Improving Business Cycle Forecasts’ Accuracy - What Can We Learn from Past Errors?



10


Roland Dohrn

Table 1

Accuracy of Forecasts of Annual GDP Growth in Germany
1991-2004

Institution

Forecast
horizon
(quarters)

Month
published

Mean
Absolute
Forecasting
Error
(MAFE)

Mean

Squared
Forecasting
Error
(MSFE)

BIAS

RWI (RWI-7)

7

July

1.26

2.87

1.03

IMF

6

September

1.22

2.93

0.96

GD (GD-6)

6

October

0.99

1.47

0.64

Sachverstandigenrat

6

November

0.84

1.17

0.47

EU

6

November

0.91

1.32

0.56

OECD

6

December

0.92

1.35

0.45

Jahreswirtschaftsbericht

5

January

0.75

0.90

0.49

RWI (RWI-4)

4

February

0.76

0.98

0.54

IMF

4

April

0.53

0.45

0.11

GD (GD-4)

4

April

0.57

0.48

0.09

EU

3

May

0.55

0.50

0.11

OECD

3

June

0.49

0.45

0.06

RWI (RWI-3)

3

July

0.40

0.30

0.16

GD (GD-2)

2

October

0.17

0.04

0.05

Author’s computations.

paper is restricted to the years after 1991, taking the German unification as a
“natural” break. From 1991 to 1994 forecasts for Western Germany only and
thereafter for the unified Germany are considered. The sample period ends
2004.

National accounts data are often revised quite substantially after the first pub-
Iication (Braakmann 2003; Oller, Hansson 2002). Therefore it is difficult to de-
termine what figures should serve as “realisations” to compare the forecast
with. In the following, the first published quarterly national accounts are taken
as a yardstick. This procedure has been employed in other forecast evaluations
too (e.g. Kirchgassner, Savioz 2001:358). It seems plausible since the first “of-
ficial” data are based mostly on the information which also forms the back-
ground of the forecast. Later revisions may be substantial, but they hardly
could have been anticipated, in particular if they emerge from new definitions
in the national accounts or from changed methods to compile data.

Table 1 compares some measures of forecast accuracy of the six GDP forecasts
under scrutiny with a sample of other projections, partly from international in-
stitutions. It shows that the Mean Absolute Forecasting Error (MAFE) as well
as the Mean Squared Forecasting Error (MSFE) and the BIAS do not differ to
much from the values calculated for other forecasts published at the same
time. Hence, the predictions analysed here seem to be state of the art.



More intriguing information

1. Measuring and Testing Advertising-Induced Rotation in the Demand Curve
2. The name is absent
3. The name is absent
4. The name is absent
5. The Values and Character Dispositions of 14-16 Year Olds in the Hodge Hill Constituency
6. The name is absent
7. Quality practices, priorities and performance: an international study
8. Conditions for learning: partnerships for engaging secondary pupils with contemporary art.
9. he Effect of Phosphorylation on the Electron Capture Dissociation of Peptide Ions
10. AGRICULTURAL PRODUCERS' WILLINGNESS TO PAY FOR REAL-TIME MESOSCALE WEATHER INFORMATION