62 M. E. Kreinin
version, the comparison is rather imperfect. Finally, the accuracy of projections will be compared with that of a “Naive” forecast.
Of the several possible “Naive” forecasts, I have chosen to follow the book of Ecclesiastes Chapter 1, verse 9:
“That which has been is that which shall be And that which hath been done is that which shall be done
And there is nothing new under the sun.”
In other words, the Naive forecasts assumes that last year’s development e.g., growth rate or inflation will be the forecast
for the next year. OECD forecasts are generated by “country desks,” and are
based on: models, the desk-official judgment, accounting checks, anticipated shocks, and the like. Each “forecasting round” lasts
3 months, and contains two to three iterations; it begins with exogenous assumptions about exchange rates and other variables.
The OECD is restricted to using governmental projections of each respective public sector. Estimates are made of each country’s
domestic demand and import demand, and a global model is used to check the consistency of a country’s imports with other
countries’ exports. That adds to the accuracy of the trade balances projections, as shown in Section 5 below. Consistency of interest
rates between countries and that of other variables are also checked. Reliance is made on none-OECD sources such as the
IMF and the World Bank. In particular, the OECD reads the IMF projections that appear earlier.
2. REAL GDP GROWTH RATES
For each year t the OECD Economic Outlook makes growth rate projections for every OECD member in the preceding year
t 2 1, where the “forecasting round” is done during September– November; and in the middle of the year itself t 2 12, where
the “forecasting round” is done during March–May. The actual growth rate is published in the following year t 1 1. For example,
the projections for 1990 appear in 1989 and then in mid-1990, while the actual 1990 growth rates are published in 1991.
To estimate the accuracy of the projections, the actual growth rate G
a
is regressed on the “year earlier” t 2 1 projection and separately on the midyear t 2 12 projection:
G
a
5 a 1 b P
t
2
1
1 e G
a
5 a 1 b P
t
2
12
1 e 1
A coefficient b 5 1 and a 5 0 would indicate an unbiased
ACCURACY OF OECD AND IMF PROJECTION 63
forecast. Regressions are estimated in a time series for each coun- try, cross-sectionally for each year across countries, and for all
years and countries combined. Individual country errors can offset each other in the second and third approaches.
Table 1 presents the regression results for all country–year com- binations. The midyear OECD projections are more accurate than
the year-earlier forecasts, but in neither case do the coefficient 1 and the constant zero fall within the 95 percent of their respective
confidence intervals. On the other hand, the projections are supe- rior to the results of the “Naive” model. Similarly, the IMF projec-
tions appear superior to its corresponding “Naive” model. But the OECD and IMF forecasts are not comparable because they
differ both in the number of countries and years of coverage.
More important are the individual country projections. Those made by the OECD for the eight largest countries are shown in
Table 2. For seven of the countries all except Italy the year- earlier projections are unbiased in a sense that the dual criterion
is met: the coefficient 1 and the constant 0 fall within their respec- tive 95-percent confidence intervals. The U.S. projections appear
“best” in terms of the statistical properties. In general, the midyear projections are superior to the year-earlier ones, as might be
expected.
In the following smaller countries the coefficient 5 1 and the constant 5 0 falls within their respective 95-percent confidence
intervals: Austria, Belgium, Denmark, Finland, Greece, The Neth- erlands, Portugal, Spain, Sweden, Switzerland, and Turkey. Nei-
ther condition is met in Norway. Only one of the conditions coef- ficient 5 1 is met in Iceland, Ireland, and Luxemburg.
In the cross-sectional regressions not shown both conditions are met in 15 out of 27 years, but there is no discernible improve-
ment over time. Also the midyear projections do not represent a marked improvement over the year-earlier ones.
That these projections are far superior to the forecast by the Naive model is shown in Table 3. In none of the eight “large”
countries is the dual criterions met. Nor is it met in any of the smaller countries. Using the same criteria for the cross-sectional
data covering the years 1962–94, in 18 years neither of the two criteria is met; in 8 years both criteria are met; in 5 years only the
constant 5 0 criterion is met; and in 1 year only the coefficient 5 1 criterion is met.
Table 4 shows the IMF projections of growth rates for the G-7 and other OECD countries. With only 10 observations, the
64
M. E.
Kreinin
Table 1: Real GDP Growth Rates Projections: Time Series and Cross-Section Combined
95 95
t 2 1
t Conf.
t 2 12
t Conf.
Coef. Stat.
R
2
interv. N
Coef. Stat.
R
2
interv. N
A OECD: i. Projections Coefficient
0.7 15.1
0.3 0.6–0.8
611 0.9
27.9 0.5
0.83–0.96 644
Constant 0.9
5.8 0.6–1.2
0.6 5.1
0.4–0.9 i.i Naive Model same as last year
Coefficient 0.4
13.5 0.2
0.3–0.47 893
Constant 2.0
14.7 1.7–2.3
B IMF: i Projections Coefficient
0.7 3.1
0.1 0.1–1.1
80 0.6
9.0 0.4
0.5–0.75 112
Constant 0.6
1.0 2
0.6–1.9 0.9
4.7 0.5–1.3
i.i Naive Model same as last year Coefficient
0.4 5.3
0.2 0.3–0.6
112 Constant
1.3 5.5
0.9–1.8
ACCURACY
OF OECD
AND IMF
P ROJECTION
65
Table 2: OECD Projections of Real GDP Growth Rates–Time Series
95 95
t 2 1
t Conf.
t 2 12
t Conf.
Country Coef.
Stat. R
2
interv. N
Coef. Stat.
R
2
interv. N
United States Coef.
0.9 7.0
0.7 0.7–1.2
27 0.8
10.0 0.8
0.6–0.9 27
Constant 0.0
0.0 2
0.9–0.9 0.4
1.4 2
0.2–1.0 Japan
Coef. 0.8
4.5 0.4
0.4–1.1 26
0.8 8.0
0.7 0.6–1.0
28 Constant
0.6 0.6
2 1.4–2.7
0.8 1.3
2 0.5–2.1
Germany Coef.
1.0 3.7
0.3 0.4–1.6
26 0.9
9.0 0.7
0.7–1.1 28
Constant 0.1
0.2 2
1.5–1.8 0.6
1.7 2
0.1–1.2 France
Coef. 0.8
4.6 0.4
0.4–1.2 26
0.8 8.8
0.7 0.6–1.0
28 Constant
0.5 0.8
2 0.7–1.7
0.7 2.1
0.0–1.3 United Kingdom
Coef. 0.9
3.7 0.3
0.4–1.4 26
1.1 10.1
0.8 0.9–1.3
28 Constant
0.4 0.6
2 0.8–1.6
0.4 1.3
2 0.2–0.9
Canada Coef.
1.4 5.3
0.5 0.8–1.9
26 1.0
7.5 0.7
0.7–1.3 28
Constant 2
1.0 2
1.1 2
2.9–0.8 0.2
0.5 2
0.8–1.3 Italy
Coef. 0.6
3.6 0.3
0.3–1.0 26
0.9 7.0
0.6 0.6–1.2
28 Constant
1.4 2.2
0.1–2.6 0.7
1.5 2
0.2–1.6 Australia
Coef. 0.6
2.0 0.1
2 0.0–1.2
22 0.9
5.0 0.5
0.5–1.2 23
Constant 1.0
1.0 2
1.1–3.3 0.3
0.5 2
1.0–1.6
66 M. E. Kreinin
Table 3: OECD Projections of Real GDP Growth Rate–Time Series Naive Model
95 t 2
1 t
Conf. Country
Coef. Stat.
R
2
interv. N
United States Coef.
0.3 1.6
0.0 2
0.1–0.6 34
Constant 2.1
3.2 0.8–3.4
Japan Coef.
0.5 2.9
0.2 0.1–0.8
34 Constant
3.2 2.9
1.0–5.5 Germany
Coef. 0.3
1.9 0.1
2 0.0–0.7
33 Constant
2.0 3.1
0.7–3.3 France
Coef. 0.6
4.5 0.4
0.3–0.9 33
Constant 1.2
2.2 0.1–2.4
United Kingdom Coef.
0.3 1.9
0.1 0.0–0.7
33 Constant
1.5 2.9
0.4–2.6 Canada
Coef. 0.4
2.5 0.1
0.1–0.7 33
Constant 2.3
3.1 0.8–3.9
Italy Coef.
0.3 2.2
0.1 0.0–0.7
33 Constant
2.1 3.1
0.7–3.5 Australia
Coef. 0.3
1.7 0.1
2 0.1–0.6
33 Constant
2.8 4.0
1.4–4.3
statistical properties are rather poor, but they improve consider- ably in the midyear projections. Yet the dual criterion is met
in most cases. In the cross-sectional regressions, with only eight observations, the dual criterion is met in 10 out of 15 years.
These results are superior to forecast by the Naive model shown in Table 5.
Where the OECD and IMF projections do poorly is in forecast- ing the turning points. Table 6 shows poor statistical properties,
and coefficients at variance with expectations. The only possible exception is in the IMF midyear forecast for the downturns and
upturns combined.
3. INFLATION