Forecasting unemployment in Brazil: A robust reconciliation approach using hierarchical data

Abstract

Hierarchical forecasting methods take advantage of the hierarchical structure of the data through base forecast reconciliation, generating results that are usually unbiased and more accurate than those provided by benchmark methods. When combining base forecasts through regression-based reconciliation strategies, however, some forecasts may behave like outliers, causing distortions to the reconciliation process. This work introduces the concept of robust estimation for hierarchical forecast reconciliation methods. We formalize two different robust-based approaches applied to unemployment data from multiple labor force surveys in Brazil. In doing so, we address a significant gap in the modelling and forecasting of unemployment, taking into account the hierarchical structure of the data. To demonstrate the potential and validity of the proposed approaches, we compare their performance with those from traditional and state-of-the-art methods. Overall, the robust approaches show promising forecasting results under multiple settings and through the lens of different evaluation metrics. Furthermore, the methodology developed herein is flexible, in the sense that it can be readily applied to other time series and deliver equally reliable results.