To determine if the model is well-calibrated, we perform a Kupiec test (two-tailed) or calculate the expected upper bound of exceptions using the normal approximation. The expected number of exceptions for a 98% VaR model over 252 days is: $\mu = p \times N = 0.02 \times 252 = 5.04$

The variance is: $\sigma^2 = p \times (1-p) \times N = 0.02 \times 0.98 \times 252 = 4.9392$ Standard deviation $\sigma = \sqrt{4.9392} \approx 2.222$

Using a normal approximation for a 95% confidence level (two-tailed), the critical value $Z = 1.96$. The upper limit of the non-rejection region is: $\text{Upper Bound} \approx 5.04 + 1.96 \times 2.222 = 9.39$

Since 10 is the first integer outside the non-rejection region ($10 > 9.39$), 10 daily losses will lead to the conclusion that the model is incorrectly calibrated.

Alternatively, using Kupiec's approximate test statistic $LR \approx \frac{(x - \mu)^2}{\sigma^2}$ with a 95% chi-square critical value of $3.841$(for 1 degree of freedom): For$x = 10$,$LR \approx \frac{(10 - 5.04)^2}{4.9392} = 4.98$, which is$> 3.841$. Therefore, we reject the null hypothesis at 10 losses.