Christoffersen conditional coverage test
Tests whether threshold exceedances occur at the expected rate and with the intended independence.
How it works
A conditional coverage test asks whether VaR exceedances occur at the stated rate and whether they cluster beyond what a simple independent sequence would imply. The likelihood comparison uses the numbers of transitions between non-exceedance and exceedance days. Correct overall frequency can coexist with poor timing of risk warnings.
Worked example
Take two 100-day samples that both have five exceedances. In one, the five occur consecutively. In the other, each is separated by many quiet days. An unconditional frequency test sees the same count. The transition table for the first series contains several consecutive exceedances, which the conditional test can detect.
Validation and limitations
Match the forecast window and outcome alignment and check whether overlapping horizons induce dependence. Rare events and small samples can make asymptotic inference unreliable. Consider finite-sample simulation or exact procedures as appropriate. Compare Kupiec coverage.