| Popis: |
In the last decade, natural language processing (NLP) methods have received increasing attention for applications in behavioral economics. Such methods enable the automatic content analysis of large corpora of financial disclosures, e.g., annual reports or earnings calls. In this setting, a conceptually interesting but underexplored variable is linguistic uncertainty: Due to the unpredictability of the financial market, it is often necessary for corporate management to use hedge expressions such as “likely” or “possible” in their financial communication. On the other hand, management can also use uncertain language to influence investors strategically, for example, through deliberate obfuscation. In this dissertation, we present NLP methods for the automated detection of linguistic uncertainty. Furthermore, we introduce the first experimental study to establish a causal link between linguistic uncertainty and investor behavior. Finally, we propose regression models to explain and predict financial risk. In addition to the independent variable of linguistic uncertainty, we explore a psychometric and an assumption-free model based on Deep Learning. |
