In this study, we attempt to investigate how to extract useful information from news headlines and utilize it to predict financial market. We propose four models namely ARIMA, SVM, RNN and LSTM to predict firm-level stock price trend (whether the specific stock price will be up or down in the next day) using information from stock price or news headlines. Specifically, we first employ a time series model, ARIMA model, to predict price trend of seven technology firms, which are Apple, Amazon, Facebook, Google, Microsoft, IBM, and Tesla, using only stock price data. Meanwhile, we also adopt three machine learning methods namely SVM, RNN and LSTM to conduct the price trend prediction task for the same seven firms, using only news headlines. According to our experimental results, we uncover four major findings: firstly, three machine learning methods which only use news headlines data all outperform the conventional method ARIMA model which only uses stock price data, indicating that news headlines can help predict financial market; secondly, the sentiment-based method SVM outperforms the other three none-sentiment-based models in predicating stock price trend, indicating that sentiment is one factor that can be used to predict price trend; thirdly, different news show different predictive power, indicating that only news headlines that are relevant to the firm can makes significant difference in financial market trend prediction ; fourthly, time lags exist in stock price trend forecasting when using news headlines and different firms show different days of time lags, indicating that different investor groups exist in different firms and special news released during that time period.
Details can be found in our report Text Driven Stock Market Prediction Based on Machine Learning Methods -V2.1.pdf.pdf.
We highly refer to another repository EmielStoelinga/CCMLWI
WEI, HU; Yidi, LIU; Dongliang, SHENG; Qiping, WANG; Kai, YANG