Textual analysis in stock picking: additional evidence on (dis)similarity

This study investigates the predictive capability of textual similarity within the context of financial news, drawing upon empirical evidence from prior research. Two hypotheses are formulated and subsequently validated through analysis: firstly, that the similarity among financial newspaper articles concerning the same company can forecast future stock performance, and secondly, that a negative relationship exists between textual distance (TD) and future returns. The analysis demonstrates that employing textual similarity as a selection criterion can result in effective stock selection. This observation holds true across various similarity measures and remains robust even in out-of-sample tests. Additionally, the study emphasizes the significance of specific similarity measures, specifically Minmax and Jaccard, in predicting abnormal stock returns. Finally, we assess the ability of TD to construct stock portfolios capable of generating excess performance, finding that the low TD portfolio consistently exhibits statistically significant excess performance compared to the high TD portfolio.