Deep stock ranker: A LSTM neural network model for stock selection

Abstract

Stock prediction is a great challenge for the past decades because of the fact that it is a non-stationary, noisy, chaotic environment. Traditional stock prediction models including statistical and machine learning based methods almost use handcrafted features as input. With the development of deep learning, end-to-end models achieve state-of-the-art in many other tasks. However financial time series data is too noise to apply end-to-end models straightly, instead of predicting stocks’ absolute future return, we propose a novel stock selection model DeepStockRanker to predict stocks’ future return ranking. Experimental results show that our method is able to extract information from raw data to predict stocks’ future return ranking and achieves much better performance compared with several advanced models.