Research On The Application Of Time Series Clustering Model In Finance

Time series data is a basic and important data object which exists widely in many fields.Information analysis of time series data has gradually become an important research branch in the field of data mining.This paper explores a variety of feasible methods for shape-based time series clustering,including multiple clustering processes such as distance measurement,prototype function selection,and cluster evaluation.Based on dynamic time warping distance and shape-based distance,we introduce related optimization algorithms and compare the application scope of various algorithms and computing resources.On this basis,we perform time series clustering on the stock prices of300 stocks and compare the empirical effects of various distance measures and prototype functions horizontally.Combined with a cluster evaluation procedure,it is considered that the DTW distance has a better effect on the evaluation of clustering effect.Combined with external industry classification labels,we find that the clustering results reflect certain industry aggregation characteristics,but there are still mixed phenomena of different industries in the same category.Further,this paper applies the above clustering results and finds that the stocks of different clusters still show a certain degree of income difference in the next 60 days.Based on this phenomenon,an equal-weight portfolio is constructed to verify that the time series clustering model can help We reduce the risk of portfolio volatility.