Research On The Clustering Analysis Based On Time Series Characteristics In Margin Trading And Stock Trading

The clustering analysis of time series is different from the clustering analysis of general panel data,but based on the classification algorithm of unsupervised learning,it can excavate the information of deeper and higher dimensions in the study of the correlation between different sequences,Common in all kinds of fractal and pattern recognition based on quantitative transactions.The traditional time series similarity determination method,through the test sequence and the target sequence corresponding to the point of time between the Euclidean distance to determine the completion of clustering.However,the traditional methods have different defects on the missing points caused by different sampling frequencies and the processing of abnormal outliers.The algorithm can cause large deviation and the clustering effect is unsatisfactory.In this paper,we give a new method of similarity measure-feature factor discrimination.By extracting the feature characteristics of time series and non-linear feature,we can map one-dimensional time series to high-dimensional space,The eigenvector is constructed and the multicollinearity problem between the eigenvectors is eliminated by principal component analysis.Then,the K-means unsupervised learning algorithm is used to classify and complete the clustering analysis.In this paper,the relationship between the transaction data and the stock price trend of the financial margin is taken as an example.By extracting the feature vector of the characteristic factor,the sample matrix is divided into the sample and the sample matrix is used as the input sample of the K-means algorithm,The unsupervised learning classification,to achieve the effect of cluster analysis,theoretically,the clustering analysis of the similarity of the portfolio should be consistent with the trend of the market index has a strong positive correlation,but also has a strong trend Trend,and for the trend of the object,to quantify the timing of the study provides a feasibility.Compared with the traditional method,the influence of the missing points of different length time series data and the anomalous outliers is reduced,and the similarity degree of cluster analysis and the similarity of time series identification are improved.On this basis,to explore its application in the actual market,the portfolio structure clustering analysis,explore the relationship between the index and the market in the sample and explore the trend of the price difference,through the combination of low delay average system EMA index is constructed for correction,for the price difference of transaction timing observation its performance in the sample,we confirmed the feasibility of similarity through cluster analysis are combined and quantified by selecting low latency and the rationality of the average system.