🤖 AI Summary
Detecting structural breaks in multivariate time series—particularly complex, nonlinear changes in serial dependence and marginal distributions that cannot be characterized by shifts in means or parametric forms—remains challenging.
Method: We propose a model-free, nonparametric segmentation framework. Its core innovation is the first use of the joint characteristic function of the time series and its lagged variables to capture higher-order dependence changes without distributional assumptions. The method combines sliding-window kernel estimation with adaptive thresholding for end-to-end detection of multiple change-point types.
Contribution/Results: We establish theoretical consistency for both the number and locations of estimated change points. Empirical evaluations demonstrate significant improvements over state-of-the-art methods across diverse break scenarios. The approach is successfully applied to earthquake signal identification and macroeconomic analysis, accurately pinpointing critical structural turning points.
📝 Abstract
Modern time series data often exhibit complex dependence and structural changes which are not easily characterized by shifts in the mean or model parameters. We propose a nonparametric data segmentation methodology for multivariate time series. By considering joint characteristic functions between the time series and its lagged values, our proposed method is able to detect changepoints in the marginal distribution, but also those in possibly nonlinear serial dependence, all without the need to prespecify the type of changes. We show the theoretical consistency of our method in estimating the total number and the locations of the changepoints, and demonstrate its good performance against a variety of changepoint scenarios. We further demonstrate its usefulness in applications to seismology and economic time series.
