Bootstrap-based inference for multiple variance changepoint models.

Area of Science:

• Statistics
• Data Science
• Time Series Analysis

Background:

• Variance changepoints are common and significant in diverse fields like economics, finance, biomedicine, and oceanography.
• Accurate detection of these changepoints is essential for reliable data analysis and modeling.

Purpose of the Study:

• To develop an advanced technique for constructing confidence intervals for variances in sequences with multiple changepoints.
• To enhance the accuracy and reliability of changepoint detection in noisy data.

Main Methods:

• A novel approach combining bootstrapping with the weighted sequential binary segmentation (WSBS) algorithm and the Bayesian information criterion (BIC).
• Introduction of an intensity score from bootstrap replications to identify potential changepoint locations.
• Derivation of asymptotic properties for the proposed changepoint estimation method.

Main Results:

• Simulated results demonstrate superior performance compared to current state-of-the-art segmentation methods.
• The proposed method effectively constructs confidence intervals for variances in the presence of multiple changepoints.
• Validation of the method's efficacy across various datasets including stock prices, oceanographic data, DNA copy numbers, and traffic flow.

Conclusions:

• The proposed combined bootstrapping and WSBS method offers a robust and accurate solution for detecting multiple variance changepoints.
• This technique provides improved confidence intervals for variances, enhancing analytical capabilities in diverse scientific domains.
• The method's applicability to real-world data underscores its practical significance in economics, finance, biomedicine, and environmental science.