Problem
1. Why is bagging based on random sampling with replacement? Would bagging still reduce a forecast's variance if sampling were without replacement?
2. Suppose that your training set is based on highly overlap labels (i.e., with low uniqueness, as defined in Chapter 4).
(a) Does this make bagging prone to overfitting, or just ineffective? Why?
(b) Is out-of-bag accuracy generally reliable in financial applications? Why?