personalised investment, tax or legal advice. Examples are hypothetical, not market forecasts, and investors can incur losses.