| Episode | Dur. | Growth %/yr |
Income gain |
Leading sector |
|---|
Every sustained growth acceleration in the postwar record — found by the Hausmann – Rodrik – Pritchett filter, ranked by size and duration, and decomposed by sector where the data allow.
137 episodes | 97 economies | Penn World Table 11.0, 1950–2023 | How the filter works
| Episode | Dur. | Growth %/yr |
Income gain |
Leading sector |
|---|
An episode enters the atlas if it passes the growth-acceleration filter of Hausmann, Pritchett and Rodrik (2005), applied here to real GDP per person (rgdpna/pop) in Penn World Table 11.0. Growth in year t must satisfy three conditions over an eight-year horizon: least-squares growth from t to t+8 of at least 3.5 per cent a year; an increase of at least 2 percentage points over growth in the preceding eight years; and income at t+8 above its pre-episode peak, so that pure recovery from collapse does not count as a miracle. Where several adjacent years qualify, the start year is the one that maximizes the F-statistic of a spline regression with a slope break at t — the year the trend most decisively turns. Following the original paper, countries with fewer than one million people are excluded, and accelerations less than five years apart are merged.
Hausmann, Pritchett and Rodrik dated the starts of accelerations; this atlas also dates their ends, so that episodes can be ranked by duration and size. An episode is extended one year at a time beyond the eight-year minimum for as long as growth stays miraculous: average growth since the start remains at least 3.5 per cent, trailing eight-year growth stays above 3 per cent, and income avoids collapse (it may not fall more than 5 per cent below its within-episode peak). Size is the ratio of income per person at the end of the episode to income at the start; duration is simply end year minus start year. Episodes still under way in 2023, when the data end, are marked as such — their true length is unknown.
Where the GGDC 10-Sector Database (1950–2012) or the GGDC/UNU-WIDER Economic Transformation Database (1990–2019) covers an episode, labor-productivity growth is decomposed in the manner of McMillan and Rodrik (2011): a within-sector term (each sector's own productivity growth, weighted by its initial employment share) and a reallocation or structural-change term (employment shifting toward sectors with higher productivity levels). The “leading sector” is the one contributing most to aggregate productivity growth, counting both terms. The decomposition uses constant-price value added and employment, over the longest stretch of the episode the sectoral data cover; 52 of the 137 episodes can be decomposed. The usual caveats apply — mining and real estate carry high measured productivity per worker, so their reallocation terms can be large even when employment changes are small.
| Leading sector | Episodes | Economies |
|---|
Manufacturing leads 21 of the 52 decomposable episodes — more than twice the next sector — the classic industrialization path, and the signature of nearly every East Asian miracle. Finance and business services lead nine (Mauritius, Peru, Rwanda, Nigeria's second boom), typically where a high-productivity sector gains employment share rather than raising its own productivity. Mining leads four, where the episode is a resource windfall rather than a transformation. Agriculture leads just three — Ethiopia, Nepal, and Mauritius in the 1970s.
The filter is strict in ways worth knowing. India never qualifies on these data: its acceleration was real but too gradual — the change in eight-year growth peaks at 1.8 points around 1982 and 1.6 points around 2002, always short of the required 2. Slow-burn transformations are the filter's blind spot. Bangladesh was the same story on the previous data vintage; the revised national accounts in PWT 11.0 lift its 2005 acceleration over the bar, and it now sits in the atlas.
| Near miss | Best year | Growth after | Acceleration | Required |
|---|---|---|---|---|
| India | 1982 | 3.2% | +1.8 pp | +2.0 pp |
| India | 2002 | 5.4% | +1.6 pp | +2.0 pp |
Bangladesh's promotion is a caution as much as a celebration: episode lists are sensitive to the data vintage. Moving this atlas from PWT 10.01 to 11.0 created 23 episodes, removed 17 (Egypt's infitah among them), and merged China's two reform-era episodes into one — a sensitivity documented since Johnson, Larson, Papageorgiou and Subramanian (2013) asked “Is newer better?”. Two censoring problems also shape the list. The data begin in 1950, so the first detectable start year is 1958 — the West German Wirtschaftswunder and Japan's earliest postwar surge lie just out of view (Japan's run is caught from 1958). And accelerations after 2015 cannot yet show the required eight years of results. Finally, where national statistics are doubtful — Myanmar, Turkmenistan, Somalia, Equatorial Guinea under Macías — the atlas reports the numbers and says so plainly in the text.
Hausmann, R., L. Pritchett and D. Rodrik (2005), “Growth Accelerations,” Journal of Economic Growth 10(4). McMillan, M. and D. Rodrik (2011), “Globalization, Structural Change and Productivity Growth,” NBER WP 17143. Feenstra, R., R. Inklaar and M. Timmer (2015), “The Next Generation of the Penn World Table,” AER 105(10) — PWT 11.0. Timmer, M., G. de Vries and K. de Vries (2015), GGDC 10-Sector Database. Kruse, H., E. Mensah, K. Sen and G. de Vries (2023), GGDC/UNU-WIDER Economic Transformation Database. Episode narratives were written for this atlas; they aim for orientation, not citation.