Predicting Index Returns from the Market Structure Disagreement Evidence from China

摘要

The factors with information are key to predict stock returns. Previous studies on disagreement mainly focus on the report of the analysts and the sentiment or belief of investors, as well as the trading volume or turnover, ignore the structure among stocks. In this paper, we come up with a new concept of the market structure disagreement and measure it based on the K-means clustering algorithm and Gini impurity. The experiments for the CSI100 stock market show that the market structure disagreement could improve the predicting direction accuracy of machine learning algorithms nearly 1.5%. Specifically, trading volumes and net capital inflows affect the market structure disagreement, which increases with the log dif- ference of trading volumes and decreases with the growth rate of net capital inflows. This paper proposes a new information factor, structure disagreement, which is significantly helpful for investors with market timing, especially for the investors using machine learning.