Financial data division and rules mining based on influence and AP clustering

Abstract

The amount of data in financial data is enormous and mining it has a great value. For stock market, how to effectively select stocks from a reference sector is very important for investors. Based on the co-movement effect between stocks, this paper introduces a concept which is the stock's influence and constructs the influence matrix by using the time series of stocks. Then we divide the new stock sector by using the concept of responsibility and availabilities in Affinity Propagation (AP)cluster. In our experiment, we conducted experiments on more than 2000stocks in 4 different periods of time. Experimental results show that the sector division method has better cohesion, and there are some interesting rules such as: The number of sectors changes with the market up and down, as well as part of the industry's stock often gathered together.