Sentiment Analysis for Forecasting: Exploring Volatility-at-Risk, Expected Shortfall, and Innovative Trading Strategies

This paper explores the integration of general market sentiment data and fast Environmental, Social, and Governance (ESG) sentiment in investment decision-making and realized volatility modelling. (ESG) Sentiment data is constructed using natural language models and is based on news articles and social media data. Traditional ESG scores are based on companies self reported information and hence are prone to greenwashing. Fast ESG scores are more timely and based on more independent data sources. In this paper I investigate the benefits of using fast ESG sentiment to measure ESG performances as opposed to traditional slower ESG scores. The study demonstrates how to effectively process sentiment data using the Kalman filter. I present two applications: Firstly, sentiment-based trading strategies that surpass the performance of the S&P 500 with better risk/return characteristics. Secondly, the utilization of Quantile Random Forest and GARCH-F models enhanced with sentiment to effectively model tail risk in realized volatility. I show that the Quantile Random Forest models augmented with sentiment perform better than the traditional GARCH-F in volatility-at-risk and expected shortfall modelling.