Combining Time Series and Sentiment Analysis for Stock Market Forecasting

Abstract

Objective of this research is to build a model to predict stock price using sentimental information from news headlines and historical prices, and the model is able to not only conclude better results but also minimize the difference between predicted values and actual values. News headlines show impact on stock price. Unlike previous approaches where the textual information were usually calculated into sentiment score, we apply various approaches to extract information from news headlines. On the other hand, price data through time series are also useful to predict stock prices. Hence, improvement is made with combination of sentiment analysis of news headlines and time series analysis of historical prices, and the combination is able to complement nonavailability of sentiment lexicon and lack of news. Compared to time series models and word embedding models, our combined model shows smaller or similarly error measures MAPE, MAE, and RMSE with time series models, and reduces lag in graphs.

Incorporating Multi-Source Market Sentiment and Price Data for Stock Price Prediction

“…Previous works on market sentiment mainly focused on sentiment polarity (positive/negative/neutral expression), much research has expanded on this foundation. For example, Chou split news headlines into words and then analyzed the sentiment polarity of each word to calculate sentiment scores for stock price prediction [27]. Cristescu et al analyzed the sentiment polarity of news headlines and used a regression model to predict prices [28].…”