For a 98% VaR model over $T = 252$ days, the expected number of exceptions is $p \times T = 0.02 \times 252 = 5.04$. To determine if the model is poorly calibrated at a 95% confidence level (two-tailed Kupiec POF test), we approximate using the normal distribution: $Z = \frac{x - pT}{\sqrt{p(1-p)T}}$

The standard deviation is $\sqrt{252 \times 0.02 \times 0.98} \approx 2.222$. The critical Z-value for a 95% two-tailed test is $1.96$. Solving for the upper bound limit:$x = 5.04 + 1.96 \times 2.222 \approx 9.39$.

Since exceptions must be whole integers, 10 or more exceptions fall firmly into the rejection region ($Z \approx 2.23 > 1.96$). Therefore, 10 daily losses would lead Roy to conclude the model is incorrectly calibrated (rejecting the null hypothesis).