14
Roland Dohrn
Table 3
Test for BIAS of the Forecasts Considered1
1991-2004, S-values and their level of significance
RWI-7 |
RWI-4 |
RWI-3 |
GD-6 |
GD-4 |
GD-2 | |
GDP |
12** |
10 |
8 |
10 |
7 |
8 |
Private consumption (PC) |
12** |
12** |
10 |
12** |
10 |
7 |
Government consumption (GC) |
7 |
6 |
5 |
5 |
6 |
2** |
Investment in equipment (IEQ) |
11* |
11* |
11* |
10 |
9 |
9 |
Investment in structures (IS) |
10 |
11* |
9 |
10 |
10 |
6 |
Exports (EX) |
9 |
8 |
8 |
8 |
5 |
6 |
Imports (IM) |
9 |
8 |
9 |
8 |
7 |
9 |
Author's computations. - 1See eq. (8). For abbreviations see tables 1. Level of significance: ***
1%;** 5%;* 10%.
the indicators only in a few cases. In the RWI forecasts the type II error is
more common (21 cases) than in the GD forecasts.
Addressing some specific findings, an outstanding result is the strong corre-
lation of changes in the OECD leading indicator with errors of the forecasts
with relative long horizons, in particular the GD-6 forecast. Indeed, OECD
(2002: 33) shows in hindsight that some turning points in the business cycle
were indicated by the OECD index with a rather long lead in the 1990s. This
was particularly true for the start of the downturn in 1992 and the recovery in
1993. The forecasters obviously were not aware of these interrelations in the
past. But surprisingly the result for GDP is not mirrored in any of the GDP
components.
One important source for calculating the OECD leading indicator is the ifo
business survey. Therefore, it is not very surprising that forecast errors also
covariate with ifo business expectations. However, it is abit surprising that the
survey contains some information that might help to improve even the GD-6
forecast, because the companies were asked to assess their expectation over
the next six months only. But again, this interrelation is not mirrored in the
components of GDP.
Finally, a strong correlation is also found between errors in the GDP forecast
of GD-4 and GD-2 on the one hand and share prices on the other hand. At
least in GD-4 the forecast error of exports also shows some co-variation with
changes in share prices. Other channels through which the share market can
be expected to influence GDP growth seem to have been taken into account
correctly: Neither private consumption expenditure nor investment in
equipment show any interrelation.