泊松序列中变点的贝叶斯分析

Bayesian Analysis of Change-points in Poisson Sequences

利用二分法将含有多个变点的泊松序列进行分割,将泊松序列多变点问题转化为没有变点和仅有一个变点的问题,运用贝叶斯因子进行模型比较.选用上证指数数据,结合贝叶斯因子对存在变点的模型进行了检验.

For the problem of the change-points in the Poisson sequence,a Bayeian method is proposed,which can confirm the number of change-points and confirm the position of theirs.First of all,we make use of the binary segmentation procedure to have more than one change-point in Poisson sequence,so that each short section doesn＇thave change-point or only contain a change-point,so changepoints problem is transformed into a Poisson sequence hasn＇t changed and only a change-point problem.Then,compared models by using the Bayes factor model.Last,used the Bayes factor model in Shanghai securities composite index from 3 rd of December in 2012 to 30 th of December in 2012,verified model＇s precision.