Abstract

This paper presents a statistical methodology for detecting outliers and structural breaks in a time series. The approach is implemented through three probabilistic and regularization-based models: the Inverse d-Gamma model, the t-Student-Beta2 model, and Bayesian Lasso. An empirical application is conducted using GDP data. First, GDP is specified within a dynamic linear model framework, where GDP is decomposed into the sum of potential output and the output gap. Potential output is modeled as a linear growth process, while the output gap is represented by a second-order autoregressive model (AR(2)). Based on this structural specification, the methodology is then used to perform inference on outliers and structural breaks in the GDP series. Finally, the analysis is extended by disaggregating GDP into its three main sectors—agriculture, industrial activities, and services—which yields a multivariate modeling setting. Empirical results are reported for all model specifications, and the paper concludes with the main findings and substantive implications.