Bootstrap-based inference for multiple variance changepoint models

Variance changepoints in economics, finance, biomedicine, oceanography, etc. are frequent and significant. To better detect these changepoints, we propose a new technique for constructing confidence intervals for the variances of a noisy sequence with multiple changepoints by combining bootstrapping with the weighted sequential binary segmentation (WSBS) algorithm and the Bayesian information criterion (BIC). The intensity score obtained from the bootstrap replications is introduced to reflect the possibility that each location is, or is close to, one of the changepoints. On this basis, a new changepoint estimation is proposed, and its asymptotic properties are derived. The simulated results show that the proposed method has superior performance in comparison with the state-of-the-art segmentation methods. Finally, the method is applied to weekly stock prices, oceanographic data, DNA copy number data and traffic flow data.