Modeling and Forecasting of Industrial Production in Brazil

Abstract

Industrial Production is considered a relevant measure for analyzing the economic situation and making decisions within a country. In this study, we propose several alternative short-term forecasting models for the Industrial Production series in Brazil. In our analysis, we consider the univariate ARIMA model, the dynamic distributed lag ARDL model, and the multivariate VAR model. Additionally, we incorporate various methods of forecast combination in the final selection to enhance results. Within the study, we integrate aggregate variables such as the Selic interest rate, Bovespa index, energy consumption, revenue, working hours, imports of machinery and equipment, employment, and inflation rate. The findings indicate that the ARDL model exhibited the best forecasting performance for horizons of 1, 3, 6, and 12 steps ahead. However, in comparison to forecast combination methods, the OLS-AVG model demonstrated superior outcomes, underscoring that diversifying forecasts leads to a reduction in diversifiable error.