Skip to content
DP
VaR: Validation & Alternativesadvanced

Christoffersen Conditional-Coverage Test

Christoffersen (1998) sharpened backtesting by testing two things at once. Beyond Kupiec’s correct frequency of exceptions (unconditional coverage), he added an independence test: exceptions should not cluster, because a good VaR adapts to volatility so that one violation does not predict the next. The combined conditional-coverage statistic LRCC=LRUC+LRIND\mathrm{LR}_{CC} = \mathrm{LR}_{UC} + \mathrm{LR}_{IND} is distributed chi-squared with two degrees of freedom (one for coverage, one for independence). A model can pass Kupiec yet fail Christoffersen if its exceptions bunch during turbulent periods, which is exactly the symptom of a VaR that fails to update its volatility.

Why it matters

Getting the right number of exceptions is necessary but not sufficient. If all your violations strike in the same chaotic fortnight, the model is ignoring volatility clustering even though the annual count looks fine. Christoffersen adds the timing check: do exceptions arrive like independent coin flips, or in bursts? The two pieces, right frequency and proper spacing, combine into one test with two degrees of freedom. Clustered exceptions are the fingerprint of a static VaR that should have been dynamic.

Formulas

Conditional-coverage statistic
LRCC=LRUC+LRIND∼χ2(2)\mathrm{LR}_{CC} = \mathrm{LR}_{UC} + \mathrm{LR}_{IND} \sim \chi^{2}(2)
It adds the unconditional-coverage piece (right frequency) to the independence piece (no clustering). With two restrictions it is chi-squared with two degrees of freedom; the 5% critical value is about 5.99.
Independence (Markov) statistic
LRIND∼χ2(1)\mathrm{LR}_{IND} \sim \chi^{2}(1)
Tests a first-order Markov chain on the hit sequence: the probability of an exception today should not depend on whether yesterday was an exception. Rejection signals clustering of violations.

Worked examples

Scenario

A 99% VaR shows exactly 3 exceptions over 250 days, which passes Kupiec comfortably. But all three occurred on consecutive days during a market shock. What does Christoffersen reveal?

Solution

Kupiec passes because 3 exceptions is close to the expected 2.5. However the independence test flags that exceptions cluster: after a violation, the next-day violation probability is far above 1%. So LRIND\mathrm{LR}_{IND} is large, LRCC=LRUC+LRIND\mathrm{LR}_{CC} = \mathrm{LR}_{UC} + \mathrm{LR}_{IND} exceeds the chi-squared(2) critical value of 5.99, and the model is rejected. The clustering shows the VaR is not updating volatility, a static model that should be dynamic.

Common mistakes

  • ✗Christoffersen and Kupiec test the same thing. Kupiec tests frequency only; Christoffersen adds an independence test and combines both into conditional coverage.
  • ✗The conditional-coverage statistic has one degree of freedom. LRCC\mathrm{LR}_{CC} has two degrees of freedom because it tests two restrictions, coverage and independence, with a 5% critical value of about 5.99.
  • ✗Passing Kupiec guarantees passing Christoffersen. A model can have the right number of exceptions yet have them cluster, failing the independence component and hence the joint test.
  • ✗Clustering of exceptions is harmless if the total count is right. Clustering means violations are predictable, the hallmark of a VaR that ignores volatility dynamics, so it is a genuine model defect.

Revision bullets

  • •Christoffersen (1998) tests conditional coverage = frequency + independence
  • •Independence: exceptions should not cluster (no predictability)
  • •LR_CC = LR_UC + LR_IND is chi-squared with 2 degrees of freedom
  • •5% critical value about 5.99; reject if exceeded
  • •A model can pass Kupiec but fail Christoffersen via clustering

Quick check

What does the Christoffersen test add beyond the Kupiec test?

The Christoffersen LRCC\mathrm{LR}_{CC} statistic follows a chi-squared distribution with how many degrees of freedom?

Connected topics

Dynamic VaRGARCH VaRBacktestingKupiec Test

Sources

  1. Christoffersen (1998)
    Christoffersen, P. F. "Evaluating Interval Forecasts." International Economic Review, 39(4), 841-862, 1998.
    Introduces the conditional-coverage test combining correct frequency with independence of exceptions.
  2. Jorion (2007), Ch. 6
    Jorion, P. Value at Risk: The New Benchmark for Managing Financial Risk. 3rd ed. McGraw-Hill, 2007.
    Discusses conditional-coverage backtesting and exception clustering.
How to cite this page
Dr. Phil's Quant Lab. (2026). Christoffersen Conditional-Coverage Test. Derivatives Atlas. https://phucnguyenvan.com/concept/frm-christoffersen-test
← Back to the atlasSee in the network →