Skip to content

New to Omni?·Get 20% off your first purchase with code WELCOME20 at checkout.See plans

FRM® worst-day loss versus VaR: which question is answered?

VaR sets a loss threshold for a chosen confidence level and period. A worst-day analysis looks across a holding window. Learn the distinction with a worked example.

ConceptOmni Curriculum Team6 min read

An open blank notebook, unmarked paper strips and a calculator sit on a navy desk.
On this page

A one-day value at risk, or VaR, is a loss threshold associated with a stated probability over one day. A worst-day analysis instead asks how large the largest daily loss may be during a chosen run of days. The two answers can differ even when they use exactly the same daily loss distribution, because one describes a quantile of one observation and the other describes the maximum of several observations.

Set the question before choosing a number

Start with the period to which the loss refers. A manager who asks for a daily VaR has selected a period, a portfolio, and a confidence level. The answer is a boundary in a distribution of daily losses, not the amount the portfolio will necessarily lose. The Basel Committee terminology describes VaR using a time horizon and a predefined confidence level. Neither element is optional when interpreting a reported figure.

The worst day over a window is a different random quantity. Imagine keeping the same portfolio for a sequence of trading days and recording the daily loss on every day. In each possible sequence, select the largest daily loss. Repeating the thought experiment gives a distribution of those selected maxima. A report might describe the average of those maxima, or a percentile of them, but it must say which. The window length is essential because more opportunities to encounter an adverse day change the distribution of the maximum.

This distinction is easy to lose when both figures are expressed in the same currency. A VaR report can sound like a prediction of the worst loss the desk will see. It is not: losses can cross its threshold. Conversely, the phrase “worst case” can sound like the most extreme physically conceivable outcome. A modelled worst-day result remains conditional on the losses and dependencies admitted by the model. The name alone tells you neither the probability nor the severity of an unmodelled event.

A small worked distribution

Consider an invented, illustrative portfolio whose daily loss has three possible outcomes. The loss is zero with probability 80%, two monetary units with probability 15%, and ten units with probability 5%. The probabilities add to 100%. For this teaching example only, assume the daily outcomes are independent and have the same distribution, and assume the portfolio does not change. These are simplifying assumptions rather than observations about any market.

The 90% one-day VaR is two units under the usual smallest-threshold quantile definition. Losses of zero or two occur with combined probability 95%, while a zero threshold covers only 80%. At a 90% confidence level, the smallest available loss threshold with at least that much probability below or at it is therefore two. The remaining upper-tail outcome can still be ten. Calling two a maximum loss would be an interpretation error.

Now hold the same invented portfolio through five independent days. The largest daily loss is ten if at least one day has the ten-unit outcome. Its probability is one minus 0.95 raised to the fifth power, about 22.6%. The largest loss is at least two if at least one day has either the two-unit or ten-unit outcome. That probability is one minus 0.80 raised to the fifth power, about 67.2%. The maximum is zero in the remaining sequences. These probabilities describe a five-day maximum, while each underlying loss remains a one-day loss.

E[M5]=2(1−0.805)+8(1−0.955)≈3.15E[M_5]=2(1-0.80^5)+8(1-0.95^5)\approx 3.15
Expected largest daily loss in the illustrative five-day window, in monetary units.

The formula values every sequence without listing them individually. The first term gives two units whenever the maximum is at least two. The second adds another eight units in sequences where the maximum reaches ten. The result is approximately 3.15 units. It exceeds the example’s 90% one-day VaR of two, but that ordering is a property of these chosen probabilities, confidence level, and window. A different confidence level could reverse it. The essential lesson is that the two numbers answer different questions.

Do not confuse a maximum with a sum

Five days of exposure creates at least two plausible measures that are both expressed over five days. One is the greatest single daily loss within the window. The other is the total gain or loss accumulated over the entire window. They need not agree. A sequence with one ten-unit loss and gains on other days may have a large worst day but a smaller net loss. A sequence of several moderate losses may have a modest worst day but a larger cumulative loss. Read the definition of the random variable before scaling or comparing it.

The familiar square-root-of-time shortcut sometimes used for an aggregated-horizon VaR does not convert a one-day VaR into an expected worst daily loss over many days. It is based on assumptions about a sum of independent changes and a particular kind of distribution. A maximum is obtained by selecting one observation from a sequence. In the worked example, even the discrete outcome probabilities show why multiplying a daily threshold by the square root of five would answer neither question exactly.

Independence is especially consequential. If losses cluster when markets are stressed, days with large losses may arrive together. The chance of a large maximum, and the chance of a large cumulative loss, then depend on that clustering. A calculation that treats days as independent may give a tidy result while concealing the source of risk. If the portfolio is rebalanced, risk factors shift, or exposures are closed after a loss, a fixed-distribution calculation also ceases to describe the actual path. These are reasons to state assumptions, not reasons to avoid measuring risk.

VaR, expected shortfall and a scenario

Expected shortfall answers another question. It describes the average loss beyond a VaR threshold at a specified confidence level, as the Basel Committee definition states. It supplies information about the size of losses in a tail that VaR alone suppresses. It still needs its period and confidence level. A worst-day analysis over a longer window adds a path dimension: how large the largest single observation may be while the portfolio is held.

A stress scenario is related but has its own meaning. Analysts may specify a coherent adverse change in rates, prices and liquidity, then estimate what the portfolio would lose under those conditions. The Basel Committee stress-testing principles call for scenarios that capture material risks and have internally consistent variables. Such a scenario need not be the expected maximum of a probabilistic daily-loss model. If a question says “scenario”, check whether it provides a specified shock, a distribution of possible days, or a statistical definition of the worst observation.

For exam purposes, write the requested object in words before calculating. “Threshold exceeded in a fraction of one-day outcomes” points to VaR. “Average loss in the tail beyond that threshold” points to expected shortfall. “Largest daily loss selected from each multi-day path, then averaged across paths” points to an expected worst day. “Loss under a stated combination of shocks” points to scenario analysis. Each statement specifies the data transformation the answer requires.

A practical checking routine

  • Identify whether the loss is a daily observation, a sum over days, or the largest daily observation in a window.
  • Write down both the horizon and confidence level when interpreting a VaR or expected shortfall figure.
  • For a worst-day calculation, specify the window, the statistic of the maximum, and assumptions about dependence.
  • Keep a hypothetical model output separate from a claim about an observed event or a guaranteed bound.

GARP places valuation and risk models in FRM Part I, and its study-materials page directs candidates to the learning objectives for the relevant exam cycle. Once the definitions are clear, practise translating a verbal prompt into the random variable it describes before reaching for a formula. The FRM program page gives program context, and the methodology page explains Omni’s approach to study material. The durable skill here is to distinguish a threshold, a tail average, a maximum, and a cumulative loss without letting similar currency figures obscure their meanings.

Sources

  1. Market risk terminology | Bank for International Settlements · bis.org · Retrieved
  2. Stress testing | Bank for International Settlements · bis.org · Retrieved
  3. FRM Study Materials | GARP · garp.org · Retrieved
  4. FRM Exam FAQs | GARP · garp.org · Retrieved

Explore FRM study resources

Educational content, not investment advice. This article is written to help candidates prepare for professional exams. It is not a recommendation to buy, sell or hold any security, and it does not take your personal circumstances into account.

Omni Finance Academy is an independent exam-preparation provider and is not affiliated with or endorsed by CFA Institute, GARP, CAIA Association or CFP Board. CFA®, FRM®, CAIA® and CFP® are trademarks of their respective owners. Full disclaimer.