Reaction to Philip Hans Franses’ Note ‘Why is GDP typically revised upwards?’

Abstract

In the period 1990-2005 Statistics Netherlands adjusted quarterly Dutch GDP growth (q on q −4 ) by 0.55% point on average. Although in most countries upward adjustments of GDP growth rates are more common than downward adjustments, this is not always the case. For instance, in Germany and Canada the mean adjustment is around zero (Öller and Hansson, 2002). For the Netherlands, the upward adjustment on average was first noticed by CPB Netherlands Bureau for Policy Analysis, 17 years ago (Van Vlimmeren, Don and Okker, 1991). From international perspective, the revisions of Dutch GDP growth rates seem to be rather large. Realized GDP growth rates in recent quarters provide useful information in making forecasts for economic growth in the current and following year. If, afterwards, these preliminary GDP growth figures need to be adjusted, the forecasters used wrong information, which unfavourably influences their forecast accuracy. Using a Monte Carlo simulation technique, Lanser and Kranendonk (2008) show that 15% (11%) of the forecasting error of the Dutch official forecasts for GDP growth in the current (following) year, made by CPB, can be attributed to revisions of preliminary data published by Statistics Netherlands (see also Kranendonk and Verbruggen, 2005, p. 20). Therefore, being forecasters ourselves, we are very much interested in knowledge that could help to diminish these adjustments.In his challenging note, Philip Hans Franses (PHF) proposes a simple reason why GDP is usually adjusted upwards (Franses, 2008). His hypothesis, which unfortunately cannot be tested, is based on the process of collecting data and the behaviour of the statisticians who have to fill in the missing data. We are not convinced that PHF's note sheds some light on this topic. In this short reaction, we will make some qualifications. First, we like to notice that PHF seems to mix up the concepts 'forecasting' and 'nowcasting'. When a statistician has to provide values of yet to collect data about a recent quarter, this is usually called nowcasting, where PHF uses the concept of forecasting. His assumption of an asymmetric loss or utility function is regularly mentioned to be one of the reasons for overestimating GDP growth forecasts in countries where the official forecasts are made by a 'non-independent' ministry. See, for instance, Jonung and Larch (2004), who suggest that to remedy the optimism bias in official forecasts, forecasting should be assigned to an