Panel Data Mean Change Point Detection And Application Research Based On CUSUM Test Statisti

In the era of big data,dynamic data models are widely used in the current fields of medicine,climatology and financial markets,and are often used to examine the structural properties of data.When it is found that there is structural heterogeneity in the observed data series,it is necessary to consider that the observed data may come from a different distribution of the population when performing data value analysis,and the use of segmented data models needs to be considered for statistical analysis and data inference.At this stage,many scholars have studied how to accurately and effectively detect the heterogeneous structure in the observed data,but there are still relatively few studies on the computational complexity of detection,and this study mainly discusses the problem of reducing the complexity of data change point detection and improving the detection accuracy.In the study of mean variable points of data structures,it is found that the traditional BS algorithm and the global CUSUM statistic have the disadvantages of high computational complexity in the process of sorting out the literature,and this study considers the use of local CUSUM statistic,and uses peak detection and T test in the step of distinguishing change points from non-variable points,and proposes a BS-T variable point detection algorithm.Then,the applicability of BS algorithm,Shape-Based BS algorithm and BS-T algorithm is compared through data simulation.Furthermore,the daily closing price data will be used as samples to further explore the detection efficiency of the BS-T algorithm in practical applications,and also prove the effectiveness of BS-T algorithm under large sample conditions.Considering the timeliness and continuity of variable point detection,based on the local CUSUM test statistics,the cut-off values of variable point and non-variable point are calculated by self-service test method and asymptotic test method.The critical values of self-service test and progressive test under different data structure and parameter conditions were determined by multiple Monte Carlo tests,and the test level,detection efficiency and numerical distribution status of detection change point position of the two methods under different conditions were solved,and the advantages and disadvantages of self-service test and progressive test under different conditions were compared and analyzed.The final results show that the self-service inspection method can meet more data structure conditions and parameter conditions in the process of change point detection.