Machine learning in predicting stock indexes: the role of online stock forum sentiment in MIDAS model

Abstract

This study aims to accurately predict stock indexes by combining sentiment analysis with machine learning. We apply web crawlers to collect text information from a representative Chinese stock forum, build a high-frequency investor sentiment index, and select a suitable mixed-data sampling model to make nowcasting predictions on the Shanghai Composite Index (SHA). We show that the investors’ sentiments significantly drive the SHA, and that the exchange rate is the most powerful indicator for short term SHA prediction. Additionally, no autoregressive effect exists on the SHA. These results will benefit investors and policymakers.