Not logged in.
Quick Search - Contribution
Contribution Details
Type | Bachelor's Thesis |
Scope | Discipline-based scholarship |
Title | The influence of media vs social media on the financial markets |
Organization Unit | |
Authors |
|
Supervisors |
|
Language |
|
Institution | University of Zurich |
Faculty | Faculty of Economics, Business Administration and Information Technology |
Number of Pages | 31 |
Date | 2015 |
Abstract Text | Uhl et al. (2015) show that there is substantial profit to be made from using a news article based sentiment signal. Not only do they outperform the respective market benchmark but they also decrease the volatility of the returns in the process. Chen et al. (2014) contribute to that notion, however with a slight twist in the procedure. They outperform the market with their social media sentiment indicator as a trade signal. Their returns aren’t as substantial above their market benchmark but there are some discrepancies to be considered. Besides not only using a different indicator, but also compiling their signal in a different manner, they use a different time horizon than Uhl et al. (2015) and a different set of equities to which they apply their backtest. Both of the papers prove, though, that there is certain value-added information content in the news articles and social media contributions that have not been priced into the market, yet, since both strategies outperform their respective benchmark market. On this basis, the following sentiment strategy was developed. Since it has already been highlighted that both, news as well as social media articles, prove to be effective in enhancing equity return, this paper takes a different route in testing whether returns can be improved even further by having a sentiment signal compiled of news as well as social media articles. The strategy that was developed in the framework of this paper simply used the sentiment score calculated by scanning various social media pages (finance related as well as finance unrelated such as Twitter) and news articles for their stock related sentiment. For every stock, all these pages are scanned based on their name or their stock market symbol and the sentiment is given an average score based on the weighting of each sentiment score. This signal is numerated into a scale from -5 (for the worst possible sentiment) to 5 (the best possible sentiment). The strategy is backtested for signaling a buy indication at a score of 4 or higher and signaling a sell indication at -4 or lower. There are no short sells, thus a negative sentiment signal only comes into effect when a stock is already bought and thus needs to be sold in light of the sentiment change for the worse. This signal is accumulated to a weekly average to phase out some noise that could be created through overly negative or positive sentiments. The sentiment signal was gathered for the stocks that make up the S&P500 Index. These stocks were chosen since they are some of the world wide largest companies and are thus more likely to receive a lot of news media, as well as social media attention, whereas small cap stocks are less likely to have a lot of coverage. The risk in that lies in the sentiment signal that could easily be skewed in on direction. In this case, one article can make a big difference in terms of sentiment when there aren’t other articles to average out the noise of this one article. In the backtested period from January 2013 to September 2015, the return averaged 6.015% p.a. over 336 stocks that were backtested (the sample didn’t include the full 500 stocks due to lack of sentiment signal or disruption in the price history, e.g. dilution of stocks, since these events were not captured by the sentiment). The fact that the portfolio of 336 stocks has to be adapted, given the weekly accumulated signal, transaction costs are not negligible and have to be factored into the return. The strategy could not be enhanced in terms of return by setting a minimum news volume. A given stock had to display a minimum level of news attention in order to be eligible for the signal. The idea was that a lot of media coverage or related social media articles would give the best indication of the future development of the stock based on the principle of the “wisdom of crowds”. However, the regression of news volume on return exhibited that the two characteristics where virtually uncorrelated. Volatility on the other side had a more significant explanatory power over returns when only 20% of 4-week volatility or higher was considered eligible. In this case, filtering stock based on volatility was a criterion that was able to enhance the returns of the sentiment strategy. At the same time, this criterion reduced transaction fees for the adaption of the portfolio according to the sentiment signal, since after the selection, the universe of stocks shrunk considerably. It is left to further research, to find the ideal level of minimum required volatility that yields to optimal maximized returns. As a final note, it is worth mentioning that the initial sentiment strategy, that included all 336 stocks does not outperform, but slightly underperform the market on an average annual basis, especially when factoring in transaction costs and fixed costs of subscribing to a software that generates the sentiment signal. |
Export | BibTeX |