2007-2013: This is what the indicator told us - Evaluating the performance of real-time nowcasts from a dynamic factor model

In 2007, a new indicator of economic activity for Luxembourg was elaborated at the BcL. It was developed using a large database of about 100 economic and financial time series. The methodology was based on the generalized dynamic-factor models, and the model was estimated over the period from June 1995 to June 2007. Forecast performance was evaluated on several criteria (both in pseudo-real-time and using ex-post in-sample simulations) and results were satisfactory. They gave in particular clear evidence that the indicator provides better forecasts of GDP growth than a more standard approach that relies on past GDP values only. In this paper, we present results of the real-time use of the indicator from December 2007 onwards. Special attention is given to real-time forecasts of GDP growth and the real-time assessment of the economic situation that were made during the financial crisis. The root mean squared forecast errors of the indicator-based GDP growth forecasts have decreased during the 2009-2011 ''recovery" period in comparison to the 2007-2009 period, which is an encouraging result. This paper also includes results based on an extended study period, which includes (real-time) forecasts that were produced until the end of April 2013. The mean squared errors appear to have on average decreased over the second half of this extended study period in comparison with the first half. Finally, the BcL indicator produced better forecasts on average than the benchmarks over this extended study period.

JEL Classification:  C53, E17

Keywords:  Forecasting, factor model, large datasets, real-time analysis