Measuring the Information Content of Financial News

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 3216–3225, Osaka, Japan, December 11-17 2016. Measuring the Information Content of Financial News Ching-Yun Chang1, Yue Zhang1, Zhiyang Teng1, Zahn Bozanic2, Bin Ke3 (1) Singapore University of Technology and Design (2) Fisher College of Business, The Ohio State University (3) NUS Business School, National University of Singapore chang.frannie@gmail.com, yue zhang@sutd.edu.sg, zhiyang teng@mymail.sutd.edu.sg, bozanic.1@fisher.osu.edu, bizk@nus.edu.sg Abstract Measuring the information content of news text is useful for decision makers in their investments since news information can influence the intrinsic values of companies. We propose a model to automatically measure the information content given news text, trained using news and corre- sponding cumulative abnormal returns of listed companies. Existing methods in finance literature exploit sentiment signal features, which are limited by not considering factors such as events. We address this issue by leveraging deep neural models to extract rich semantic features from news text. In particular, a novel tree-structured LSTM is used to find target-specific representations of news text given syntax structures. Empirical results show that the neural models can outperform sentiment-based models, demonstrating the effectiveness of recent NLP technology advances for computational finance. 1 Introduction Information has economic value because it allows individuals to make choices that yield higher expected payoffs than they would obtain from choices made in the absence of information. A major source of information is text from the Internet, which embodies news events, analyst reports and public sentiments, and can serve as a basis for investment decisions. Measuring the information content of text is hence a highly important task in computational finance. For investors such as venture capitals, information content should reflect a firm’s intrinsic value, or potential of future growth. However, measuring the information content of text can be challenging due to uncertainties and subjectivity. Fortunately, for public companies, the stock price can be used as a metric. Our goal is thus to leverage such data on public companies to train a model for measuring the information content of news on arbitrary companies. Finance theory suggests that stock prices reflect all available information and expectations about the future prospects of firms (Fama, 1970). Based on this, empirical studies in economics and finance litera- ture have exploited statistical methods to investigate how stock returns react to a particular news or event, which is called event studies or information content effect (Ball and Brown, 1968). A standard analysis is to measure the cumulative abnormal return (CAR) of a firm’s price over a period of time centered around the event date (termed the event window) (MacKinlay, 1997). Conceptually, a daily abnormal return represents the performance of a stock that varies from the expectation, normally triggered by an event, and can be positive or negative depending on whether the stock outperforms or underperforms the expected return. A CAR is the sum of all abnormal returns in an event window, formally described in Section 2. We study how news affects a public firm’s CAR for training a model …

Content truncated to 500 words