Analyzing and Utilizing Thematic Stocks based on Text Mining Techniques and Information Flow-Based Networks: An Example of the Republic of Korea’s Mask-Themed Stocks

Abstract

In this study, based on the theme sentiment index (TSI) derived from text mining techniques, we measured the causal relationship from the theme sentiment index to the stock price’s abnormal returns for detecting thematic stocks. We selected mask-themed stocks as the experiment subject and set 20 candidate stocks as candidates, considering the frequency of appearance as the associated search term for a mask. Then, we collected search volumes for the keyword “mask” and related keywords from December 1, 2016, to November 30, 2020, to construct the TSI. In addition, we scraped 15,337 text data, such as articles, editorials, and economic broadcast scripts. We also used the abnormal return data of selected 20 stocks derived from the market model. Results show that 19 stocks have statistically significant causal relationships from the TSI to the abnormal returns of their stock prices when the effective transfer entropy is used. We constructed a thematic stock network using the 19 stocks to detect their inner causal relationships. Network-and node-level measures were measured for the constructed thematic stock network for selecting core stocks in the thematic stock network. In addition, two experiments were conducted using the configured thematic stock network. Results confirmed the thematic stock network’s change in behavior and interconnectivity and confirmed that abnormalities such as listing stock misinformation could be detected and empirically analyzed.