The research in this thesis has been split into two chapters. Both deal with textual information and its impact on stock markets. Chapter 1 contains a discussion of the development of a database of textual news that are processed and creates a number of news characteristic variables to measure sentiment, novelty, relevance, ambiguity as well as the topics and entities mentioned in the news. The main contribution of this part is the database itself. We briefly show the descriptive statistics of the data set and also discuss the potential use of data.

The second chapter, based on a draft written jointly with Pedro Saffi, builds on the author's work in the first chapter by testing news variables around earnings announcements. It finds that they indeed have an effect on abnormal returns after them. It focuses on post-earnings announcement drift, trying to evaluate the information value of news around these scheduled events. We look, in particular, at whether our news data can shed some light on information asymmetry and heterogeneous beliefs.