FRM Part II · FRM Exam Part II

Backtesting VaR for FRM Part II

Backtesting VaR compares a model's predicted loss limit with actual daily results to check that exceptions occur as often as the confidence level implies. Count the days when the loss exceeds VaR, test that count with the binomial or Kupiec test, check for clustering with Christoffersen, then place the result in a Basel zone and interpret it.

What this chapter covers

Backtesting is how you find out whether a VaR model works. A 99% one-day VaR should be exceeded on about 1% of days. You count the actual exceptions and ask whether that count is believable under the model. This chapter gives you the tools: the binomial logic, the Kupiec test for frequency, the Christoffersen test for independence, and the Basel traffic light framework that links exceptions to capital.

The chapter is mostly about interpretation. You need to compute an expected number of exceptions, run a likelihood ratio test against a chi-square critical value, and then say what the result means. Questions also ask why exceptions happen, such as a flawed model, trading inside the day, or plain bad luck, and what a risk manager should do next.

It connects to the rest of Part II in three ways. It tests the VaR and expected shortfall models from Market Risk Measurement and Management. It feeds the Basel capital multiplier, so it links to regulatory capital. It also reflects model risk, which comes up again in Current Issues, for example with AI models.

Backtesting questions are applied and fairly predictable. You are given a number of exceptions, a sample size and a confidence level, and you must pick the correct conclusion. Most of the work is a few short formulas and a clear decision rule, so this chapter rewards practice more than memory. Candidates who learn the rules with their exact conditions, such as degrees of freedom and zone boundaries, pick up reliable marks in an exam where every question carries equal weight.

Backtesting VaR: topics in the order to study them

1Purpose and Objectives of VaR BacktestingStart here to learn what is being tested and why, including the difference between actual and hypothetical P&L.
2VaR Exceptions and the Binomial TestExceptions follow a binomial distribution, and every later test builds on that idea.
3Kupiec Unconditional Coverage TestThis turns the binomial idea into a likelihood ratio test with a chi-square decision rule.
4Christoffersen Conditional Coverage and IndependenceIt extends Kupiec by adding a test for clustering, so learn Kupiec first.
5Type I and Type II Errors in BacktestingOnce you know the tests, you can judge their weaknesses, mainly low power at high confidence levels.
6Basel Traffic Light ApproachThis applies the error trade-off in a regulatory rule with fixed zones and capital add-ons.
7Causes of Exceptions and Model ImprovementFinish with diagnosis and action, which pulls the whole chapter together in case-style questions.

How to prepare Backtesting VaR

Work from the logic to the numbers, then practise decisions. Keep sessions short so you can revise on a phone.

Write the core idea in your own words: at confidence level c, the exception probability is p = 1 − c, and the expected count is p × N.
Learn the binomial set-up. For 250 days at 99%, the expected number of exceptions is 2.5 and the standard deviation is √(250 × 0.01 × 0.99) ≈ 1.57. Practise judging whether a count looks too high.
Memorise the Kupiec statistic and its rule. The likelihood ratio uses chi-square with 1 degree of freedom, and the 5% critical value is 3.84. Reject the model if the statistic exceeds it.
Learn Christoffersen as two parts. The independence test has 1 degree of freedom, and the conditional coverage test combines it with Kupiec and has 2 degrees of freedom. The 5% critical value for 2 degrees of freedom is 5.99.
Learn the Basel zones for 250 observations of 99% one-day VaR: green is 0 to 4 exceptions, yellow is 5 to 9, and red is 10 or more. Know that the base multiplier is 3 and that the add-on rises through the yellow zone to 1.00 in the red zone.
Practise explaining Type I versus Type II errors in one sentence each, and why a 99% backtest has little power to catch a bad model.
Finish with mixed questions. For each one, state the measure, the method and the interpretation, then name the likely cause of exceptions and the fix.

Common mistakes in Backtesting VaR

Using the wrong exception probability, such as 99% instead of 1%.
Fix: Always write p = 1 − c first, then compute the expected count and the test.
Using the wrong degrees of freedom or critical value for Kupiec and Christoffersen.
Fix: Link each test to its value: unconditional coverage 1 degree of freedom (3.84), independence 1, conditional coverage 2 (5.99), all at 5%.
Treating Kupiec as a test of clustering.
Fix: Remember that Kupiec counts exceptions but ignores when they occur. Only the independence test checks for clusters.
Mixing up Type I and Type II errors.
Fix: Type I means rejecting a correct model. Type II means accepting a wrong one. Then link the second to low power at 99%.
Misstating Basel zones or applying them to the wrong sample.
Fix: State the conditions every time: 250 observations, 99% one-day VaR, green 0–4, yellow 5–9, red 10 or more.
Assuming every exception means the model is broken.
Fix: Separate the cause first. Check data and P&L definitions, then the model's accuracy, then consider chance, before recommending a fix.

Last-day revision: Backtesting VaR

An exception is a day when the loss exceeds the VaR estimate.
Expected exceptions = p × N, where p = 1 − confidence level.
Exception count is binomial with standard deviation √(N × p × (1 − p)).
Kupiec tests the exception frequency only; it ignores timing.
Kupiec LR_uc follows chi-square with 1 degree of freedom; 5% critical value is 3.84.
Christoffersen independence tests whether exceptions cluster; 1 degree of freedom.
Conditional coverage LR_cc = LR_uc + LR_ind, with 2 degrees of freedom; 5% critical value is 5.99.
Basel zones at 250 days and 99% VaR: green 0–4, yellow 5–9, red 10 or more exceptions.
Basel base multiplier is 3; yellow adds a rising amount and red adds 1.00.
Type I error rejects a correct model; Type II error fails to reject a wrong model.
Higher confidence levels give fewer exceptions and lower test power.
Causes of exceptions include model flaws, intraday trading, and bad luck.

Backtesting VaR practice questions

Backtesting VaR in other exams

The same ground in other exams, if you are preparing for more than one or want another angle on it.

Backtesting VaR: frequently asked questions

How many exceptions are expected for 99% VaR over 250 days?

The expected number is 250 × 0.01 = 2.5 exceptions. The standard deviation is about 1.57. A count well above 2.5 casts doubt on the model, and the formal tests tell you how much.

What is the difference between Kupiec and Christoffersen tests?

Kupiec tests only whether the number of exceptions matches the expected frequency. Christoffersen adds a test for independence, which checks whether exceptions cluster. The conditional coverage test combines both and uses 2 degrees of freedom.

What are the Basel traffic light zones?

For 250 observations of 99% one-day VaR, 0 to 4 exceptions is green, 5 to 9 is yellow, and 10 or more is red. The capital multiplier starts at 3 and rises in the yellow zone. In the red zone the add-on is 1.00, giving a multiplier of 4.

Why is backtesting at 99% considered weak?

At 99%, exceptions are rare, so a short sample gives little evidence. A model that is clearly wrong may still produce an acceptable count, which is a Type II error. Lower confidence levels give more exceptions and more power.

Do I need to calculate the Kupiec statistic by hand in the exam?

You should know the formula and be able to interpret it against 3.84. Calculator-based questions are possible, but many questions give the statistic and ask for the conclusion. Practise both.