Value at Risk, or VaR, is an estimate of the loss that a portfolio should not exceed over a chosen period with a stated level of confidence. For example, a one-day 95% VaR of Rs 20 lakh means losses larger than that are expected on about 5 days in 100.
Three ways to calculate it
| Method | How it works | Weakness |
|---|---|---|
| Historical | Rank past returns and read off the loss at the chosen percentile | Only as good as the period chosen; ignores events that have not occurred |
| Parametric | Assume returns are normally distributed and use volatility times a multiplier | Normality understates extreme moves |
| Monte Carlo | Simulate many possible outcomes from a model | Depends on the model and is heavier to run |
A worked example (parametric)
A portfolio worth Rs 10 crore has a daily volatility of 1.2%. For 95% confidence the multiplier is 1.645. VaR = Rs 10 crore x 1.2% x 1.645, about Rs 19.7 lakh. The inputs are illustrative.
What VaR does not tell you
- The size of the loss beyond VaR. A 95% VaR says nothing about how bad the worst 5% of days are. Expected shortfall (the average loss on days beyond VaR) addresses this.
- Tail risk. Real returns have fatter tails than the normal assumption, so parametric VaR can be too low in a crisis.
- Stable relationships. VaR relies on past volatility and correlations, which can change sharply in a sell-off.
Where it is used
Banks and funds use VaR as a standard risk measure and for limits. In practice it is paired with stress tests and measures such as drawdown. See risk management in quant strategies and volatility targeting.
Frequently asked questions
What is Value at Risk?
It is an estimate of the maximum loss a portfolio is expected to stay within over a set period, at a given confidence level such as 95% or 99%.
What does a 95% one-day VaR mean?
It means that on 95 days out of 100 the loss should be no larger than the VaR, and on about 5 days it could be larger.
What is expected shortfall?
It is the average loss on the days when losses exceed the VaR. It describes the tail that VaR leaves out.
What are the main limitations of VaR?
It gives no information about losses beyond its threshold, can understate extreme events when it assumes a normal distribution, and relies on past data.