Value at Risk - The Probabilistic Loss Estimate
By EC Assets Research Team, Risk Analytics · Published
Value at Risk: Value at Risk (VaR) estimates the loss that a portfolio would not exceed with a stated probability over a stated time horizon. A 1-day 99% VaR of 10 million means there is a 1% chance of losing more than 10 million in a single trading day, under the assumed model.
What Value at Risk Measures
Value at Risk is the loss a portfolio is not expected to exceed with a given confidence over a given horizon. A one-day 99 percent VaR of 10 million says that on 99 days out of 100 the portfolio should lose less than 10 million, and on the remaining day it may lose more, by an amount VaR does not specify.
Every VaR figure therefore carries three hidden choices: the horizon, the confidence level and the model used to estimate the loss distribution. Changing any one changes the number, and a figure reported without all three cannot be compared with anything.
The measure spread through the industry after J.P. Morgan published its RiskMetrics methodology in 1994. Two years later the Basel Committee's market-risk amendment allowed banks to base trading-book capital on their own VaR models, which made it the most widely reported risk number in finance.
How It Works
Three methods turn positions into a loss distribution:
| Method | How it works | Main weakness |
|---|---|---|
| Historical simulation | Revalue today's positions under each day of a 250 to 500 day history; take the percentile | Blind to anything the window did not contain |
| Parametric | Estimate portfolio standard deviation from a covariance matrix; multiply by a normal quantile | Normal tails are too thin for most assets |
| Monte Carlo | Simulate many scenarios from a chosen distribution; revalue and take the percentile | Only as good as the distribution assumed |
Risk teams often run more than one method. When historical and parametric results diverge sharply, recent markets have stopped looking like the textbook.
The parametric version, with the mean set to zero as is usual for short horizons, is:
$$\text{VaR}_{c,h} = z_c ,\sigma, \sqrt{h} \times V$$
where z is the normal quantile (1.645 at 95 percent, 2.326 at 99 percent), σ the daily volatility of returns, h the horizon in days and V the portfolio value. The √h scaling is the square root of time rule, and it assumes independent days. The VaR calculator maps confidence levels to z-scores and shades the tail.
Worked Example
A 100 million portfolio has daily volatility of 1.2 percent:
| Measure | Calculation | Result |
|---|---|---|
| 1-day 95% VaR | 1.645 × 1.2% × 100m | 1.97m |
| 1-day 99% VaR | 2.326 × 1.2% × 100m | 2.79m |
| 10-day 99% VaR | 2.79m × √10 | 8.83m |
Moving from 95 to 99 percent confidence raises the figure by 41 percent; moving from one day to ten raises it by more than threefold.
Backtesting it. Over 250 trading days a correct 99 percent model should be breached about 2.5 times. Suppose the portfolio breaches seven times. If the model were right, seven or more breaches would happen with a probability of about 1.4 percent; under the Basel traffic-light rules five to nine breaches put a model in the yellow zone and ten or more in the red.
Watching diversification raise VaR. Two independent loans of 100 each default with probability 4 percent. Each loan's 95 percent VaR is zero, because default is less likely than 5 percent. Held together, the chance that at least one defaults is 1 - 0.96² = 7.84 percent, so the portfolio's 95 percent VaR is 100. The combined VaR exceeds the sum of the parts, which is what failing subadditivity means.
When It Applies (and Limitations)
It is silent beyond the threshold. Two portfolios with the same 99 percent VaR can have very different losses on the worst day in a hundred. Expected shortfall, the average loss beyond VaR, measures what VaR leaves out.
It can be gamed. A book can be arranged so that large losses sit just beyond the confidence level, for example by selling deep out-of-the-money options. VaR stays low while the tail risk grows.
Calm windows produce small numbers. Historical and parametric estimates calibrated on quiet years understate risk going into a regime change. Stressed VaR, calibrated on a period of significant stress, was added to bank capital rules after 2008 for this reason.
Rare does not mean small. Long-Term Capital Management's models suggested that a one-day loss of more than about 35 million dollars was very unlikely. On 21 August 1998 the fund lost about 553 million, more than fifteen times that figure, and its short, calm data history was a large part of the reason (Jorion, 2000). Only stress testing asks about such days directly.
Why It Matters for Institutional Investors
It is the language of capital. Under the 1996 market-risk rules, a bank's capital charge was at least three times its average 10-day 99 percent VaR, with the multiplier rising as backtesting exceptions accumulate. The Fundamental Review of the Trading Book replaces it with expected shortfall at 97.5 percent; for a normal distribution that is 2.34 standard deviations, close to the 2.33 of 99 percent VaR, and larger whenever tails are fat.
It makes limits enforceable. A single comparable number lets a risk function allocate a budget across desks, check positions before they are traded and escalate breaches the same day.
It forces assumptions into the open. To report VaR a firm must state horizon, confidence, method and data window. That discipline is valuable even when the number itself is fragile.
It is one lens of several. Daily VaR for limits, expected shortfall for tail depth, maximum drawdown for the path actually travelled, stress scenarios for the regimes the data never saw.
References
- J.P. Morgan/Reuters. (1996). RiskMetrics Technical Document (4th ed.). New York: Morgan Guaranty Trust Company.
- Basel Committee on Banking Supervision. (1996). Amendment to the Capital Accord to Incorporate Market Risks. Bank for International Settlements.
- Kupiec, P. H. (1995). Techniques for Verifying the Accuracy of Risk Measurement Models. Journal of Derivatives, 3(2), 73-84.
- Artzner, P., Delbaen, F., Eber, J.-M., & Heath, D. (1999). Coherent Measures of Risk. Mathematical Finance, 9(3), 203-228.
- Jorion, P. (2000). Risk Management Lessons from Long-Term Capital Management. European Financial Management, 6(3), 277-300.
- Lowenstein, R. (2000). When Genius Failed: The Rise and Fall of Long-Term Capital Management. Random House.
- Jorion, P. (2006). Value at Risk: The New Benchmark for Managing Financial Risk (3rd ed.). McGraw-Hill.
Frequently asked questions
Why is parametric VaR often lower than historical VaR?
Parametric VaR assumes normally distributed returns, and real returns have fatter tails. The worst 1 percent of observed days is usually further from the mean than the normal distribution predicts, so a historical simulation over a window that contains stressed days produces a larger number.
Is CVaR the same as expected shortfall?
Yes. Conditional value at risk and expected shortfall both mean the average loss in the cases where losses exceed VaR. The Basel Committee's Fundamental Review of the Trading Book adopted expected shortfall at 97.5 percent confidence as the measure for market-risk capital.
How do 1-day and 10-day VaR relate?
If daily returns are independent and identically distributed with zero mean, 10-day VaR is 1-day VaR times √10, about 3.16. Returns are not independent in stress, when volatility clusters and losses follow losses, so the scaled figure tends to understate risk exactly when it matters.
How is VaR used to set limits?
A risk function sets a firm-wide VaR limit and allocates parts of it to desks or strategies. A new position that would push a desk over its allocation is rejected before trading, and positions that breach it at the end of the day must be reduced or escalated.
Did VaR fail in 2008?
It did what it was designed to do and was asked to do more. Models estimated on calm pre-crisis windows produced small numbers that understated the risk of illiquid credit positions, and VaR by construction said nothing about the size of losses beyond its threshold. The regulatory response added stressed VaR and later moved to expected shortfall.
Stay informed
Market commentary, firm news and research from EC Assets - direct to your inbox.