|
Abstract:
|
Change point analysis is a powerful and increasingly popular branch of statistics with a broad range of applications. This analysis is traditionally carried out by targeting the mean of a time series, but there are many situations where this method proves ineffective. We instead broaden our approach to test for any distributional changes within the time series. To do this, we develop and compare three newer methods that target other properties of a distribution. In this talk, I will discuss our Bayesian method that uses prior estimates to create a posterior distribution capable of producing the likelihood that each point is a change-point. Our goal here is to compare the efficiency of this method with that of others on climatic time series via an extensive simulation study and application to real data sets.
|
