Chapter

Annex 2. The Quality of Growth and Inflation Forecasts for G20 Countries

Author(s):
International Monetary Fund. Independent Evaluation Office
Published Date:
April 2014
Share
  • ShareShare
Show Summary Details

The bulk of the quantitative analysis of the quality of IMF forecasts in the main text focuses on the entire IMF membership, treating all member countries irrespective of size. This annex concentrates on the economies in the G20.

Table A2.1 and Figure A2.1 detail the findings on forecast quality for the G20 countries. First, over-predictions of GDP growth, indicating an overall optimistic bias, are the most frequent outcome. Considering all G20 countries and forecast horizons, 75 percent of the numbers reported in Table A2.1 indicate optimistic biases, and about 45 percent of the biases are statistically significant at the 10 percent level. Figure A2.1 shows the predominance of over-predictions in GDP growth one-year-ahead forecasts, represented by negative mean forecast errors (dotted black line) in 15 out of the 20 countries.

Table A2.1.GDP Growth Forecast Errors, Spring WEO, 1990-2011
Current-YearOne-Year-AheadTwo-Years-AheadThree-Years-AheadFour-Years-AheadFive-Years-Ahead
United States0.07−0.05−0.31−0.21−0.03−0.05
(0.511)(0.845)(0.562)(0.7266)(0.9647)(0.9326)
United Kingdom−0.17−0.82−0.95−0.86−0.78−0.83
(0.11)(0.007)(0.0276)(0.0903)(0.1276)(0.117)
France−0.13−0.90−1.25−1.29−1.11−1.13
(0.0059)(0.00)(0.00)(0.000131)(0.000394)(0.001766)
Germany0.17−0.79−1.19−1.28−1.06−1.08
(0.323)(0.0072)(0.00)(0.0011)(0.0072)(0.0087)
Italy−0.40−1.26−1.55−1.62−1.57−1.62
(0.00)(0.00)(0.00)(0.0002)(0.0015)(0.002)
Canada−0.18−0.85−0.79−0.62−0.49−0.51
(0.146)(0.0132)(0.0073)(0.1451)(0.2679)(0.2612)
Japan−0.03−1.20−2.02−2.27−2.17−2.05
(0.814)(0.0027)(0.0007)(0.001)(0.001)(0.0019)
Euro area−0.16−1.02−1.44−1.47−1.34−1.48
(0.479)(0.0511)(0.0155)(0.021)(0.0698)(0.0971)
Turkey0.46−0.42−0.69−0.65−0.72−0.22
(0.207)(0.32)(0.0632)(0.504)(0.5257)(0.849)
Australia0.10−0.33−0.24−0.21−0.16−0.26
(0.561)(0.111)(0.336)(0.5317)(0.6834)(0.5207)
South Africa−0.60−0.98−1.06−0.85−0.77−0.75
(0.101)(0.0074)(0.002)(0.093)(0.0951)(0.0895)
Argentina1.010.330.350.340.230.04
(0.0988)(0.784)(0.785)(0.8459)(0.9005)(0.9837)
Brazil0.17−0.52−1.14−1.10−1.27−1.50
(0.433)(0.131)(0.0229)(0.0659)(0.051)(0.0354)
Mexico−0.28−1.52−2.28−2.52−2.48−2.60
(0.0877)(0.0005)(0.0000239)(0.0078)(0.0078)(0.0062)
Saudi Arabia1.150.870.07−0.19−0.180.26
(0.0352)(0.165)(0.84)(0.7796)(0.8127)(0.7421)
India0.260.290.460.550.740.92
(0.324)(0.472)(0.203)(0.196)(0.1044)(0.0693)
Indonesia0.10−0.83−1.36−1.78−2.06−2.22
(0.688)(0.178)(0.0737)(0.1968)(0.146)(0.1194)
Korea0.50−0.56−1.25−1.36−1.45−1.77
(0.0159)(0.0177)(0.0000323)(0.0243)(0.024)(0.0069)
Russia0.55−1.15−1.60−1.61−1.26−1.11
(0.191)(0.331)(0.181)(0.3224)(0.3739)(0.38)
China1.161.771.871.771.671.58
(0.0005)(0.009)(0.0299)(0.0741)(0.0615)(0.0529)
Note: Robust standard-error p-values in parentheses; numbers in bold blue (bold black italics) represent statistically significant—at the 10 percent level—overpre-dictions (underpredictions).Source: IEO calculations using IMF, World Economic Outlook.
Note: Robust standard-error p-values in parentheses; numbers in bold blue (bold black italics) represent statistically significant—at the 10 percent level—overpre-dictions (underpredictions).Source: IEO calculations using IMF, World Economic Outlook.

Figure A2.1.Errors in One-Year-Ahead GDP Growth Forecasts for G20 Economies

(The dotted blue line indicates the average error for the entire sample period)
(The dotted blue line indicates the average error for the entire sample period)

Source: IEO calculations using the IMF’s Spring World Economic Outlook forecasts.

Second, underestimations of inflation are much less frequent among G20 economies than in the membership as a whole. Inflation tends to be mostly overpredicted, especially for emerging market economies but also for Germany and the euro area (Table A2.2).

Table A2.2.Inflation Forecast Errors, Spring WEO, 1990-2011
Current-YearOne-Year-AheadTwo-Years-Ahead
United States−0.140.040.37
(0.227)(0.845)(0.0755)
United Kingdom−0.090.050.11
(0.421)(0.86)(0.569)
France−0.10−0.0010.12
(0.182)(0.994)(0.557)
Germany-0.16-0.32-0.25
(0.0021)(0.0044)(0.0951)
Italy−0.13−0.30−0.22
(0.191)(0.092)(0.374)
Canada−0.110.110.28
(0.329)(0.541)(0.239)
Japan−0.010.441.01
(0.948)(0.00)(0.00)
Euro area-0.29-0.46−0.30
(0.0012)(0.0339)(0.278)
Turkey−3.86-9.63-9.28
(0.105)(0.0567)(0.153)
Australia0.260.260.23
(0.419)(0.587)(0.614)
South Africa0.72−0.910.02
(0.275)(0.135)(0.975)
Argentina−103.50−4.970.33
(0.28)(0.392)(0.871)
Brazil−192.50−278.00−291.20
(0.192)(0.189)(0.21)
Mexico-1.62-4.36-5.38
(0.0645)(0.0483)(0.0465)
Saudi Arabia0.340.230.09
(0.316)(0.735)(0.919)
India-1.19-1.90-1.84
(0.0055)(0.0431)(0.0555)
Indonesia-1.07-3.83-5.65
(0.011)(0.0037)(0.0016)
Korea0.07-0.43-0.55
(0.703)(0.0926)(0.0169)
Russia−6.65-27.16-23.79
(0.244)(0.127)(0.0871)
China0.23−0.18−0.44
(0.557)(0.806)(0.695)
Note: robust standard-error p-values in parentheses; numbers in bold blue (bold black italics) represent statistically significant—at the 10 percent level—over-predictions (underpredictions).Source: IEO calculations using IMF, World Economic Outlook.
Note: robust standard-error p-values in parentheses; numbers in bold blue (bold black italics) represent statistically significant—at the 10 percent level—over-predictions (underpredictions).Source: IEO calculations using IMF, World Economic Outlook.

Statistically significant optimistic biases in GDP growth forecasts are found for most G7 economies, notably in Europe and Japan, while underpredictions of GDP growth only occur in a few cases associated with emerging market economies1—Argentina, Korea,2 Saudi Arabia (in all cases only current-year forecasts), India (for five-year-ahead forecasts only), and notably China (Table A2.1). One-year-ahead forecast errors in France, Germany, Italy, the euro area, and Japan are below the zero line most of the time, while for China the opposite is true (Figure A2.1).

The higher frequency of underpredictions for emerging market economies may explain why, as a group, these economies do not show the average optimistic bias observed for both low-income and developed economies, as discussed in the main text. Among G20 economies, these underpredictions cancel out the statistically significant over-predictions in Brazil, Korea, Mexico, and South Africa. Over-Predictions in Indonesia and Russia are highly influenced by outliers and tend not to be statistically significant.

One aspect of economic forecasts in general, and of GDP growth forecasts in particular, is the underlying uncertainty stemming from the volatility of shocks that hit a particular economy. Presumably, economies frequently hit by more volatile shocks should be more difficult to predict. Indeed, for G20 economies, forecast inaccuracy, measured by the root-mean-square forecast errors, increases for economies with more volatile output (Figure A.2.2).

Figure A2.2.Volatility of GDP Growth Versus Forecast Error in GDP Growth, One-Year-Ahead, Spring WEO 1990-2011

Source: IEO calculations using IMF, World Economic Outlook.

Finally, as in the case of the aggregate results for the entire IMF membership, IMF forecasts of GDP growth are about as accurate as forecasts from Consensus Economics. In almost all G20 economies, not only are the mean and root mean square forecast errors in IMF and Consensus forecasts very similar, but forecasts have the same overall pattern and display the same turning points in both cases (Figure A.2.3). While the striking similarity in forecast errors in IMF and private sector forecasts does not support the notion of an organizational bias in IMF forecasts, it also does not hide the occasional occurrence of large forecast errors for most G20 economies.

Figure A2.3.Errors in One-Year-Ahead GDP Growth Forecasts for G20 Economies IMF Versus Consensus Economics Forecasts

Source: IEO calculations using the IMF’s Spring World Economic Outlook and April Consensus Economics forecasts.

    Other Resources Citing This Publication