Evaluating the Role of Climate-Related Tweets in Predicting Energy Stock Returns

Abstract

As markets grow increasingly efficient, the race to price new information into stocks gets tighter and tighter. Despite millisecond timeframes within algorithmic trading, many would argue not all information is fully incorporated into each firm's share price. This paper follows popular literature, analyzing whether incorporating daily public Twitter sentiment can increase the accuracy of traditional stock prediction models. Uniquely, I analyze whether incorporating population sentiment data from climate-related tweets representing institutional risks enhances energy stock return predictions. With these sentiments aggregated, the data are added to linear and logistic regressions, Support Vector Machines (SVMs), and Neural Networks (NNs) to evaluate their predictive ability. Here, I find climate sentiment factors don't add significant predictive power to models, only increasing accuracy by an average of 1.28%. These predictive results are also applied to trading simulations, finding no clear impact of incorporating climate-related tweet factors, especially once integrating commissions and fees.