Skip to content
Criée

Value at risk

A loss, a horizon, a probability: calculate VaR with the normal distribution and from history, and see its limits.

Lesson 99 min

By the end of this lesson

  • Calculate a normal value at risk from daily volatility
  • Interpret a 99% VaR and scale it with the square root of time
  • Explain what expected shortfall adds to the VaR
In this lesson4

A bank holding shares, bonds or currencies wants to know how much it could lose on a bad day. Value at risk answers with a single amount, attached to a probability. What you will learn: calculate a value at risk in two ways, with the normal distribution and from history, and say what it does not measure.

A loss, a horizon, a probability

The 1996 US market-risk rule, issued by the three federal banking supervisors, defines value at risk (VaR) as the estimate of the maximum amount by which the value of positions could decline over a given period at a given confidence level; it requires a daily calculation, at a 99% one-tailed level, for a shock equivalent to ten business days, and allows a one-day VaR to be scaled to ten days by multiplying it by the square root of ten (Federal Register, vol. 61, No. 174, 06/09/1996). These rules apply the Basel amendment of January 1996, in force by the end of 1997 at the latest (BIS, press release of 10/12/1996).

Value at risk (VaR)

The loss a portfolio should exceed, over a given horizon, only in a small share of cases, for example 1% for a 99% VaR. It is not the largest possible loss.

In plain wordsA haulier's bad days

Imagine a haulier who says: on bad days I lose no more than so much, except one day in a hundred. That is value at risk: a loss threshold that is only rarely passed, over a chosen period. It says nothing about that one day in a hundred: on that day the lorry can lose its whole load. And the threshold is worked out from past days: if the road has never been cut, it does not foresee the landslide.

Method 1: the normal distribution

If daily returns follow a normal distribution with a mean close to zero and a standard deviation σ, the loss exceeded once in a hundred is 2.326 standard deviations; once in twenty, 1.645 standard deviations. These numbers come from the normal distribution table.

Writing VV for the portfolio's value, σ\sigma for its daily volatility, zz for the number read from the table (1.645 at 95%, 2.326 at 99%) and hh for the horizon in days, the square-root-of-time rule gives:

VaR≈z×σ×h×V\mathit{VaR} \approx z \times \sigma \times \sqrt{h} \times V

Gervais's fictitious portfolio, 10,000,000 FCFAFictional

Estimated daily volatility: 1.2% (the standard deviation of lesson 6, computed on daily returns).

  1. Standard deviation in francs: 1.2% × 10,000,000 = 120,000 FCFA a day.
  2. 99% VaR, one day: 2.326 × 120,000 = 279,120 FCFA. On one day in a hundred, on average, the loss should exceed this amount.
  3. 95% VaR, one day: 1.645 × 120,000 = 197,400 FCFA: exceeded one day in twenty.
  4. 99% VaR, ten days: 279,120 × √10 = 279,120 × 3.162 = 882,655 FCFA (with √10 unrounded).

Multiplying by √10 assumes the days are independent and alike: variances add up, so the standard deviation grows with the square root of time.

A one-day 99% VaR is 500,000 FCFA. Over 250 trading days, how many days of larger losses should you expect to see, if the model is right?

Method 2: history

Without assuming any distribution, you rank past returns from worst to best and read off the threshold: over 250 days, the 99% historical VaR is the loss on the 3rd worst day (1% of 250 = 2.5, rounded up to the next day). The animation does this on 250 fictitious days, four of them crisis days added to a normal draw.

Value at risk: 250 days of a portfolio10,000,000 FCFA
Confidence:
−6−4−2024

Daily return, in %. In red: the days beyond the threshold.

Historical VaR, 1 day, 99%

390,000 FCFA

the 3rd worst day of 250: −3.90%

Normal VaR, 1 d

279,120 FCFA

2.326 × 1.2% × √1 × 10 M

Expected shortfall, 1 day

463,333 FCFA

average of the 3 worst days

Fictitious portfolio and returns (normal draw with 1.2% daily volatility, plus four added crisis days). Scaling to h days by √h assumes independent, identically distributed days.

Compare the one-day 99% panels: the historical VaR (390,000 FCFA, the 3rd worst day at − 3.9%) exceeds the normal VaR (279,120 FCFA). The crisis days weigh in the history; the normal formula, which knows only the standard deviation, all but ignores them.

What it does not measure

VaR says nothing about the size of losses beyond the threshold. That is why the current Basel framework asks banks using internal models to calculate, every day, the expected shortfall, at a 97.5% one-tailed level, from a base horizon of 10 days (Basel Committee, MAR33.2 to 33.4, version in force from 01/01/2023).

Expected shortfall

The average of the losses that exceed the VaR: the answer to "when things go wrong, by how much on average?".

A VaR computed on a calm past cannot see breaks. A real example from the region: the CFA franc is tied by a fixed parity, and its rate against the anchor currency does not move from day to day; a currency VaR computed on that past would have been almost zero. Yet on 12 January 1994 the CFA franc was devalued by 50%: a pension denominated in CFA francs lost half its value converted into French francs (French Senate, answer to written question No. 09202, 1998). Since the euro, the parity has been 1 euro = 655.957 FCFA (Banque de France, factsheet of 23/04/2025). VaR is a measuring tool for professionals; it promises no ceiling on losses, and a leveraged product can lose far more, as the Trading section shows.

From memory: write the normal VaR of a portfolio worth V, with daily volatility σ, at 99%, over h days. What does the √h factor assume?

Your turn

Portfolio of 5,000,000 FCFA, daily volatility 2%. Its one-day 95% normal VaR is about...

1Portfolio of 5,000,000 FCFA, daily volatility 2%. Its one-day 95% normal VaR is about...

A one-day 99% VaR of 300,000 FCFA means that...

2A one-day 99% VaR of 300,000 FCFA means that...

A one-day VaR is 100,000 FCFA. With the square-root-of-time rule, over 4 days it is...

3A one-day VaR is 100,000 FCFA. With the square-root-of-time rule, over 4 days it is...

The current Basel framework supplements VaR with a measure that looks at...

4The current Basel framework supplements VaR with a measure that looks at...

Read next: Federal Register, vol. 61 n°174 (1996), règle sur le risque de marché

A question? Ask your teacher, or an AI assistant, giving it the title of this lesson.

Share:WhatsAppLinkedIn

Going further