| It is the goal of most investors to obtain high profits in the securities market,which depends on the correct judgment of the value of securities.However,due to the different abilities of information acquisition,judgment and analysis,investors come to different conclusions in the face of huge trading data,leading to the deviation of securities price from its own value,A large number of statistical arbitrage operations can correct the pricing and improve the efficiency of the market to a certain extent.Therefore,this paper intends to provide investors with a method of value judgment and propose a reasonable and effective statistical arbitrage strategy based on future-spot arbitrage.In this paper,when we determine the spot portfolio of the duplicate index,considering that the price trend of the index HS300 and its constituent stocks are highly nonlinear and nonstationary,and the data dimension is high,we first use the variational mode decomposition and independent component analysis to extract features of the data and reduct the dimension.Then the dynamic time warping technology is used to measure the similarity between the characteristics of each stock and the HS300 index,which is used for the later improved cluster analysis.In order to improve the clustering effect,this paper integrates particle computing into the clustering algorithm to obtain two different classifications.Finally,the stock is selected according to the industry and the circulating market value,so as to achieve the purpose of a small number of stocks to copy the index.Then,under the setting of dynamic window,the capital matching of the selected stock portfolio is carried out according to the minimum tracking error method,and the no-arbitrage interval is calculated,which is used as the trading signal to carry out arbitrage trading.The empirical results show that the return of the spot portfolio obtained by the two clustering algorithms can reach 34.40%and 33.98%respectively.Compared with the results of ETF portfolio arbitrage,the proposed strategy is more effective.The innovations of this paper can be classified as follows:Firstly,before clustering analysis,variational mode decomposition method and independent component analysis method are used to extract features from high-dimensional data,which improves the efficiency and accuracy of operation.Secondly,the paper introduces the clustering method based on granular computing in the construction of spot portfolios,putting forward two improved clustering algorithm:One,on the base of the K-Means clustering algorithm,determine the initial center under different particle size of division.Then the ability of attribute resolution is introduced in the computing of the inner-class distance and between-class distance,which determine the criterion function as a clustering validity criterion.Two,given that we have already done the feature extraction,from the perspective of clustering character description,the paper uses a simple unsupervised feature selection method.Describe each sample accordingly and then select exemplar description according to the relative frequency.Finally cluster the data under the guide of the selected exemplar description.Thirdly,the sliding window is applied in the process of calculating the spread and no-arbitrage interval of the future-spot arbitrage,and the cost issues involved in the transaction process are comprehensively considered,so as to ensure that the empirical analysis in this paper is closer to the reality. |
