Sulkhan Metreveli, University of Zurich - Institute of Mass Communication and Media Research
Abstact
Recent years have seen an increasing interest in 'Big Data'. This interest has been driven by regulatory challenges on the one hand (e.g., privacy issues), and the enormous potential to uncover and predict social behaviour on the other. This paper contributes to the latter: it examines the influence of media information on price changes in stock exchange and foreign exchange markets. It tests the possibility of forecasting price changes in international stock exchange and foreign exchange markets by uninterrupted processing and analysis of incoming data streams from about 300 media outlets. The results demonstrate a major correlation between media sentiment and market prices. Moreover, differentiation of media outlets shows that third generation information delivery systems (such as Bloomberg Terminal or Reuters Xtra 3000) have the most significant effect on price changes, while first generation information delivery systems (such as printed newspapers) have no influence. The effects of news on price changes strongly depend on the speed of information diffusion. The observed effects give rise to questions of political implications. e.g., the question of accountability for news reporting in highly influential third generation information delivery systems, the problem of information diffusion asymmetry in terms of equal access to information, and, more generally, questions regarding the political governability of an information ecosystem in which human action is partly replaced by technological architectures.
Background and Research Question
In recent years, the amount of news data available to researchers has reached immeasurable levels, providing new opportunities for social science research. ‘Big Data’ allows for real-time monitoring of social action to uncover social behavioural patterns. More comprehensive approaches even discover major factors of influence on social change and allow for prediction of future societal trends. From a communications science point of view, questions regarding the influence of media information on social behaviour have been at the core from the beginning. To what extent and how does media information affect social awareness, attitude and action? The availability of ‘Big Data, and innovations in data processing and analysis may provide new answers to old questions. This paper contributes to the analysis of media effects. In particular, it examines the role and the influence of news media on price changes in the world’s most important international markets – the New York Stock Exchange and the Foreign Currency Exchange.
Method
The analysis builds on a database that covers 70,000 news from over 300 different media outlets. The news sources include, among many others, the news agencies Reuters and Bloomberg and their corresponding terminals. Analyses are conducted using fully automatic quantitative content analysis based on a novel news gathering and analysis tool - Calfor (Computer Assisted Linguistic Forecast). The tool, programmed in C++, enables automatic searching and gathering of huge amounts of information from targeted sources in a very short time. In addition to the data gathering function, Calfor contains a built-in instrument which automatically archives all available data and detects its sentiment (positive, negative or neutral). In an eight-month field-phase Calfor searched for up-to-date information from a wide range of World Wide Web sources on a minute-by-minute basis.
Using Calfor, we tested if it was possible to forecast price changes in stock exchange and foreign exchange markets by uninterrupted processing and analysis of incoming data streams. The study utilized “Granger causality” in order to detect causal relationships between the sentiment in media reporting on the one hand and market price action on the other.
Findings
The results of the analysis show a major correlation between media sentiment and market prices, and demonstrate that markets as well as media are highly self-referential. Additionally, the findings show that news effects on market price changes strongly depend on the type of media outlet in question and on the data in use. The analysis identifies three forms of news and data delivery systems, which differ according to the channels of distribution and the type of data provided to the end-user. First generation information delivery systems (such as printed newspapers) use “traditional” channels for information delivery and are the slowest in reaching the end-user. The information delivery time varies here between 2 to 24 hours on average. Second generation information delivery systems (such as websites of news agencies, information portals, and social media applications) are fast and their information delivery time varies between several minutes to one hour. Third generation information delivery systems (such as Bloomberg Terminal or Reuters Xtra 3000) are the fastest, and their data delivery time varies between several seconds to several minutes.
According to our findings, the first generation systems occurs to have no effect on market prices, while third generation systems have the strongest and the most significant effects. Moreover, the findings show that first and second generation information delivery systems are highly affected by price actions. In contrast, third generation information systems have significant effects on price changes. Accordingly, our research indicates that the effects of news on price changes heavily depend on the speed of data delivery or, as we call it, information diffusion.
Policy Implications
The strong influence of third generation information delivery systems raises questions about policy implications. First, the question of accountability for news reporting in fast and highly influential third generation information delivery systems; second, the problem of information diffusion asymmetry in terms of social equality regarding access to information; third, and more generally, the question of governability of a complex adaptive information ecosystem in which human actors are partly replaced by technological architectures.