Value at Risk
Value at risk (VaR) estimates the loss a portfolio is not expected to exceed over a set horizon at a chosen confidence level. Formula, methods and limits.
Contents
Value at risk (VaR) is a statistical estimate of how much a portfolio could lose over a set time period at a given confidence level, under normal market conditions. A one-day 95% VaR of $1 million means there is a 5% chance of losing more than $1 million in a day; put the other way, losses should remain below $1 million on about 95 days out of 100. Banks, asset managers and corporate treasuries use VaR to measure and limit market risk.
VaR has three parts: a time horizon (often one day or 10 days), a confidence level (often 95% or 99%) and a loss amount. It is popular because it converts many positions and risks into a single figure in currency, but it marks the threshold of the bad outcomes, not how bad they get beyond it.
Value at risk formula (parametric method)
VaR = z × σ × V
- z is the number of standard deviations for the confidence level, from the normal distribution: 1.645 at 95% and 2.326 at 99% (one-tailed).
- σ is the standard deviation of the portfolio's returns over the horizon, its volatility.
- V is the current value of the portfolio.
This version assumes returns are normally distributed with an expected return of zero, a common simplification for short horizons. To scale a one-day VaR to a longer horizon, multiply by the square root of the number of days: VaRT days = VaR1 day × √T. The square-root-of-time rule assumes daily returns are independent and identically distributed.
Value at risk example
A portfolio is worth $10 million and has a daily volatility of 1.2%.
- One-day 95% VaR: 1.645 × 1.2% × $10,000,000 = $197,400.
- One-day 99% VaR: 2.326 × 1.2% × $10,000,000 = $279,120.
- 10-day 99% VaR: $279,120 × √10 (≈ 3.162) ≈ $882,655.
The 95% figure means that on about one trading day in 20, the portfolio is expected to lose more than $197,400. Over a year of about 250 trading days, that is roughly 12 or 13 days. The VaR figure says nothing about how much more it might lose on those days.
Three ways to calculate VaR
Parametric (variance-covariance) method
Estimate the volatility of each position and the correlations between them, combine them into a portfolio volatility and apply the formula above. It is quick and simple, but it assumes normally distributed returns and positions whose value moves in line with the risk factors, so it handles options and fat-tailed markets poorly.
Historical simulation
Revalue today's portfolio under each day's actual market moves over a past window, such as the last 500 trading days, and sort the resulting gains and losses. The 95% VaR is the loss exceeded in only 5% of the scenarios, about the 25th-worst of 500. It requires no distribution assumption and captures any fat tails that occurred in the window, but it assumes that window represents the future.
Monte Carlo simulation
Generate thousands of random scenarios from a statistical model of the risk factors, revalue the portfolio in each and read the VaR from the distribution of results. It handles options and complex positions well but depends on the model selected and takes the most computing time.
How VaR is used
- Risk limits: trading desks and portfolio managers work within VaR limits, often alongside limits on the Greeks for options books.
- Regulatory capital: banks have long used VaR models to calculate capital for market risk, although newer rules have moved to expected shortfall.
- Corporate risk: treasury teams use VaR, and the related cash-flow-at-risk measure, to size the potential cost of currency or commodity exposure and to assess how much hedging is needed.
Limitations of value at risk
- It says nothing about tail size. Two portfolios with the same 99% VaR can have very different losses on their worst 1% of days.
- Normal assumptions understate extremes. Market returns have fatter tails than the normal distribution, so parametric VaR tends to underestimate how often large losses occur.
- It looks backward. Volatilities, correlations and historical windows come from the past and can change rapidly in a crisis, when correlations often rise together.
- It can penalize diversification. VaR is not always subadditive: the VaR of a combined portfolio can exceed the sum of the VaRs of its parts.
- It ignores liquidity. A one-day VaR assumes positions can be exited or hedged within a day, which may not hold for large or illiquid positions.
Because of these shortcomings, risk managers pair VaR with stress tests, sensitivity analysis and backtesting. Backtesting counts how often actual losses exceed VaR: at 99%, about 2.5 exceptions are expected in 250 trading days, and many more suggests the model understates risk.
Value at risk vs. expected shortfall
Expected shortfall (ES), also called conditional VaR (CVaR), measures what VaR leaves out: the average loss on the days when losses exceed the VaR threshold. Under a normal distribution, 95% expected shortfall is about 2.063 × σ × V, or about $247,500 for the example portfolio, compared with its 95% VaR of $197,400. At 97.5%, expected shortfall is about $280,500, close to the 99% VaR of $279,120 when returns are normal but bigger when tails are fat. Because it captures the size of the tail and is subadditive, the Basel Committee's revised market risk framework for banks replaced 99% VaR with 97.5% expected shortfall.
Frequently asked questions
What does a 95% VaR mean?
A one-day 95% VaR of $1 million means there is a 5% probability that the portfolio loses more than $1 million in one day under normal conditions. It is not the maximum possible loss.
Is VaR the maximum loss?
No. VaR is a threshold that losses exceed with a specified probability, and on the worst days losses can be much larger.
How do you convert one-day VaR to 10-day VaR?
Multiply the one-day figure by √10, about 3.162, assuming returns are independent from day to day. A one-day 99% VaR of $279,120 becomes about $882,655 over 10 days.
What is the difference between VaR and standard deviation?
Standard deviation measures the spread of returns in both directions. VaR turns that spread, or the actual distribution of past or simulated outcomes, into a loss threshold in currency at a selected confidence level, which is easier to compare with capital and risk limits.