| In the era of big data,time series have the characteristics of high dimensionality,mass,randomness and low signal-to-noise ratio.The traditional time series model analysis has been unable to meet the requirements of modern analysis.Therefore,it is of great significance to extract information from time series and reduce data dimension and complexity.The HP filtering method and similarity measure used in this paper can extract the long-term trend,fluctuation cycle,similar fluctuation and other information in the data,which is conducive to better time series analysis.The filtering technology originated in the field of electronic engineering.It filters the band of specific frequency in the signal to remove the interference signal.HP filtering is a mature,simple and easy-to-operate high-pass filtering method,which can separate high-frequency components from other components of the signal.Using HP filter to decompose the time series,the long-term growth trend component can be removed and the stable periodic fluctuation component can be obtained.Economists often use HP filter to decompose economic indicators,so as to get the implicit information of economic fluctuations,analyze the economic cycle,and study the relationship between economic fluctuations and growth.At present,the research on HP filtering is mostly limited to direct application,but it still has some shortcomings.Firstly,there is a lack of uniform criteria for smoothing parameters.Secondly,the decomposition results are affected by the sequence length,and the tail solution is unstable.Chapter 4 of this paper puts forward effective solutions to these two problems.In terms of smoothing parameters,a method is proposed to select smoothing parameters by using the coincidence rate of extreme points between the decomposed fluctuation sequence and the volatility sequence.In the modification of tail solution,an extended sequence method is proposed,and the number of extended periods is determined by the modified distance of the wave series decomposed by the extended sequence.In this paper,using the improved HP filtering to filter the annual data of GDP.The fluctuation components are more consistent with the actual economic fluctuation,and the macroeconomic development and change rules can be analyzed more carefully.Similarity measurement is the basis of data mining such as classification,clustering and pattern discovery.In the long time series,the effective similarity measure method is used to search the similarity of its subsequence segments,find similar fluctuations and eliminate redundant information,which can effectively optimize the effect of data analysis.In the fifth chapter,a hierarchical similarity measurementmethod considering the trend and degree of sequence fluctuation is proposed,and the similarity search is realized by programming.In the empirical analysis,firstly,the improved HP filtering method in Chapter 4 is used to filter the stock price time series,and the pure fluctuation sequence which removes the long-term growth trend is obtained.Then,the similarity measurement and search of the pure fluctuation sequence are carried out.Finally,the S-BP neural network model is established based on the similarity search results.The results show that this method has a good similarity measurement effect.It can predict the future stock price by using the stock price of similar fluctuation period in history,which is better than using the recent historical stock price to predict the future stock price trend.In this paper,the prediction analysis based on HP filter decomposition and traditional time series model is proposed for macroeconomic index series,and the prediction analysis based on similarity measure and neural network model is proposed for stock price time series.Empirical results show that the analysis method proposed in this paper for short time series with implied periodic fluctuations and long time series with similar fluctuations has better effect in future value prediction. |
