Is the predictive power of news sentiment embedded in implied volatility?
No Thumbnail Available
URL
Journal Title
Journal ISSN
Volume Title
School of Business |
Master's thesis
Authors
Date
2020
Department
Major/Subject
Mcode
Degree programme
Finance
Language
en
Pages
54
Series
Abstract
Recent findings show that the predictive power of Google searches on volatility vanishes after controlling for implied volatility. This phenomenon has been shown to apply for equity index returns, commodities and currencies (Basistha, Kurov, & Wolfe, 2017). This study aims to evaluate whether these findings extend to the field of textual analysis, by evaluating the predictive power of negative news sentiment, while controlling for realized-implied volatility spread (RVIV). The findings of this study show that studying the impact of negative news on returns without controlling for RVIV spread can lead to biased results. The dataset consists of 137,737 daily observations for S&P500 stocks from September 10, 2018 to November 1, 2019. A news negativity variable is derived from 429,000 news headlines utilizing Loughran and Mcdonald’s (2011) word sentiment list and a fraction-of-negative-words method. Hypotheses are tested with similar VAR models as Tetlock (2007) to asses the impact of negative news on individual S&P500 stock returns. VAR models are tested separately with OLS regressions under the assumption of independent error terms. Initially, results suggest that after negative news days, returns are 4.2 basis points lower compared to days when there is no negative news, which is aligned with what Tetlock (2007) finds for negative news (-4.4 basis points). However, after controlling for RVIV spread the impact of negative news decreases by 45% to -2.3 basis points and is not significantly different from zero. Simultaneously the impact of RVIV spread is not remarkably affected by negative news sentiment. The findings of this study show that RVIV spread contains additional information about future returns when compared to negative news sentiment. These results suggest that option implied measures, such as RVIV should be controlled for while studying the impact of news sentiment on stock returns.Description
Thesis advisor
Torstila, SamiKeywords
sentiment analysis, implied volatility, forecasting, returns