The Relationship between Media Information and Stock Returns Based on Text Semantic Mining Algorithms

Abstract

We use text semantic mining algorithms based on intelligent search engine framework to obtain media information data of stocks. Since media is significantly related to firm size, industry affiliation and whether belongs to important index, we adopt event study and use the residual attention model to examine the relationship between abnormal media information and stock returns with a special sample. We find that relative to stocks with high abnormal media information, those stocks with low abnormal media information have higher returns. The "media effect" exists in Chinese stock market. A long-short trading strategy can earn significant positive cumulative excess returns in the following 10 days. Furthermore, our findings show that the excess return from "media effect" is due to the significantly low returns of high abnormal media information stocks. We suggest that the explanation of this asymmetry phenomenon is possibly the stock price's overreaction to media reports caused by investor sentiment, which yields lower expected returns.