Analysis · Forecasting

How Wrong Is the IMF? We Graded 30,740 WEO Growth Forecasts

We scored every real GDP growth forecast in 34 IMF World Economic Outlook vintages (2008–2025) against what actually happened. The typical year-ahead forecast misses by 3.1 percentage points, the errors lean optimistic and get worse with horizon — and of more than a thousand recessions, the IMF saw only 9% coming a year ahead.

WorldPulse ResearchAugust 9, 2026

3.1pp

Typical year-ahead miss

Mean absolute error, all countries, 2008-2024

+1.2pp

Year-ahead optimism bias

Mean forecast minus outcome; 57% overshoot

Recessions called a year ahead

1,026 country-year contractions graded

35%

Forecasts within ±1pp

Year-ahead forecasts landing near the outcome

Grading of 30,740 country-year forecasts from WEO vintages April 2008–April 2025, computed August 2026. Source: IMF World Economic Outlook.

Twice a year, the IMF’s World Economic Outlook sets the reference growth numbers that governments budget against, investors trade on, and journalists cite as consensus. The projections are precise to a tenth of a percentage point. Almost nobody goes back and checks them.

We did. WorldPulse holds the IMF’s current projections in its knowledge graph, and the IMF’s own archives preserve every past edition of the WEO database. Comparing the two turns the world’s most-cited forecast into something rarer: a forecaster with a public, gradable track record. The verdict, in one line: useful for ordinary years, systematically optimistic beyond a year, and almost blind to turning points.

1. How we graded 17 years of forecasts

A forecast error here is forecast minus outcome, in percentage points of real GDP growth — positive means the IMF was too optimistic. The horizon is the gap between the vintage year and the target year: the April 2019 WEO’s projection for 2020 is a one-year-ahead forecast (h=1); its projection for 2024 is five years ahead (h=5).

We graded 34 WEO vintages (April 2008 through April 2025) for 194 economies at horizons of zero to five years — 30,740 individual forecasts, including 5,797 at the one-year horizon — against realized growth as reported in the April 2026 WEO database. Targets run from 2008 to 2024, a sample that contains the global financial crisis, the euro crisis, a pandemic and a war in Europe.

2. A year out, the typical forecast misses by three points

Accuracy decays fast with horizon — and then, surprisingly, stops decaying. By the October WEO of the target year itself the average absolute miss is 1.7 points and the bias is mildly pessimistic(−0.3pp): once real data accumulates, the IMF is a careful nowcaster. One year out, the mean absolute error roughly doubles to 3.1 points, the bias swings to +1.2 points of optimism, and only 35% of forecasts land within a point of the truth. Beyond one year the error curve goes flat at about 3.3 points: a five-year-ahead WEO projection is no more accurate than a one-year-ahead one.

The median absolute error at one year is 1.7 points — half the mean — because the distribution has violent tails: crises, commodity busts and wars produce double-digit misses that ordinary years never cancel out. Even excluding the 2009 and 2020–22 shock years, the year-ahead bias remains +0.5pp and the typical miss 2.3 points.

Forecast horizon	Bias (pp)	Mean abs. error (pp)	% too optimistic	% within ±1pp
Same year, October WEO	−0.33	1.69	41%	50%
Same year, April WEO	+0.04	2.21	47%	43%
Year ahead, October WEO	+1.08	3.01	56%	36%
Year ahead, April WEO	+1.23	3.16	58%	34%
2 years ahead (April)	+1.18	3.14	61%	32%
3 years ahead (April)	+1.28	3.28	61%	32%
4 years ahead (April)	+1.32	3.39	63%	30%
5 years ahead (April)	+1.27	3.32	63%	31%

All 194 economies, WEO vintages 2008–2025, targets 2008–2024. Bias = mean (forecast − outcome); positive = optimistic. Source: IMF World Economic Outlook archives.

3. The optimism is an emerging-market problem

Split the sample and the famous “IMF optimism bias” almost vanishes for rich countries. At the one-year horizon, advanced-economy forecasts overshoot by +0.6pp on average (absolute error 2.5pp); emerging and developing economies overshoot by +1.3pp (absolute error 3.3pp). At five years the gap is starker still: the advanced-economy bias is 0.0pp, while the emerging-market bias is +1.6pp, with 64% of forecasts too high.

That pattern fits what IMF self-evaluations have long conceded: medium-term projections often embed program assumptions and reform scenarios rather than unconditional predictions. A five-year WEO number for a developing economy is closer to a target than a forecast — worth remembering when the WEO’s 2026–2030 tables are quoted as the future.

4. The IMF has almost never called a recession in advance

Our sample contains 1,026 country-years in which growth turned out negative. A year in advance, the WEO projected negative growth in just 9.1% of them. The average such episode was a contraction of −4.9%; the average year-ahead forecast for those same episodes was +3.0% growth. Strip out the two global shocks (2009 and 2020) and the hit rate is still only 13%. Only within the target year itself does detection become the norm — 64% of contractions were flagged by the WEO rounds published that year.

The two great global misses bookend the sample. The October 2008 WEO — finalized after Lehman Brothers collapsed — still projected +3.0% world growth for 2009; the outcome was −0.4%, the only global contraction in the modern PPP-weighted series before 2020. The October 2019 WEO projected +3.4% for 2020; COVID-19 delivered −2.7%, a 6.1-point miss. The pattern then inverts: having missed the crash, the IMF missed the rebound, underestimating 2010 world growth by 3.3 points (April 2009 vintage) and 2021 growth by 1.5 points (October 2020 vintage).

One consolation: errors diversify. Averaged across all 31 graded vintage-target pairs, the world-aggregate year-ahead forecast missed by just 1.2 points — individual country errors partly cancel. The IMF is a much better forecaster of the world than of any country in it.

5. The country report card

Per-country grades at the one-year horizon (up to 31 forecasts per country, 2008–2025 vintages) separate three stories. The wildest optimism is concentrated in conflict and collapse states — Libya tops the table with a +19.3pp average overshoot. The persistent pessimism outliers are growth surprises the IMF kept doubting: Ireland (−2.9pp, inflated by multinational accounting), Ethiopia (−1.9pp) and Turkey (−1.8pp), whose credit-fueled booms repeatedly beat forecasts the IMF thought unsustainable.

And the accuracy champion is telling: the IMF’s single most forecastable major economy is China (0.95pp average miss, +0.2pp bias). Given how unusually smooth China’s official GDP series is — a question we examined in Does China’s 5% Hold Up? — forecasting a managed number is the easy version of the game.

Most optimistic

Country	Bias	MAE
Libya	+19.3	22.7
South Sudan	+14.2	16.0
Yemen	+9.2	9.6
Ukraine	+5.4	6.1
Venezuela	+4.7	8.4

Most pessimistic

Country	Bias	MAE
Ireland	−2.9	5.3
Malta	−2.6	4.6
Tajikistan	−2.2	2.6
Ethiopia	−1.9	2.3
Turkey	−1.8	3.2

Most accurate

Country	Bias	MAE
Tanzania	+0.2	0.65
Bangladesh	+0.4	0.86
Australia	+0.5	0.90
China	+0.2	0.95
Cameroon	+0.7	0.95

Major economies: year-ahead forecast grades, 2008–2024

Economy	Bias (pp)	Mean abs. error (pp)
China	+0.22	0.95
Egypt	+0.45	1.09
South Korea	+0.37	1.23
Indonesia	+0.93	1.23
United States	+0.25	1.25
France	+0.65	1.39
South Africa	+1.19	1.58
Japan	+0.90	1.65
Poland	−0.24	1.72
Italy	+0.86	1.83
United Kingdom	+0.72	1.87
Germany	+0.78	1.95
India	+0.76	2.33
Mexico	+1.39	2.36
Brazil	+0.75	2.40
Nigeria	+1.08	2.50
Saudi Arabia	−0.24	2.84
Turkey	−1.79	3.23
Russia	+0.75	3.32
Argentina	+1.53	4.61

Sorted by accuracy. Bias = mean (forecast − outcome) at the one-year horizon; positive = IMF too optimistic. 31 graded forecasts per economy. Source: IMF World Economic Outlook.

6. What the IMF expects now — read with error bars

The charts below are live from the WorldPulse knowledge graph and show the current WEO database: solid for reported history, dashed for the IMF’s projections through 2031 (April 2026 release as of publication). The track record above is the instruction manual for reading them. A US projection deserves a ±1.3pp mental error bar; a German one ±2.0pp; and any number more than a year out has shown a persistent optimistic lean — except when the next recession arrives, in which case none of these lines will have seen it.

United States: real GDP growth, actual vs IMF forecast

IMF forecast for 2026: 2.3%. Typical year-ahead miss (2008–2024): ±1.3pp (bias +0.25pp).

China: real GDP growth, actual vs IMF forecast

IMF forecast for 2026: 4.4%. Typical year-ahead miss (2008–2024): ±0.9pp (bias +0.22pp).

Germany: real GDP growth, actual vs IMF forecast

IMF forecast for 2026: 0.8%. Typical year-ahead miss (2008–2024): ±1.9pp (bias +0.78pp).

India: real GDP growth, actual vs IMF forecast

IMF forecast for 2026: 6.2%. Typical year-ahead miss (2008–2024): ±2.3pp (bias +0.76pp).

The verdict: trust the level, not the turn

As a nowcaster, the IMF is good: by the autumn of the year in question its forecasts are accurate to under two points and, if anything, slightly cautious. As a one-year forecaster it is mediocre in a way that is at least predictable — knock roughly a point off any year-ahead growth projection for an emerging economy, and expect a typical miss of about three points either way.

What the WEO cannot do is see corners. Recessions arrive in its tables only after they have arrived in the data — a 9% detection rate a year out — and recoveries are then underestimated on the way back up. The practical rule for readers of the April 2026 projections: the levels are a reasonable baseline for normal times; the turning points will not be there. If the forecast says 3.1% and the year is ordinary, expect something near 3. If the year is not ordinary, the forecast was never going to tell you.

Frequently Asked Questions

How accurate are IMF GDP growth forecasts?

Across 194 economies and WEO vintages from 2008 to 2025, the mean absolute error of a year-ahead IMF growth forecast is 3.1 percentage points (median 1.7). Only about 35% of year-ahead forecasts land within one point of the outcome. For large advanced economies the typical miss is smaller — around 1.2 points for the United States and 2.5 points for advanced economies as a group.

Is the IMF systematically optimistic?

At horizons of a year or more, yes. Year-ahead forecasts overshoot realized growth by 1.2 percentage points on average, and 57% of them are too high. The bias is concentrated in emerging and developing economies (+1.3 points at one year, +1.6 at five years); for advanced economies the five-year bias is roughly zero. By the October WEO of the year itself, the bias flips slightly pessimistic (-0.3 points).

Did the IMF predict the 2009 or 2020 recessions?

No. The October 2008 WEO — published after Lehman Brothers failed — still projected +3.0% world growth for 2009; the outcome was -0.4%. The October 2019 WEO projected +3.4% for 2020; COVID-19 delivered -2.7%. Across all 1,026 country-year recessions in our sample, only 9% were forecast as negative growth a year in advance.

Which countries does the IMF forecast best and worst?

Best (lowest year-ahead absolute error, 2008-2024): Tanzania (0.65pp), Bangladesh (0.86pp), Australia (0.90pp) and China (0.95pp) — though China’s accuracy partly reflects how smooth its official GDP series is. Worst: conflict-hit economies like Libya (22.7pp), South Sudan, Yemen and Venezuela. The largest pessimistic bias is Ireland (-2.9pp), whose multinational-driven GDP repeatedly surprised to the upside.

Where does the data in this analysis come from?

Forecasts come from 34 archived IMF World Economic Outlook databases (April 2008 through April 2025), retrieved via DBnomics. Realized growth is taken from the most recent WEO database stored in the WorldPulse knowledge graph (the April 2026 release). Errors are forecast minus outcome, so positive numbers mean the IMF was too optimistic.

Data & methodology. Forecasts are drawn from 34 archived editions of the IMF World Economic Outlook database (April 2008 through April 2025; the October 2011 edition is absent from the archive), retrieved via DBnomics. Realized growth is the “actual” series of the April 2026 WEO database as stored in the WorldPulse knowledge graph; for the United Kingdom and Greece, realized values come from the April 2025 WEO edition. Errors are forecast minus outcome in percentage points of real GDP growth; horizons are calendar-year gaps between vintage and target (an April year-ahead forecast is issued ~21 months before the target year closes). Realized growth is itself revised over time, so grades depend mildly on the actuals vintage used — the same convention the IMF’s Independent Evaluation Office applies. World aggregates use the WEO’s PPP-weighted world growth. Advanced/ emerging split follows the current WEO classification. Live charts show IMF WEO data served by WorldPulse and update as new releases are ingested; grading statistics are fixed as of August 9, 2026. Explore growth data for all countries on the GDP growth topic page. This article is analysis, not investment advice.

← All articles Browse topics →