fulifeng / Temporal_Relational_Stock_Ranking

Code for paper "Temporal Relational Ranking for Stock Prediction"
https://arxiv.org/pdf/1809.09441.pdf
GNU Affero General Public License v3.0
445 stars 167 forks source link

future function #13

Open noke8868 opened 3 years ago

noke8868 commented 3 years ago

"we normalize the price of each stock via dividing it by its maximum value throughout the entire 2013-2017 dataset" in the paper, you used this way to normalize the value. however maximum value during the period is a future function which can be known after the whole period. this is very common mistake to use future function, did you try other normlizeing method? thanks

fulifeng commented 3 years ago

Yes, before we use the maximum value throughout the entire 2013-2017 dataset, we have tried to use the maximum value during the training period. The results are quite close.