FRM Part II · Market Risk · Free Lesson

Validating Bank VaR Models: Beyond Backtesting

Free GARP FRM Part II lesson in Market Risk. 20 min read, ~2,955 words.

Exception-count backtesting is necessary but not sufficient. A model can pass Kupiec with seven years of clean data, then crater on the eighth when the regime shifts. Supervisors and modelers want validation that goes deeper: sensitivity testing, distribution-based goodness-of-fit, governance review. The probability integral transform turns the entire forecasting distribution into a uniform random variable; if the distribution is wrong anywhere, the PIT histogram bends.

Backtesting based on VaR breaches is binary: the realized loss either exceeded VaR or didn't. That throws away most of the information in the forecast distribution. A model that produces a 2-sigma forecast when the true outcome is right at the mean is wrong but never registers a breach. A model that nails the median and badly misses the tail might pass Kupiec for years before the tail event arrives.

Validation frameworks built around Lynch and post-2010 supervisory guidance push beyond exception counts in three directions:

Read the full lesson, free →
Worked examples and practice. Free with a free account, no card.

Common mistakes

Bottom line

Exam shortcut

When a question describes Kupiec passing but distribution-based tests rejecting, the correct interpretation is that the model has approximately correct exception counts but wrong shape. The remediation is distribution change (Student-t, historical) or vol input adjustment, not exception-count workarounds. PIT histograms with U-shape mean thin tails; peaked PITs mean fat tails. Read the deviation, then remediate.

The full lesson (about 2,955 words, 20 min read) adds 2 worked examples, all 6 common mistakes, a self-check, free in the app.

Learning objectives

Browse all free FRM Part II lessons or jump into free FRM Part II practice questions.