Empirical studies indicate that the relative performance of order statistics and bootstrapping depends on various factors, such as the size of the dataset, the characteristics of the data, and the specific implementation of each method. Neither method has consistently shown to outperform the other in producing tighter confidence intervals for historical simulation VaR. Both techniques are valid but have strengths and weaknesses depending on the context.

A is incorrect: While order statistics can perform well, particularly with smaller datasets, its performance is not universally better than bootstrapping. For larger datasets or when more resampling is possible, bootstrapping may yield competitive or superior results.

B is incorrect: Bootstrapping does not always outperform order statistics. Bootstrapping relies on resampling, which may not provide tighter confidence intervals in small datasets or in cases with high sampling variability.

D is incorrect: Bootstrapping is commonly applied to historical simulation by resampling the historical dataset to generate confidence intervals for VaR estimates. Its application is not limited to parametric methods.