Analyse économétrique et compréhension des erreurs de prévision

Abstract

This article presents recent results regarding the econometric approach to economic forecasting. We aim to establish here what constitutes a "good" forecast. A forecast taxonomy helps understand how to obtain forecasts that prove robust to the most detrimental source of error: structural breaks that affect the data generating process. The concepts of accuracy, precision and certainty applied to forecast models show that evaluation criteria are paramount in the model design stage. An empirical application to forecasting French imports of goods and services provides an illustration.