Stock Movement Prediction with Deep Learning, Finance Tweets Sentiment, Technical Indicators, and Candlestick Charting

Abstract

Stock prediction has been a popular research topic. Due to its stochastic nature, predicting the future stock market remains a difficult problem. This thesis studies the application of Deep Neural Networks (DNNS) in investment from following perspectives: sentiment, stock technical indicators and candlestick charting. In our first experiment, we use DNN to process collective sentiment on the news dataset from Kaggle, and then compare the performance between DNN and traditional machine learning approach. In our second experiment, we build our own dataset that covers 80 stocks from the US stock market. Our attention-based LSTM model shows overall accuracy of 54.6% and MCC of 0.0478 on the aggregate dataset and the best individual stock achieve 64.7% of accuracy. Our third experiment studies the Japanese candlestick charting. In this experiment, harami patterns shows predictive power in our dataset and CNN model on candlestick charting shows great potential in stock market prediction.