Hey man, I wouldnt trade on that in real life. The article is written nicely, thanks, but, you have made several severe errors that lead to overfitting and overperformance of your model. One particular error that I havent seen mentioned in the comments yet is the structure of your training data. You mention that you train on a variety of stocks to predict apples stock price. If the training data that you use on the other stocks overlaps in time with the test set that you use on apple, your model suffers from lookahead bias; you are essentially feeding information about other stock’s prices to predict apple at a time when this information had not been available, yet. As these stocks are sometimes highly correlated to apple, of course a model overfit to the traininh set will learn to predict the same patterns on apple quite well, as it sees them. This would not work in actual real-time applications, sadly, but i think that nevertheless there are many correct ideas in here that one can follow.