摘要
平稳性检验是时间序列回归分析的一个关键问题,已有的检验方法在处理海量时间序列数据时显得乏力,检验准确率有待提高。采用分类技术建立平稳性检验的新方法,可以有效地处理海量时间序列数据。首先计算时间序列自相关函数,构建一个充分非必要的判定准则;然后建立序列收敛的量化分析方法,研究收敛参数的最优取值,并提取平稳性特征向量;最后采用k-means聚类建立平稳性分类识别方法。采用一组模拟数据和股票数据进行分析,将ADF检验、PP检验、KPSS检验进行对比,实证结果表明新方法的准确率较高。
Stationarity test is a key problem of time series regression analysis,existing methods of stationarity test can hardly deal with the massive data,the test accuracy needs to be improved.Based on the analysis of classification,this paper would build a new method for stationarity test,which could effectively deal with massive time series data.Firstly,it calculates the autocorrelation function and then construct a fully and non-necessary criterion;secondly,establish a kind of quantitative analysis of sequence convergence,the optimal value of the sequence convergent parameter is given out,and then take out the characteristic of stationarity;finally,the k-means clustering algorithm is used to establish stationarity classification and recognition method.This paper makes empirical analysis with a set of simulated data and agroup of stock data,which results show that our method＇s accuracy is not only higher than that of the ADF test,and than that of KPSS test and PP test.
出处
《统计与信息论坛》
CSSCI
北大核心
2016年第4期3-8,共6页
Journal of Statistics and Information
基金
教育部青年基金项目<海量金融时间序列数据平稳性检验方法研究>(13YJCZH044)
关键词
时间序列
平稳性
特征提取
分类
time series
stationarity
feature extraction
classification
