Generalized Lambda Distribution-Based Change Point Detection Technique: A Comparative Study with Pruned Exact Linear Time (PELT) Approach

Identifying structural changes in time series data is essential for accurate modeling, forecasting, and informed decision-making in many fields. Classical change point detection methods often assume specific distributional forms, such as normality, which can limit their effectiveness when dealing with real-world data exhibiting skewness or heavy tails. The limitation is mitigated in this study through the introduction of a methodological enhancement by introducing a change point analysis (CPA) approach based on the Generalized Lambda Distribution (GLD), which is known for its flexibility in modeling diverse data distributions. The primary objective is to develop and evaluate a GLD-based CPA (GLDCPA) method and compare the performance with the widely applied Pruned Exact Linear Time (PELT) algorithm. Simulation results demonstrated that the GLD-CPA approach consistently achieved higher accuracy in detecting all true change points across all sample sizes, particularly outperforming the PELT method, where PELT frequently failed to identify multiple change points. The proposed methodology is broadly applicable to real-world time series data, particularly those exhibiting non-normal characteristics. To illustrate the utility and effectiveness of the method, currency exchange rate series were selected. The findings highlight the advantage of adopting flexible distributional modeling in CPA, offering improved detection capabilities in both simulated and empirical settings. The proposed GLD-CPA method is a reliable framework for analyzing time series data characterized by complex distributional structures, offering practical utility for both applied and theoretical investigations.