FRM Part II · FRM Exam Part II · Beyond Exceedance-Based Backtesting of Value-at-Risk Models
A risk team reviewing its backtesting framework notes that Expected Shortfall (ES) is not elicitable, whereas VaR is. Which statement best describes the practical implication of this property for backtesting?
Non-elicitability means no scoring function is minimized by the true ES alone, so ES forecasts cannot be compared with a single scoring rule. ES is still testable, for instance jointly with VaR, which is jointly elicitable. It does not mean ES is untestable or incoherent.
- AES cannot be backtested in any form, so regulators must rely solely on VaR exceedance counts
- BES has no scoring function that it alone minimizes, so ES forecasts cannot be ranked directly by a single scoring rule, though ES can be evaluated jointly with VaRCorrect
- CES can be elicited only if the confidence level is above 99.9%
- DES is non-elicitable because it is not a coherent risk measure
Explanation
Elicitability means a statistic is the unique minimizer of the expected value of some scoring function. ES lacks this on its own, but the pair (VaR, ES) is jointly elicitable, so joint scoring and other approaches work. The claim that ES cannot be backtested at all is wrong, and ES is in fact coherent.
Did you get it right without looking?
One question tells you little. A timed set on Beyond Exceedance-Based Backtesting of Value-at-Risk Models shows your real accuracy, how long you take and where you lose marks.
More Beyond Exceedance-Based Backtesting of Value-at-Risk Models questions
- When using a scoring function to compare competing VaR forecasts, which property makes a scoring function 'strictly consistent' for the quan…
- A risk analyst evaluates a bank's daily 99% VaR model using the probability integral transform (PIT). For each day, she computes the value o…
- Two models forecast 97.5% ES for a portfolio. Model A and Model B both pass a standard exceedance-count test on VaR at 97.5%. Over the sampl…
- A risk manager wants to test whether a bank's daily trading P&L is consistent with the full predictive distribution produced by its VaR mode…
- A bank's two VaR models both pass the Kupiec unconditional coverage test at 99%, each with 2 or 3 exceedances in 250 days. A validator wants…
- An analyst applies the Kolmogorov-Smirnov (KS) test to 250 PIT values to check uniformity. The test statistic is defined as: