A Generalized Dynamic Factor Model for the Belgian Economy

This paper aims to identify the Belgian business cycle and forecast GDP growth based on a large data base of short-term conjunctural indicators. The data base consists of 509 indicators containing information on surveys of Belgium and its neighbouring countries, macroeconomic variables and some worldwide watched indicators such as the US ISM and OECD confidence indicators. The statistical framework used is the One-Sided Generalized Dynamic Factor Model developed by Forni, Hallin, Lippi and Reichlin (2003). The model reduces the variables to their core business cycle information, defined as the part of variation of the variables common to the data set. Well-known indicators such as the EC economic sentiment indicator and the NBB overall synthetic curve contain a high amount of business cycle information. Furthermore, the richness of the model allows to determine the cyclical properties of the series and to forecast GDP growth all within the same unified setting. We classify the variables into leading, lagging and coincident with respect to a reference business cycle defined as the common variation contained in quarter-on-quarter GDP growth. 22% of the variables are found to be leading. Amongst the most leading variables we find asset prices and international confidence indicators such as the ISM and some OECD indicators. In general, national business confidence surveys are found to be coincident, while consumer confidence seems to lag. Although the model captures the dynamic common variation contained in the data set, forecasts based on that information are insufficient to deliver a good proxy for GDP growth given a non-negligible idiosyncratic part in GDP's variance. Lastly, we explore the dependence of the model's results on the data set and show through a data reduction process that the idiosyncratic part of GDP growth can be dramatically reduced. However, this does not improve the forecasts.