Growth slowdowns: Middle-income trap vs. regression to the mean
7 May 2015
By Lant Pritchett and Lawrence H. Summers.
[Re-posted from VOX – CEPR’s Policy Portal.]No question is more important for the living standards of billions of people, or for the evolution of the global system than the question of how rapidly differently economies will grow over the next generation. We believe that conventional wisdom makes two important errors in assessing future growth prospects.
First, it succumbs to the extrapolative temptation and supposes that, absent major new developments, countries that have been growing rapidly will continue to grow rapidly, and countries that have been stagnating will continue to stagnate. In fact, when it comes to growth, our research (Pritchett and Summers 2014) suggests that the past is much less the prologue than is commonly supposed.
Second, conventional wisdom subscribes to the notion of a ‘middle-income trap’ – the idea that when countries reach some intermediate income threshold, growth becomes much more difficult. Our work suggests that any tendency of this type is very weak, and that what is often ascribed to the middle-income trap is better thought of as growth rates reverting to their means.
Sports writers refer often to the phenomenon of the ‘sophomore slump’ – the tendency for outstanding rookies to perform less well in their second seasons. In that same vein, reference is often made to the ‘cover of Time magazine curse’ – the observation that public figures or celebrities seem to be on the cover of Time at the peak of their careers, and that it is downhill from there. Both of these phenomena reflect the statistical principle of mean reversion. In any process where there are transitory random fluctuations, there will be a tendency for increases to be followed by decreases, and above-average levels to be followed by declines.
In Pritchett and Summers (2014), we apply this frame to the analysis of national growth rates and find that regression to the mean is a robust regularity. We corroborate earlier findings (e.g. Easterly et al. 1993) that find the correlation across decades in national growth rates is surprisingly low, typically in the range of 0.2 to 0.3. This is in sharp contrast to typical forecasts, which assume much more persistence in growth rates – with both success and failure expected to continue. It is also inconsistent with many prevailing theories of growth that seek to explain growth performance in terms of highly stable national features like culture, institutional quality, or the degree of openness. We suggest that the prevailing pattern of regression to the mean in growth rates should create substantial doubt about extrapolative forecasts of China’s growth, and we believe that there is a significant risk of a major growth slowdown in China at some point over the next decade.
It is natural to ask whether there is any content to the idea of a ‘middle-income trap’, over and above regression to the mean. Since countries that have recently entered into any classification of ‘middle-income’ are, almost by definition, countries that have had recent rapid growth, it is obviously difficult to distinguish between these two phenomena. We think there are three ways in which ‘middle-income trap’ is overstated as a concern relative to regression to the mean.
Which countries should count as ‘middle-income’?
First, we find it difficult to understand the meaning of the ‘middle-income trap’ when it is used to discuss countries that range from Latin American countries to Russia to China to Indonesia to India to Vietnam to Ethiopia. We agree that all of these (and other) rapidly growing countries should be greatly concerned about the risk of a growth deceleration. But, it is not at all clear what the ‘middle-income trap’ means – outside of some arbitrary threshold the World Bank set decades ago for concessional lending – if countries at less than 11% of US GDP per capita (Indonesia 10.2, India 8.5, Vietnam 8.0 and Ethiopia 1.8) qualify and are considered at risk. The 2011 world average GDP per capita was $14,467 and the median was $8,491. Countries like Mexico (average GDP $12,709), Malaysia ($13,468), and Turkey ($14,437) are clearly ‘middle-income’ countries, and having reached these levels of economic sophistication might make continued progress more difficult. But, India could grow for 20 more years at 6% per capita before reaching Mexico’s 2011 level of $12,709, and for Ethiopia it would take 47 years. It is impossible to disentangle the ‘middle-income trap’ from regression-to-the-mean growth dynamics if all growing countries – regardless of their current level of income – are considered at risk of a ‘middle-income trap’.
Income level is a poor predictor of growth slowdowns
Second, if we run an empirical horse race of the correlates of growth decelerations (discrete transitions to lower growth), we find rapid growth a much more powerful predictor of the likelihood of a deceleration than level of income. In our recent paper, Pritchett and Summers (2014), we the use the timing of growth episodes from Kar et al. (2013) to estimate a dummy for a growth deceleration, and regress that on current growth and a quartic in the level of per capita income.
There are three points, all visibly detectable in Figure 1, that plot the predicted values for a country at the average growth rate, at India’s growth rate of 6.3% per annum, and at China’s growth rate of 8.6% per annum.
First, we take an F-test of whether the income terms are jointly significant at the 0.037 level. We find significance, but this is entirely driven by the difference between the very highest income countries and the rest of the sample – the tapering risk for high-income countries. If one estimates the same regression for just those under $25,000 (or any lower threshold), the income terms are not jointly significant.
Second, while the quartic has some ‘middle-income trap’ features – particularly, that decelerations are more likely at higher income ranges up to a point, and less likely above that point – the peak to trough difference is small. For a country at the average growth rate, the smallest likelihood of a deceleration is 4.3% at an income level of roughly $5,500, which then increases, but only to 5.7% at income of roughly $19,000. Going from the range of incomes of Mongolia or Jamaica all the way to the upper middle-income range of Hungary or Bahrain would increase the predicted deceleration in any given year by only 1.4%. Not surprisingly, given the low value of the F-test, this is much smaller than the standard error of the prediction, so within this range, the highest and lowest risks of the ‘middle-income trap’ are not significantly different.
Rapid growth is a strong predictor of future slowdowns
Third, in contrast to the very weak impact of level of income, the impact of the current growth rate on the likelihood of deceleration is large, significant and important. A 1% higher growth rate leads to a 1.46% higher risk of deceleration – equal to the trough to peak rise due to moving into the ‘middle-income’ range. The T-statistic on past growth is 13.3. As Figure 1 shows, the predicted likelihood of a slowdown in China is 14.4% at its current growth rate of 8.63%, and at the country sample growth rate of 1.84%, but the same income level it is only 4.5% – a 10 percentage point difference.
So, by these predictions of the correlates of growth slowdowns, there is a substantial risk of slowdown for countries that are growing rapidly, but it is almost entirely due to regression to the mean; whereas the effect of the ‘middle-income trap’ is small, and even moves to the peak ‘middle-income trap’ risk on these estimates empirically add little value.
Figure 1. Risk of growth slowdown from ‘middle-income trap’ is small compared to risk from regression to the mean
Source: Pritchett and Summers 2014.
These three findings – low statistical significance, small empirical magnitudes, much larger differences in predicted change in growth – also hold true for simple panel regressions of growth acceleration rates on lagged growth rates and a quartic in level of income at five-year frequencies. There is some increase in the likelihood of acceleration up to a point, and then a decline above that point – but the magnitude of pure ‘middle-income’ decreasing acceleration in growth at income levels above $11,000 is small compared to the regression-to-the-mean effect. The coefficient on lagged growth is 0.83 – consistent with strong mean reversion.
Why do we care whether the impetus for adopting policies to sustain growth is based on a narrative of regression to the mean as opposed to the ‘middle-income trap’? Both emphasise that sustaining rapid growth is hard and that the quickest way to slow growth is complacency during rapid growth. However, we think there are three important differences.
Three important differences for policy thinking
First, the cause of the sophomore slump cannot be found by asking, “What went wrong in athletes’ second years?” The cause of sophomore slumps may just be exceptionally good luck in their first years. It was not that success was qualitatively harder in their second year, or that they got worse, just that things tend to even out. As we have learned excruciatingly from experience with the financial crisis in the US, the time when things seem to be going well is the time it is most important to be prudent about the future. Labelling the problem of sustaining growth as the ‘middle-income trap’ might suggest that the risk of a growth slowdown is much more controllable by policy than it actually is.
We have had a number of people suggest that China will not have a growth slowdown because a slowdown would be politically costly and hence Chinese policymakers have every incentive to avoid that. Thus, they will act to prevent it – and a known ‘trap’ should be easily avoidable. But acting to sustain an unsustainably high growth rate may lead subsequent adjustment to be much harsher. As our friend Ricardo Hausmann puts it, “the path from 8 percent growth to 4 percent growth often goes through negative 2 percent.”
Second, labelling the generic risk of growth slowdowns a ‘middle-income trap’ risks posing current challenges in the wrong light – of how to handle the final stages of the transition from middle-income into developed economy, with the accompanying ‘bright lights’ temptations of picking cutting-edge industries to attract, for example, biotech and software engineering firms. But, the complexity for most countries in the World Bank ‘lower-middle-income’ band is that the 21st century beckons, but there are still 19th century problems to address. Sixty percent of Indians still practice open defecation – because the urban water and sanitation is so inadequate. The average food share in Vietnam in 2010 was still 50% – and so core agriculture issues must remain on the agenda.
Third, one of the factors that makes sustaining rapid growth so difficult is that growth depends on both parts of creative destruction, but only one is popular with governments and economic elites. Everyone loves the ‘creative’, as it brings new investments, new industries, and new profits. But sustained economic growth typically relies on continued structural transformation in which new industries arise, but also old industries shrink – sometimes just relatively, but sometimes absolutely. While essential to sustained economic growth, neither governments nor existing firms like destruction – with its geographic shifts, employment shifts, and firm exits that are a necessary part of weeding out uncompetitive industries. This makes it much harder to sustain rapid growth (which few do) than to get it going (which many do). While this happens at ‘middle-income’ levels, it is also a part of economic transformation at every stage – from the challenges posed by the threat of new industrialists to the landed aristocracy in the Industrial Revolution to the difficulties facing the most advanced countries today. Sustaining growth is not a ‘middle-income’ problem – it is the fundamental challenge of progress at all stages.
At the end of the day, our reading of the statistical evidence is that rapid growth is not something that can be taken for granted, even for those who have enjoyed for a long time. Its continuation requires the constant renewal of good policy, along with good luck. This is the challenging reality for all countries, not just those who have been deemed to have middle incomes.
Easterly W., M. Kremer, L. Pritchett and L. Summers (1993), “Good policy or good luck: Country growth performance and temporary shocks”, Journal of Monetary Economics 32(3), 459–483.
Kar S., L. Pritchett, S. Raihan and K. Sen (2013), “Looking for a break: Identifying transitions in growth regimes”, Journal of Macroeconomics 38(B), 151–166.
Pritchett, L. and L. H. Summers (2014), “Asiaphoria meets regression to the mean”, NBER Working Paper 20573.