[amy214]

The paper ignored a big honking chunk of the field, as is used in finance often - you take the delta between adjacent timepoints, something like the derivative, now you have something centered on zero and going up and down "randomly". Something like PCA also typically may use centering although that's not exactly the same. I would have loved to see an investigation of window size and such and deltad time series (usually either x2-x1, or x2/x1). In this sense the paper is just recapitulating, "if you don't delta your time series bad things happen"