An IB Classification Algorithm for Quantitative Investing Multi-factor Model

This paper proposes a new classification algorithm for quantitative investing multi-factor model. The proposed algorithm is derived from the Information Bottleneck (IB) clustering method. To achieve the goal of categorize stock objects into different class of price changes, two procedures are introduced into the original IB algorithm. By the aid of merging an unknown object into training samples, IB method clusters the data. After that, the class label of the unknown object is voted by all the other samples in the same cluster. Meanwhile, the discretization method is initially applied to satisfy the requirement of IB algorithm. Experimental results show that the proposed algorithm is feasible to classify the stock change level, and the equal width discretization of the class label is superior to the equal frequency discretization.