Bootstrapping involves repeatedly sampling with replacement from the historical data or residuals to build an empirical distribution. To properly capture the confidence intervals for a GARCH VaR model, the GARCH parameters (e.g., variance equation weights) must ideally be re-estimated for every single generated bootstrap sample. Since GARCH estimation relies on iterative maximum likelihood procedures, repeating this for thousands of bootstrap samples is highly computationally intensive.

Option D is incorrect because bootstrapping is specifically used to avoid relying on the assumption of normally distributed residuals. Option C is incorrect because bootstrapping historical data actually incorporates the extreme events present in that data. Option A is not the primary significant challenge compared to computational intensity.