| The financial market is a complex system,where various assets are related and interact with each other.Based on the complex network theory,combine it with financial analysis to further analyze the portfolio model optimization.This paper summarizes the distribution characteristics of the overall return rate of the current domestic stock market based on the analysis of the closing price and the rise and fall of the CSI 300 Index from January 2,2014 to October 17,2022;At the same time,this paper collected the daily closing prices of 200 stocks in the Shanghai Shenzhen 300 Index from February 11,2014 to October 17,2022,a total of 2115 days.Based on Pearson correlation coefficient,Spearman rank correlation coefficient,Kendall rank correlation coefficient,Ising rank correlation coefficient,this paper constructed a multilevel stock correlation network,and constructed different levels of stock correlation networks from linear correlation coefficient,nonlinear correlation coefficient,and tail risk correlation coefficient,and carried out topology analysis,Analyze from the perspective of network connectivity,aggregation and network centrality;Based on the constructed multi-level stock association network,calculate the local clustering coefficient of each layer of network,and use the local clustering coefficient matrix to replace the correlation coefficient matrix to optimize the portfolio model,so as to obtain the optimized portfolio model of different stock association layers.According to the Sharpe ratio,turnover rate,annualized rate of return and daily rate of return indicators outside the sample,further analyze whether the optimized portfolio model has improved the investment indicators compared with the pre optimized portfolio model.Compared with the traditional portfolio model,the optimized portfolio model based on multi-level stock association network takes into account the complex relationship between assets and can consider more diversified investment objectives.It can not only process large-scale and high-dimensional data,but also be more robust for processing non-normal distribution data,outliers,etc.Finally the following conclusions are drawn:1.The time series stock data generally reflect the significant characteristics of nonstationary and nonlinear;2.The topological structures of different correlation layer networks are different.The out of sample Sharpe ratio and average annual rate of return of the portfolio model optimized by different correlation layer networks are better than the traditional portfolio model.In the bull market stage,the portfolio model optimized by the network can obtain higher daily rate of return;3.The factors that affect the Sharpe ratio of the portfolio model mainly include the number of stocks selected and the ratio of holding period to window period length.The number of stocks selected affects the risk dispersion and profitability of the portfolio model after the optimization of the network of different correlation layers.The increase of holding period length mainly affects the profitability of the portfolio model after the optimization of the network of different correlation layers. |
