Abstract
This paper proposes a novel application of a dynamic threshold regression model to three-way panel data, in which the third dimension influences both the identification of regimes and their associated thresholds. Estimation is carried out using the first-difference generalized method of moments with instrumental variables, and a criterion is proposed for the identification of change points in the temporal dynamics. The performance of the proposed methods is assessed through an extensive Monte Carlo simulation study, evaluating the accuracy of both coefficient and threshold estimates, as well as the precision of the detected change points. The methodology is then applied to an innovative three-way panel dataset consisting of fruit bioimpedance measurements collected across a range of frequencies using two types of electrical impedance analyzers: a bench-top device and a state-of-the-art portable one. The analysis provides insights into ripening dynamics by estimating thresholds and identifying change points across the frequency spectrum. In addition, it offers new evidence regarding the performance of the portable device, supporting its practical relevance for post-harvest monitoring, processing, and optimisation within the fruit supply chain. The proposed three-way panel model is implemented in the R package PanelTM.