Analysis of the forecasting performance of the threshold autoregressive model

Abstract: In this investigation, we analyze the forecasting performance of the threshold autoregressive (TAR) model. To this aim, we find the Bayesian predictive distribution from this model, and then, we conduct an out-of-sample forecasting exercise, where we compare forecasts from the TAR model with those from a linear model and nonlinear smooth transition autoregressive, self-exciting threshold autoregressive and Markov-switching autoregressive models. For this empirical forecast evaluation, we: i) use the U.S. and Colombian GDP, unemployment rate, industrial production index and inflation time series, which lead us to estimate and forecast forty models; and, ii) use evaluation criteria and statistical tests that are mostly employed in literature. We also compare the in-sample properties of the estimated models. For the overall comparison, we find a satisfactory performance of the TAR model in forecasting the chosen economic time series, and a shape changing characteristic in the Bayesian predictive distributions of this model that may capture the cycles in the economic time series. This gives important signals about the forecasting ability of the TAR model in the economic field.

• Autores: Vaca González, Paola Andrea
• Fecha de publicación: 2018
• Institución: Universidad Nacional de Colombia
• Repositorio: Universidad Nacional de Colombia
• Idioma: spa
• OAI Identifier: oai:repositorio.unal.edu.co:unal/64780
• Acceso en línea: https://repositorio.unal.edu.co/handle/unal/64780; http://bdigital.unal.edu.co/65786/
• Palabra clave: 5 Ciencias naturales y matemáticas / Science; 51 Matemáticas / Mathematics; Bayesian predictive distributions; Forecasts comparison; Threshold autoregressive model; Linear model; Nonlinear model; Distribuciones predictivas Bayesianas; Comparación de pronósticos; Modelo autorregresivo de umbrales; Modelo lineal; Modelo no lineal
• Rights: openAccess
• License: Atribución-NoComercial 4.0 Internacional