Page Title


Tout ce qui fait ou fera l’actualité du BFP est présenté dans cette rubrique : qu’il s’agisse de ses dernières études, des communiqués de presse en relation avec celles-ci, d’articles ou encore l’annonce de ses prochaines publications, workshops, colloques.

Article (21/05/2013)


La version de l'article disponible ci-dessus en format HTML ne contient généralement pas l'ensemble des informations reprises dans l'article en format pdf. Pour une version complète de l'article (avec les graphiques et tableaux), veuillez télécharger l'artcle en format pdf disponible dans le cadre "PDF & downloads" situé à droite.

Forecasting Belgian GDP in times of international crisis

The Federal Planning Bureau (FPB) is responsible, within the National Accounts Institute, for producing the economic budget, i.e. the macroeconomic forecasts used to establish the federal government budget and perform budgetary control exercises. In order to ensure transparency, ex post evaluations of the quality of these forecasts are undertaken at regular time intervals. In the latest assessment, the one-year-ahead forecast errors for economic growth in 2009 and, to a lesser extent, in 2010 appeared as outliers. In this article we analyse the
impact of world trade forecast errors on Belgian GDP forecasts.

Setting the scene

The economic budget of September 2008 was finalised on 10 September and foresaw a deceleration in Belgian GDP growth from 1.6% in 2008 to 1.2% in 2009. The collapse of Lehman Brothers just a few days later created a worldwide panic on financial markets with devastating consequences for international trade. The year 2009 turned out to be the most severe recession recorded in Europe since World War II. In Belgium, GDP shrank by 3.0%, according to the first release of the National Accounts. This exceptionally strong downturn and the elevated levels of uncertainty caused forecasting institutions to be extremely cautious for 2010. Accordingly, in the economic budget of September 2009, Belgian GDP was projected to grow by a mere 0.4% in 2010. Against all the odds, a robust growth of 2.2% was recorded. These forecast errors, calculated as the difference between forecasts and outcomes, are put in perspective in Graph 1. The forecast error for the year 2009 appears to be by far the largest made since the launch of the economic budget in 1994. For the year 2010, the magnitude of the error is, in absolute terms, more in line with those recorded during the period 2000-2003, but it nevertheless represents the greatest underestimation of growth in the sample.

The forecasting process at the FPB

While the quarterly econometric model MODTRIM constitutes the corner stone of the production process of the economic budget, it has to be supplemented with information that is produced outside the model. These so-called exogenous variables are assumed not to be affected by the model results. The most important exogenous variables in the context of this analysis are those that form the international environment, which is crucial for a small open economy. Assumptions for world trade, commodity prices and financial variables are founded on international forecasts and future market quotations. Other exogenous variables are related to the Belgian socio-demographic context, the development of wages and fiscal and social policies. Once the trajectory for all exogenous variables is settled, the MODTRIM model can be simulated. The model outcome is then confronted with external information such as business cycle indicators, near-term forecasting methods or any other type of recent information that cannot be inserted directly into the model equations. The model solution can then, if necessary, be amended using so-called “add-factors”. These adjustment variables are introduced in such a way that the consistency of the model is fully respected. This external information is of importance merely in the very short run, with their impact fading with the lengthening of the forecasting horizon. Why were the GDP forecasts for 2009-2010 so wrong? Within the bulk of exogenous variables, the development of foreign export markets is essential to determine Belgian GDP growth. The export markets are computed using a reweighted average of world trade reflecting the geographical orientation of Belgian exports. To perform these calculations, the FPB uses world trade forecasts produced by international organizations such as the IMF, the OECD or the European Commission. As can be seen in Graph 2, errors on foreign export market and GDP growth are highly correlated. The coefficient of determination of the estimated linear equation indicates that almost 90% of the variance in the errors on GDP growth is explained by errors on potential export market growth.

However, it is clear that the equation presented here is a reduced form which captures not only international trade surprises but also the impact of other international variables correlated with world trade, such as oil prices, asset prices or interest rates. It may even capture the effects of fiscal policy if the latter is pro-cyclical. Therefore, to isolate the impact of export market growth errors on GDP forecasts, a standard simulation measuring the sole impact of a shock on foreign export markets was produced with the MODTRIM model. This simulation indicates that a 1% rise in export markets raises GDP by 0.25% after four quarters. The unadjusted GDP forecast errors and those adjusted for errors in export market growth are presented in Graph 3.

After correction, neither 2009 nor 2010 appear as outliers. In other words, if the dramatic downturn in world trade had been properly anticipated in September 2008, Belgian GDP growth for 2009 would have been overestimated by only 0.5 %-points (instead of 4.3 %-points). For 2010, the GDP forecast error would have been 0.6 %-points (instead of -1.8) with an accurate evaluation of the upswing in export markets. The poor forecasting performance of world trade during this period is related to the well-known difficulty of predicting turning points. Moreover, recessions are even harder to foresee and may even be considered by their nature as impossible to anticipate.

Implications for the statistical properties of the forecast errors

It is also interesting to examine the implications of this adjustment on the overall properties of the forecast errors in all economic budgets. Therefore several forecasting rounds are distinguished: September of the year t-1 to forecast year t (round 1), February of year t forecasting the current year (round 2) and September of year t for that same year (round 3). The most intuitive indicator in measuring forecast errors is the mean absolute error (MAE), which provides the average deviation in absolute terms of the forecast from the outcome. Table 1 shows that adjusting for errors on export market growth reduces the MAE by more than 60% for round 1 and by 40% on average for the two following rounds. The decline in the MAE from one round to a subsequent one is significantly reduced with the correction, especially from round 1 to 2. This implies that an essential difference between the initial economic budget (round 1) and that prepared for the budgetary control (round 2) lies in a better assessment of the international business cycle stance. Another frequently used indicator is the root mean square error (RMSE), which penalises more large errors. The RMSE is reduced even more after correction and is much closer to the MAE, indicating that large errors on GDP growth are indeed primarily caused by errors on export market growth.

Another important parameter used to gauge the quality of the forecasts is the mean error, which indicates the average of by how much the projected growth rates were overestimated (positive sign) or underestimated (negative sign). A desirable property of forecasts is unbiasedness, meaning that positive and negative forecast errors should offset each other, on average. The optimistic bias observed for round 1 is more than halved after adjustment for export markets, while the absence of bias is confirmed for the following rounds.


This article validates the importance of the international scenario and, in particular, the evolution of world trade for forecasting Belgian GDP accurately. This said, the other international exogenous variables (oil prices, interest rates, exchange rates) and national policy variables are important as well, in particular to forecast inflation and the evolution of public finances.

  Documents associés






  Mots clés


Please do not visit, its a trap for bots