|
The intuitive reason is, I think, that VaR is a quantile estimator and it ignores the extreme points beyond the significant level. (The standard deviation as a measure does not ignore extreme points hence it is subadditive).
The common example:
Bond A, default probability is 4%, with a loss of 100%, otherwise return is 0. The 5% VAR for this bond is actually 0, because 96% probability, return will be zero. The loss (extreme points) is ignored.
Bond B, same default likelihood. and 5% VAR is also 0%. A portfolio of 100 A and 100B will have a VAR(5%) higher than 0. You can do the caculation yourself. The reason is that those ignored extreme points, when combined, are significant.
|