To evaluate how well a VaR model is performing, analysts usually perform backtesting, such as the Kupiec two-tailed test, which looks at the frequency of exceptions (days when the actual loss exceeds the calculated VaR).

For a one-day 95% VaR evaluated over a 250-day period, the expected number of exceptions is 5% of 250, which equals 12.5 exceptions. A statistically sound model should yield exceptions that fall within an acceptable non-rejection confidence interval (typically around 6 to 21 exceptions for a 95% confidence level over 250 days).

• Option A (0 exceptions): Having zero exceptions indicates the actual losses never exceeded the VaR. This implies that the calculated VaR of GBP 110 million is excessively conservative and overstating the risk, trapping too much capital. This provides convincing evidence of a poorly performing model.
• Option D (7 exceptions): Seven exceptions falls well within the acceptable non-rejection range, meaning it wouldn't serve as evidence of poor performance.
• Option B and C: VaR is a frequency metric (it doesn't bound the magnitude of tail losses). A single massive loss (Option C) or a specific mean loss figure (Option B) are not strictly criteria used to reject the standard frequency-based VaR backtesting, whereas 0 exceptions statistically proves the frequency parameter is wrong.