What Is VaR? Value at Risk Meaning in Finance

Search Fintrev

Search our guides, reviews and glossary.

Value at risk (VaR) is a statistical estimate of the loss threshold a portfolio, position or firm should not exceed over a set time horizon at a chosen confidence level, assuming normal market conditions and an unchanged portfolio. A one-day 99% VaR of $2 million means that on 99 out of 100 trading days the model expects losses to stay at or below $2 million. On the remaining day, the loss can be larger, and VaR alone says nothing about how much larger.

VaR stands for value at risk. In statistics and econometrics, VAR in capitals also means vector autoregression, an unrelated forecasting model.

How VaR works

Every VaR figure is a statement with three parts:

1. Horizon. The period over which the loss is measured: typically one day for trading desks, 10 days in older bank capital rules, and a month or longer for asset managers.
2. Confidence level. Usually 95% or 99%. The complement (5% or 1%) is the probability that losses exceed the VaR figure in any one period.
3. Loss amount. Expressed in currency or as a percentage of portfolio value; multiply a percentage VaR by the position’s current market value to get a dollar figure.

The probability applies to the whole horizon. A 5% one-month VaR of 8% means a 5% chance that the portfolio falls by 8% or more over the month. It does not mean that 5% of the individual days in that month will show losses.

Calculating VaR

Historical method

The historical method applies actual past returns to today’s positions. Each risk factor’s daily percentage change over the look-back window (often 252 trading days, about one year) is applied to current market values, producing one scenario per historical day. The scenarios are ranked from worst loss to largest gain, and VaR is read at the chosen percentile: with 1,000 days of history, a 95% VaR is roughly the 50th worst day and a 99% VaR roughly the 10th worst.

The appeal is that no distribution has to be assumed: fat tails and correlations come straight from the data. The weakness is that the result can only reflect what happened in the window. A calm year produces a low VaR, and a crisis that happened outside the window is invisible.

Variance-covariance (parametric) method

The parametric method assumes returns follow a normal distribution and expresses the loss threshold as a number of standard deviations from the mean. For a single position, with the expected return over a short horizon treated as zero:

VaR = z × σ × V

where z is the standard normal value for the confidence level (1.645 for 95% one-tailed, 2.326 for 99%), σ is the standard deviation of returns over the horizon, and V is the position’s value. For a portfolio, σ comes from the covariance matrix, so correlations between assets reduce the total when they are below 1.

This is fast and easy to explain, and it works reasonably well when return distributions are close to normal and the inputs are estimated from enough data. It performs poorly with small samples, with options and other nonlinear instruments, and whenever real returns have fatter tails than the normal curve allows.

Monte Carlo simulation

Monte Carlo VaR generates thousands of hypothetical future paths for the risk factors from a specified statistical model, revalues the portfolio on each path, and reads the loss at the chosen percentile. It handles nonlinear payoffs and custom distributions better than the parametric method, but it is computationally heavy and only as good as the distribution and correlations fed into the model.

Interpreting VaR

With a 95% one-day VaR, a portfolio is expected to breach the figure about one day in 20, or roughly 12 to 13 trading days a year. A breach is not a model failure in itself; a model that never breaches is probably overstating risk. Risk managers check this through backtesting, counting how often actual losses exceeded VaR. At 99% daily VaR, about 2 to 3 exceptions over 250 trading days are expected; many more suggest the model understates risk.

Example of VaR

Hypothetical example for illustration only. An investor holds a $1,000,000 stock portfolio whose daily returns have a standard deviation of 1.5%. Assume a mean daily return of zero and normally distributed returns.

• 95% one-day VaR: 1.645 × 0.015 × $1,000,000 = $24,675
• 99% one-day VaR: 2.326 × 0.015 × $1,000,000 = $34,890

Moving from 95% to 99% confidence raises the figure by about 41% without any change in the portfolio. That is why VaR numbers from different sources cannot be compared unless the confidence level and horizon match.

To estimate a longer horizon, the one-day figure is often scaled by the square root of time. A 10-day 99% VaR would be $34,890 × √10 ≈ $110,330. This shortcut assumes returns are independent from day to day, volatility stays constant and the portfolio is not traded during the period. All three assumptions tend to break down in stressed markets.

Why VaR is used

• One number across asset classes. Stocks, bonds, currencies, commodities and derivatives are summarized on one scale, which is useful for comparing desks and reporting to senior management.
• Limits and capital. Banks set VaR limits for traders and desks, and firm-wide VaR helps judge whether capital covers potential losses or whether concentrated holdings should be cut.
• Regulation. The 1996 amendment to the Basel Capital Accord let banks use internal VaR models (99% confidence, 10-day horizon) to set market risk capital. The Fundamental Review of the Trading Book later shifted the core measure to expected shortfall.
• Availability. VaR became an industry standard after J.P. Morgan published its RiskMetrics methodology in 1994 and is built into most professional risk and portfolio software.

Limitations of VaR

It ignores the tail. VaR marks the edge of the worst outcomes, not their depth. A 95% VaR of 20% is “validated” by a 50% loss, because the model only claimed that losses of 20% or more would happen about one period in 20.

Inputs drive the answer. There is no single standard protocol for the look-back window, weighting, distribution or horizon. Data from a low-volatility period understates both the likelihood and the size of future losses, and normal-distribution assumptions downplay extreme events.

It is not always subadditive. Portfolio VaR is not the sum of the VaRs of its parts. Diversification usually lowers the total, but with concentrated credit or option exposures VaR can show a combined portfolio as riskier than the sum of its pieces, so VaR is not a “coherent” risk measure in the technical sense.

It assumes positions can be held or exited normally. VaR rarely reflects liquidity: in a crisis, positions may not be sellable at the prices the model uses.

The 2008 crisis. The Financial Crisis Inquiry Commission found that VaR models underestimated both the probability and the magnitude of losses on subprime mortgage portfolios. That supported extreme leverage and left institutions unable to absorb billions of dollars in losses when mortgage values collapsed. The episode is the main source of the criticism that VaR can create a false sense of security.

Because of these gaps, firms pair VaR with stress tests and scenario analysis, which ask what specific shocks (a rate spike, a 2008-style credit event) would do to the portfolio regardless of probability.

VaR compared with related measures

VaR vs. standard deviation and variance

Standard deviation measures how widely returns scatter around their average, in both directions; variance is its square. Neither is a loss amount, and neither specifies a horizon or confidence level. VaR focuses on the loss side and expresses it in currency. In the parametric method VaR is a multiple of standard deviation, but historical and Monte Carlo VaR can diverge sharply when returns are skewed or fat-tailed. The Sharpe ratio uses standard deviation as its risk input.

VaR vs. expected shortfall

Expected shortfall (ES, also called conditional VaR or CVaR) is the average loss in the cases where VaR is exceeded. If the 97.5% VaR is $1 million, ES might be $1.4 million, the mean of all losses beyond that line. ES captures tail depth, is subadditive, and is the measure modern bank capital rules use, but it needs more data to estimate reliably and is harder to backtest.

Marginal, incremental and component VaR

• Marginal VaR is the change in portfolio VaR from a small change in one position, a sensitivity used to see which holdings add risk at the margin.
• Incremental VaR is the change in portfolio VaR from adding or removing an entire position, computed by running the calculation with and without it.
• Component VaR splits total VaR into contributions from each position so that the parts sum to the whole, which makes it useful for allocating risk budgets.

FAQs

The Bottom Line

VaR gives risk managers, banks and portfolio owners one loss figure that spans asset classes and supports limits and capital decisions. It marks where bad outcomes begin and leaves their depth unmeasured, and it depends heavily on its inputs and assumptions, as the 2008 crisis showed. Compare VaR figures only when horizon and confidence match, and pair them with expected shortfall and stress tests.