Value at Risk
A bad-day loss: the one-day fall that was only exceeded on the worst 5% of days.
5 min readLesson 8 of 57
In short
A one-day 95% VaR of 2% means that over the past year, only about 1 day in 20 lost more than 2%. Expected shortfall goes further: the average loss on those worst days. Neither is a limit; the worst days can be much worse.
Why it matters
Volatility says how much an investment swings on a typical day, in both directions. Most people care more about one direction: how much could I lose on a bad day? Value at Risk, or VaR, answers that in money or percent, which makes it easier to picture than a standard deviation.
It's one of the most widely used risk numbers in finance. Banks report it, fund managers set limits with it, and regulators have built rules around it. Knowing what it says, and especially what it leaves out, keeps you from reading it as a worst case, which it isn't.
How it works
MarketLab works out a one-day, 95% historical VaR. Take the last year of daily returns, about 252 of them, and sort them from worst to best. The VaR is the loss at the 5% line: only about 1 day in 20 lost more. Most tools, MarketLab included, interpolate between the two days either side of that line, so the figure can sit between two actual returns.
VaR says nothing about how bad those worst days were. Expected shortfall, also called conditional VaR, fills that gap: it's the average loss on the days beyond the line. It's always at least as big as the VaR, and on a stock with a few terrible days it can be much bigger.
There's a shortcut, called parametric or normal VaR: assume returns follow a bell curve, and the 95% VaR is about 1.645 standard deviations of daily return below the average. It's a handy cross-check, but real returns have fat tails, meaning extreme days happen far more often than a bell curve predicts. So the normal version tends to understate the worst losses.
MarketLab shows losses as positive percentages: a VaR of 2.1% means a loss of 2.1%. Historical VaR also only knows the year it's built from. If that year was calm, the number will be small, however rough the next year turns out.
95% VaR = the loss at the 5th percentile of daily returns; expected shortfall = the average of the losses beyond it
The loss that only the worst 1 day in 20 went beyond, and how much those worst days lost on average.
A worked example
To keep the numbers small, take a portfolio worth $50,000 with 100 days of returns. Sorted from worst, its five worst days were −4.0%, −3.2%, −2.8%, −2.5% and −2.1%.
- 5% of 100 days is 5 days, so the 95% line sits at the 5th-worst day: a loss of 2.1%. That's the one-day VaR.
- In money: 2.1% of $50,000 = $1,050. On about 1 day in 20, the portfolio lost more than that.
- Expected shortfall is the average of the five worst days: (4.0 + 3.2 + 2.8 + 2.5 + 2.1) ÷ 5 = 14.6 ÷ 5 = 2.92%, or $1,460.
- Cross-check with the normal shortcut. If the daily standard deviation is 1.2% and the average return about 0, the normal VaR is 1.645 × 1.2% ≈ 1.97%.
- The worst single day, −4.0%, was almost twice the VaR, and more than 3 standard deviations from the average, which a bell curve says should almost never happen.
Result: The VaR says a typical bad day costs about $1,050, and the expected shortfall says the worst 5% of days averaged $1,460. Neither caps the loss: the worst day here cost $2,000, and the next bad year can have worse days than the last.
How to read it
- Under 1.5%
- A calm investment over the last year. Broad index funds often land here in quiet years.
- 1.5% to 3%
- Typical of a single large stock. A bad day costs a noticeable amount, but it's normal.
- Above 3%
- A volatile stock, or a very turbulent year. Expect daily losses this size a dozen or so times a year.
- Expected shortfall far above VaR
- The worst days were much worse than the line. That points to fat tails: rare but severe drops, such as crashes after earnings news.
In the real world
- Banks have reported VaR for their trading desks since the 1990s, and regulators long used it to set how much capital banks must hold. After the 2008 crisis, when losses blew far past reported VaR figures, international bank rules moved toward expected shortfall, because it looks at the size of the worst losses, not just where they begin.
- Fund managers often set VaR limits, such as "no more than 2% one-day VaR", and cut positions when a portfolio drifts above them. Because VaR rises after turbulent days, those limits can force selling just after a fall.
- Big losses beyond VaR are not rare surprises. A 95% one-day VaR is expected to be broken on about 12 or 13 days in a normal year, and in a crash several breaks can come in a row.
Common mistakes
- Reading VaR as the most you can lose. It's where the bad days start, not where they end; the worst day can be several times bigger.
- Trusting a VaR built from a calm year. Historical VaR only knows the days it has seen, so it's lowest just before trouble arrives.
- Assuming returns follow a bell curve. The normal shortcut tends to understate how often extreme losses happen.
- Scaling one-day VaR to a year by assuming the days are independent. Bad days tend to cluster, so losses over weeks can be worse than that math suggests.
See it in MarketLab
Stock pages →
The Risk card shows the 1-day VaR (95%) and the expected shortfall for the last year, with a histogram of daily returns and a line at the VaR. Compare NVIDIA with a calmer stock like Procter & Gamble (PG).
Portfolio risk analysis →
Under Risk analysis, the Risk tab shows the same two numbers for your whole portfolio, worked out from its daily returns at today's weights.
Test yourself
Optional: 4 quick questions on the words, using the idea, and how it plays out in real markets.
Loading your answers…