Sentiment-aware jump forecasting

Li, Shaoyu Ning Kaixuan Zhang Teng

This study models the return distributions of the Shanghai Security Composite Index (SSCI) by adding sentiment-aware variables (attention, sentiment, and disagreement), which may affect the jump intensity dynamics or changing the jump size variance, into the GARJI model of Maheu and McCurdy (2004). Textual analysis with some state-of-art machine-learning and deep-learning algorithms is used to select investor sentiment-aware variables with better performance. The extended models (GARJI-sentiment models), which incorporate the sentiment-aware variables into GARJI model, have better forecasting powers on volatilities and extreme events than the benchmark GARJI model. The significant influence of sentiment-aware variables on the jumps and conditional variances implies bounded rationality of investors. Our case study further provides some evidence that Black Swan events, including the implementation of the circuit breaker rule and the lockdown of Wuhan during the COVID-19 epidemic, could affect market jump risks and conditional variances by influencing the sentiment-aware variables, especially investor attention.