In 2003, real economic growth in Belgium amounted to 1.1% thanks to the recovery registered in the second half of the year. World trade growth, which has been remarkably strong since the last few months of 2003, should weaken and the impact of the more expensive euro should make itself more profoundly felt. The pace of exports and GDP growth should then slacken a little by the end of this year. All in all, GDP at constant prices should grow by 2.0% in 2004.
Last year, solid domestic demand combined with disappointing exports led to a considerable negative contribution of net exports to GDP growth. Thanks to the strong recovery of exports and the weaker growth of domestic demand, that negative contribution should be transformed into a slightly positive contribution this year. Households will only reduce their savings rate when the situation on the labour market becomes noticeably brighter. However, the unemployment rate should only stabilize by the end of 2004, thereby preventing a further fall in the savings rate.
This year, a gradual increase in employment should be registered. By the end of the year, employment should be 16,500 units higher than the level at the end of last year. Due to the low starting point at the beginning of this year and the fact that the increase is taking place gradually, employment in annual average should exceed last year’s level by only 7,000 units.
Headline inflation should increase by 1.5% in 2004, as compared with 1.6% last year. On the one hand, underlying inflation should drop significantly as a result of the past appreciation of the euro and the moderate evolution of unit labour costs. On the other hand, the downward impact of the abolition/ reduction of radio and television license fees has been almost exhausted.
Since February 1994, the FPB has produced short-term economic forecasts twice a year. These forecasts are referred to as 'Economic Budgets' and create the macroeconomic context for the drafting of the Federal Government’s Revenue and Expenditure Budget. In this special topic, FPB short-term projections are subjected to a quality control in which the size, nature and source of forecasting errors will be considered.
The economic budget is released twice a year, once in the summer of year t-1 for the preparation of the budget for year t and once in February of year t for the budgetary control of year t. The quarterly macroeconometric model Modtrim is used as a central tool in producing the economic budget. The model’s results, however, are adjusted on the basis of experts’ views and are adapted to take into account the latest business cycle-related information, taken for instance from leading indicators.
The economic budget supplies forecasts for a large range of economic variables. In this analysis, only two variables, namely real GDP growth and CPI inflation, will be examined. These are two global series that are important to users of the economic budget. One initial choice that must be made is to define what is considered as the outcome. In the case of inflation this is quite straightforward, since the monthly published CPI is never revised. GDP growth figures, however, are subject to significant revisions. To avoid a situation where unforeseeable factors in the national accounts (such as methodological changes) affect the analysis, outcomes are defined as the figures published in the first version of the national accounts of the year in question.
In addition to the problem of what should be chosen as outcomes, post-mortem analyses are generally beset by other problems. The unchanged policy hypothesis undeniably constitutes a drawback in terms of the accuracy of the forecasts, although the importance of this element for short-term GDP and inflation forecasts should not be exaggerated. The limited sample size should lead to a cautious interpretation of the results, which may be influenced by some outliers.
In this article, the accuracy of the FPB short-term economic forecasts will be assessed. It should be remembered that accuracy is only one aspect of quality. Other important features of quality are coherence and completeness. A full quality assessment should take all these aspects simultaneously into account.
Is the size of FPB forecast errors acceptable?
The most intuitive indicator to use in evaluating the size of forecast errors is the mean absolute error (MAE), which yields the average difference between forecasts and actual figures in percentage points (see table 1). The MAEs of the first round forecasts, which are not negligible, demonstrate the limits of a forecasting exercise. A look at the declining profile of the MAEs shows that the additional information that becomes available between the first and the second forecasting round significantly increases the accuracy of the projections, thus proving the usefulness of the February updates of the economic budget.
Projections made by international organisations are used to some extent as exogenous inputs for FPB forecasts. Since the economic budget is released a few months later than EC and OECD forecasts, forecast errors in the economic budget should be somewhat smaller, due to the integration of more recent and more detailed information. The results reported in table 1 generally confirm this expectation.
Another way to evaluate the accuracy of the FPB’s projections is to compare them with naive forecasts. Three kinds of naive forecasts have been examined: a random walk (Theil 1), a 10-year moving average growth rate forecast (Theil 2) and an ARIMA projection (Theil 3). Theil coefficients are calculated as the ratio between the root mean square error of the reference forecasts and the naive forecasts. Table 1 shows that naive forecasts are usually less accurate than the economic budget. Theil coefficients exceed unity in some cases (Theil2 and Theil3 for GDP growth during the first round and Theil3 for CPI inflation during the second round). This is, however, not systematically the case over the whole sample, but only for very recent years. Theil coefficients fall below unity if these specific years are removed from the sample period. In conclusion, our findings confirm the commonly accepted view that over a 4 to 6-quarter horizon, structural model based forecasts outperform naive forecasts.
On the basis of the above criteria, it can be said that FPB forecast errors are well within acceptable territory. Moreover, GDP forecast errors should be put in perspective by comparing them to the revisions in the outcomes. The mean absolute revision of GDP growth over the 1993-2002 period was 0.5%-point, which is half of the first round forecast error. [More in the publication ...]
STU 1-04 was finalised on 3 March 2004.