A Generalised Dynamic Factor Model for the Belgian Economy - Useful Business Cycle Indicators and GDP Growth Forecasts

Abstract

This paper aims to extract the common variation in a data set of 509 conjunctural series as an indication of the Belgian business cycle. The data set contains information on business and consumer surveys of Belgium and its neighbouring countries, macroeconomic variables and some worldwide watched indicators such as the ISM and the OECD confidence indicators. The statistical framework used is the One-sided Generalised Dynamic Factor Model developed by Forni, Hallin, Lippi and Reichlin (2005). The model splits the series in a common component, driven by the business cycle, and an idiosyncratic component. Well-known indicators such as the EC economic sentiment indicator for Belgium 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 common component of the variables into leading, lagging and coincident with respect to the common component of 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 coincide with Belgian GDP, while they lead euro area GDP and its confidence indicators. 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 as a result of a nonnegligible 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's quarter-on-quarter growth can be dramatically reduced. However, this does not improve the forecasts.

Keywords: Dynamic factor model, business cycle, leading indicators, forecasting, data reduction

JEL Classification: C33, C43, E32, E37