Key Details
The September 2026 research paper, hosted in the Reserve Bank of Australia’s discussion-paper series, assesses both forecast accuracy and whether successive projections make efficient use of available information.
Finding | What the Evidence Shows |
|---|---|
Forecasts examined | Real GDP growth projections from the IMF, World Bank, European Commission, OECD and the mean of private forecasts collected by Consensus Economics |
Time coverage | Varies by forecaster, extending as far back as 1985 and up to 2025 |
Accuracy comparisons | Statistically significant differences from IMF forecasts appear for only a small number of individual economies |
Cross-economy pattern | Consensus forecasts recorded lower errors than IMF forecasts in roughly 75% of economies at the current- and one-year-ahead horizons |
Longer horizons | Accuracy differences grow mainly because the average forecast becomes less reliable, with increased disagreement playing a smaller role |
Information use | Rationality is rejected for most forecasters in most economies under the paper’s statistical tests |
Forecast revisions | International organisations more often make revisions that are too large; Consensus revisions more often appear too small |
India | No statistically significant accuracy difference was found between IMF forecasts and the available OECD, World Bank or Consensus forecasts at the one- or two-year horizons |
Formal Tests Find Few Clear Winners
The paper compares IMF World Economic Outlook forecast errors with those from the World Bank, European Commission, OECD and Consensus Economics.
Few comparisons show statistically significant differences in accuracy. This does not mean forecasters are equally accurate; some samples, particularly for the World Bank and European Commission, may be too short to detect modest differences.
Across economies, however, a pattern emerges: Consensus forecasts have lower errors than IMF forecasts in about three-quarters of economies at current- and one-year horizons, while IMF forecasts outperform OECD forecasts in roughly three-quarters at the one-year horizon. These are descriptive patterns rather than formal cross-country tests because errors are correlated across economies.
Longer Horizons Produce Larger Common Errors
The paper separates accuracy differences into disagreement among forecasters and the error in their average forecast.
As horizons lengthen, errors grow mainly because the common forecast moves further from the eventual outcome. Disagreement also increases, but contributes less.
The result suggests that longer-range forecasting problems arise less from choosing the wrong institution than from multiple forecasters missing the same economic turning points.
Some Forecast Errors Are Predictable
Rationality tests examine whether information available when a forecast was made could have predicted its eventual error. The paper finds that:
forecasts are too extreme for a sizeable minority of economies;
international organisations often make larger revisions than subsequent outcomes justify;
Consensus forecasts tend to revise too cautiously; and
significant average bias is uncommon, but when present is usually optimistic.
These are statistical tests of information use. A failed rationality test does not establish that a forecast was unreasonable or poorly prepared.
Forecast Design Can Explain Some Apparent Irrationality
Forecasts often depend on assumptions about oil prices, exchange rates, interest rates, fiscal policy and geopolitical developments. When these assumptions change, subsequent errors can appear predictable even if the original forecast was internally consistent.
Institutions may also publish the most likely outcome (mode) rather than the mean of all possible outcomes. With substantial downside risks, the mode can exceed the mean and appear systematically optimistic.
Across 145 economies, the paper estimates that the mode exceeded the mean in 91 cases, by an average 0.16 percentage point. The authors caution that these distributions are difficult to estimate from short samples.
India Is Included but Not Separately Assessed
India’s forecasts are adjusted to its April–March financial year to make institutional projections comparable.
The paper finds no statistically significant accuracy difference between IMF forecasts for India and available OECD, World Bank or Consensus forecasts at the one- or two-year horizon. This does not mean individual forecasts were equally accurate in every year.
The study does not assess India’s domestic forecasting institutions, national-accounts methodology or official growth projections. Its India finding is limited to comparisons among external forecasters.
What Is Forecast Rationality?
A forecast is statistically rational when it uses information available at the time consistently with its forecasting objective. If past errors or revisions systematically predict future errors, some available information may not have been fully incorporated. The result depends, however, on what the forecaster is trying to predict, its conditioning assumptions and whether it publishes a mean, mode or another point in the forecast distribution.
Policy Relevance
For Indian fiscal, monetary and business planning, the findings favour comparison over reliance on a single headline projection. Few formally significant accuracy gaps make simple league tables less useful than examining why forecasts differ.
Three elements deserve closer attention:
Assumptions: External projections should be read alongside their expectations for oil prices, global demand, exchange rates and domestic policy.
Revision history: A large upgrade or downgrade is itself informative, but the study shows that international organisations may revise too aggressively.
Risk distribution: Publishing only one central number can conceal whether downside risks make the mean outlook weaker than the most likely outcome.
For the Ministry of Finance and RBI, forecast evaluation could therefore track not only the eventual error, but also which assumptions changed and whether successive revisions consistently overshot or undershot the outcome. Clearer documentation of forecast vintages would also help researchers separate economic surprises from weaknesses in the forecasting process.
Follow the Full Paper Here: Real GDP Forecasts by International Organisations