Time series with non-constant variance often have a long-tailed distribution. The data is left- or right-skewed. This can impair the learning process of some algorithms. Methods such as deep neural networks are affected by this issue, unlike tree-based approaches.
