A comprehensive survey on deep neural networks for stock market: The need, challenges, and future directions

Abstract

The stock market has been an attractive field for a large number of organizers and investors to derive useful predictions. Fundamental knowledge of stock market can be utilised with technical indicators to investigate different perspectives of the financial market; also, the influence of various events, financial news, and/or opinions on investors’ decisions and hence, market trends have been observed. Such information can be exploited to make reliable predictions and achieve higher profitability. Computational intelligence has emerged with various deep neural network (DNN) techniques to address complex stock market problems. In this article, we aim to review the significance and need of DNNs in the field of stock price and trend prediction; we discuss the applicability of DNN variations to the temporal stock market data and also extend our survey to include hybrid, as well as metaheuristic, approaches with DNNs. We observe the potential limitations for stock market prediction using various DNNs. To provide an experimental evaluation, we also conduct a series of experiments for stock market prediction using nine deep learning-based models; we analyse the impact of these models on forecasting the stock market data. We also evaluate the performance of individual models with different number of features. We discuss challenges, as well as potential future research directions, and conclude our survey with the experimental study. This survey can be referred for the recent perspectives of DNN-based stock market prediction, primarily covering research spanning over years 2017-2020.