On XLE Index Constituents’ Social Media Based Sentiment Informing the Index Trend and Volatility Prediction

Abstract

Collective intelligence represented as sentiment extracted from social media mining found applications in various areas. Numerous studies involving machine learning modelling have demonstrated that such sentiment information may or may not have predictive power on the stock market trend. This research investigates the predictive information of sentiment regarding the Energy Select Sector related XLE index and of its constituents, on the index and its volatility, based on a novel robust machine learning approach. While we demonstrate that sentiment does not have any impact on any of the trend prediction scenarios investigated here related to XLE and its constituents, the sentiment's impact on volatility predictions is significant. The proposed volatility prediction modelling approach, based on Jordan and Elman recurrent neural networks, demonstrates that the addition of sentiment or sentiment moment reduces the prediction root mean square error (RMSE) to about one third. The experiments we conducted also demonstrate that the addition of sentiment reduces the RMSE for 24 out of the 36 stocks/constituents, representing 87.9% of the index weight. This is the first study in the literature relating to the prediction of the market trend or the volatility based on an index and its constituents' sentiment.

Predicting S&P 500 Based on Its Constituents and Their Social Media Derived Sentiment

“…We prove that our constituent and sentiment based approach is efficient in predicting S&P 500, and thus may be used to maximise investment portfolios regardless of whether the market is bullish or bearish 4 . This study extends our previous recent work on XLE index constituents' social media based sentiment informing the index trend and volatility prediction [13].…”

“…We currently develop further work on exploring the extension of this approach and of the approach proposed in our recent work [13], for several stock market indices.…”

A Two-Step Optimised BERT-Based NLP Algorithm for Extracting Sentiment from Financial News

“…This aspect supports the rationale behind the time sensitivity of the statistical significance of sentiment variables. Financial news could be the source of new information, expressed as sentiment, which has proved to be useful in enhancing stock market prediction with statistical and machine learning approaches (Smales [20], Shiller [25], Olaniyan et al [26], and Marechal et al [12]). Advanced NLP approaches are powerful tools that can be used to reliably and effectively extract sentiment polarity information from financial news, and we propose such a novel approach here based on adapting and extending the BERT algorithm [16].…”