Improving Business Cycle Forecasts’ Accuracy
RWI Essen publishes a GDP forecast twice a year, in February and July3, pro-
viding predictions for the current and for the next year (RWI : Konjunktur-
berichte). The February report contains a forecast for the next year only since
1998. Therefore, the forecast published in February for the following year
cannot be analysed here due to an insufficient number of observations. The
July forecast has a horizon of7 quarters. All in all, three forecasts per year can
be analysed subsequent. They are named according to their forecast horizon in
terms of quarterly GDP figures to be projected:
- The July forecast for the following year with a horizon of 7 quarters
(RWI-7),
- the February prediction of the current year with a horizon of 4 quarters
(RWI-4), and finally
- the July forecast of the current year with an horizon of 3 quarters (RWI-3).
The forecast of the GD is a result of the joint effort of six (up to 1993 five)
leading German economic research institutes (Arbeitsgemeinschaft, var.
years), one of them being RWI Essen. It is commissioned by the Federal
Ministry of Economics and published twice a year, in April and October. In
April it comprises a forecast horizon of8 (before 1997 of4) quarters, the first
quarter of the current year being still a forecast, in October the forecast period
is 6 quarters. Again, the April forecast for the next year does not offer a suf-
ficient number of data. Therefore, three forecasts from the GD can be
analysed in this study:
- The October forecasts for the following year with a horizon of 6 quarters
(GD-6),
- Those issued in April of the current year with a forecast horizon of 4 quar-
ters (GD-4).
- Finally, the October forecasts for the current year with a horizon of2 quar-
ters (GD-2).
From a statistical point of view, it is desirable to analyse forecast accuracy over
a long period to have a sufficient number of observations. However, the meth-
odology as well as the data background of the forecasts underwent consid-
erable changes over the last decades. National accounts data as well as many
business cycle indicators are published more up to date today than they have
been in the past. Furthermore, forecast methods as well as the personal and
professional background of the forecasters might have changed. An analysis
over a very long period could be spoilt by these factors too much. Hence, this
3
For the RWI forecast the publication date varies to some extent, but in each year it includes the
data for the same number of quarters from the national accounts. The flash forecast for the year
ahead published in December will be neglected subsequently, as it did not differ substantially
from the February forecast.