Return Decomposition

The decomposed approach is proposed as future work rather than completed in the current session. Any log return can be decomposed into direction multiplied by magnitude. The decomposition approach assigns direction prediction to a binary classifier and magnitude prediction to a regression model. A magnitude regression exercise would predict the absolute size of future log returns using the same lag features and temporal split as the classifier. A decomposed return prediction exercise combines the classifier's predicted sign with a magnitude regressor to produce a full predicted log return.