Research On Method Of Stock Clustering Based On Comovement

In recent years,Chinese financial market has been developed rapidly and becoming more and more perfect,playing an important role in the national economic system.Chinese people has formed the conception of investment for the purpose of finance management,which leads to the funds flowing towards the stocks market,main part of the financial market in China.In the investment field,the profit is always accompanied by risks.The higher the return of the investment is,the greater the implicit investment risk is.The most effective method to reduce the investment risk is to avoid the concentration o f investment risk.Stock comovement,one of the most common phenomena in the stock market,is very likely to cause the concentration of investment risks in the investment portfolios,which has attracted the attention of many investors and researchers.At present,t he prediction of the stock comovement phenomenon mainly depends on the category attribute of stocks.Investors predict the arrival of stock comovement phenomenon in advance by studying the herd behaviors of stocks.However,the traditional stock classification system can not satisfy the needs of both investors and researchers.The efficiency of manually completing stock classification is too low to catch up the market updating in time,and the traditional stock classification has relatively bad ability to describe the stock market performance.This paper proposes a numerical indicator that describes the comovement between stocks,making the quantitative analysis of the comovement phenomenon realizable.By applying deep learning technique,the numerical comovement indicator trend can be predicted.At the same time,a proposed fuzzy clustering algorithm is designed to realize the cluster analysis of the comovement phenomenon existing in the stock market,using the distance transformed from the numerical comovement indicator values.The research content of this paper can be seperated into four aspects:The construction of the stock time series data sets.This article collects stock market data from web pages and public data interfaces through crawlers,verifies data consistency through multiple data sources,and preprocesses data formats and structures.The design of the stock comovement numerical indicator.Combining the partial correlation coefficient and the similarity calculated by the dynamic time warping distance,the numerical comovement indicator shows the numerical correlation as well as the morphological similarity.The prediction of the comovement indicator numerical sequences between stocks.The wavelet transform technique and the denoising autoencoder are used to denoise and reconstruct the input samples,and the predicting performance of the LSTM model for the stock comovement indicator numerical sequence has been well improved.The prediction results are used to construct the stock comovement matrix.The realization of the stock multi-categories fuzzy clustering algorithm.Using the comovement indicator between stocks as the similarity and spatial distance between samples,the fuzzy clustering algorithm has been developed to cluster the stocks in the Chines stock market,to explore the potential high-linked stock groups,and to deal with the problem of stocks with multiple categories.