Improving Business Cycle Forecasts’ Accuracy
13
All these indicators are available at least back to 1979, CSI being the only ex-
ception (since 1985). The calculate the “centered” indicators according to (7),
the median of the observations over the last 10 years is subtracted from the
figure included in the test; the median has been calculated for monthly data.
Before 1991 West German data have been used, thereafter data for unified
Germany.
4. Results of the Tests
In a first step it is tested whether the forecasts considered are unbiased
(table 3). As this is not the case in particular for the projections with longer ho-
rizons and for most of the forecasts of private consumption expenditure (PC),
the correction described in (6) has been applied to all forecasts. After this
transformation the rank signed test is calculated.
Table 4 summarizes the results of the orthogonality test. In 100 out of 798 com-
binations of indicators, economic aggregates and forecasts co-variation was
found which was significant at least at a 10%-level. Before making reference
to specific results, some more general conclusions can be presented first:
- In many cases (45 out of 100) a co-variation between forecast errors and
economic indicators is significant at a 10%-level only, giving a rather weak
indication for possibilities to improve forecasts.
- There is no indication that the two forecasters exploit the data in a different
way; in 44 cases a co-variation of indicators with the RWI forecast errors
was found, in 56 cases with the GD forecast errors. χ 2 shows that the differ-
ences are not significant.
- Comparing forecasts of different horizons, a closer look at the indicators
might help to improve the forecasts with a 4 months horizon in particular
(40 cases). A smaller number of significant correlations was found for the
forecast with longer (RWI-7, GD-6: 29 cases) as well with a shorter horizon
(RWI-3, GD-4: 31 cases).
- Most correlations (30) are observed with errors in GDP forecasts. On the
other hand, only in 6 cases errors in investment in struktures forecasts
covariate significantly with any of the indicators considered. For goverment
consuption the number of correlations is quite small (9), too. For invest-
ment in equipment, which is the forecast with the highest average error, 15
cases of significant correlations are observed.
- Ifa correlation is detected between an indicator and GDP, the same indica-
tor as a rule does not covariate with one of the demand side components of
GDP et vice versa.
- Type I errors (72 out of 100) predominate, i.e. forecasters make insufficient
use of some indicators. On the other hand, they over-estimate the impact of