Analysis of the results of the paper "Predicting the direction of stock market prices using random forest" (2016) by Khaidem et al. (https://arxiv.org/abs/1605.00003)
Their model is overfitted, since they are splitting the data randomly, using scikit-learns train_test_split
. When the data is splitted, such that the first 80% are used for training, and the remaining 20% are used for testing, the results vanish.
Reddit post: https://www.reddit.com/r/algotrading/comments/cv83yh/overfitting/