When machines read the news: Using automated text analytics to quantify high frequency news-implied market reactions

We examine high-frequency market reactions to an intraday stock-specific news flow. Using unique pre-processed data from an automated news analytics tool based on linguistic pattern recognition we exploit information on the indicated relevance, novelty and direction of company-specific news. Employing a high-frequency VAR model based on 20 s data of a cross-section of stocks traded at the London Stock Exchange we find distinct responses in returns, volatility, trading volumes and bid-ask spreads due to news arrivals. We show that a classification of news according to indicated relevance is crucial to filter out noise and to identify significant effects. Moreover, sentiment indicators have predictability for future price trends though the profitability of news-implied trading is deteriorated by increased bid-ask spreads.