What’s in a Word: Economic Applications of Text Mining Methods

This dissertation demonstrates how text can be used to both shape and answer research questions in business and economics research. It contains three papers, which each use a range of natural language processing methods, varying from sentiment to topic analysis. The first chapter investigates whether volatilty in user-generated-content can spill over to volatility in stock returns, and which company events drive these dynamics. The findings show that in particular new product launches increase the volatility in user-generated content and that this volatility can spill over to stock returns. The second chapter of the thesis examines to which extent the Federal Reserve has balanced its dual mandate of price stability and maximum sustainable employment, by studying the topics of speeches by members of the Board of Governors of the Federal Reserve System. In addition, it provides evidence that after the financial crisis the Federal Reserve has unofficially adopted a third mandate of financial stability. The final chapter studies the role of the media as an intermediary of financial information. It compares central bank communications to the articles reporting on these communications to investigate if (and if so, in what way) the media distort information. The findings show that there is a large discrepancy between the original central bank communications and newspaper articles concerning those texts. The type of news outlet, the publication form (online versus print), type of financial text and journalist characteristics (e.g., relevant previous experience and high-quality degrees) all help explain the observed information distortion.