VoxEU Column Development Global economy Poverty and Income Inequality

Identifying prisoners of the middle-income trap

The World Bank has identified 37 countries as being in a ‘middle-income trap’, but few formal tests of the middle-income trap hypothesis exist. This column presents a new test based on a more nuanced observation that incorporates information on a country’s long-run growth path. Only seven out of 46 middle-income countries are found to be potentially ‘trapped’. Some countries that are usually considered to be trapped may just be growing very slowly.

The disappointing post-crises growth record of the East Asian Tiger economies has given rise to the notion of a middle-income trap (Gill and Kharas 2007). This was eloquently highlighted by a World Bank (2012) study showing that of 101 middle-income economies in 1960, only 13 had become high-income by 2008.

Nevertheless, there are few formal tests of the hypothesis. One approach is to see how the probability of a growth slowdown changes as income levels rise (Eichengreen et al. 2012, 2013).1 Naturally, however, this has limitations. For example, Pritchett and Summers (2014) note that some slowdowns may simply reflect a reversion of growth to the long-run trend. Moreover the World Bank’s study shows that not all ‘trapped’ countries experienced slowdowns. Countries such as Turkey and Colombia simply grew slowly throughout the entire sample period.

A trap as an equilibrium path

These considerations suggest that a test for the presence of a middle-income trap should incorporate information on a country’s long-run growth path. To that end, in a new paper (Ye and Robertson 2015), we use time series techniques to characterise countries’ per capita income time paths. Since the notion of a ‘trap’ implies an equilibrium, we consider whether a country’s steady-state income level lies in the middle of the world income distribution.

Specifically, for a country in a middle-income trap, we would expect that the long-term forecast of per capita income relative to a wealthy reference country is:

  1. time invariant; and,
  2. lies within the middle-income band.

Together, these imply that the gap in the log of per capita income, relative to any country growing at the world growth rate, is level-stationary.

Critically, a country would not satisfy these necessary conditions if its growth path contains a unit root, since the expected mean per capita income gap is not time invariant. Likewise a deterministic trend would suggest that the country is on a transition path, so cannot meet these necessary equilibrium conditions.

Trapped in time

Conditions (i) and (ii) are fairly general and one could operationalise them in different ways. We chose a parsimonious route of taking the US per capita income path as indicative of the long-run growth of the world productivity frontier. We then applied standard unit root tests to the log difference of GDP per capita between each country and the US, allowing for endogenous structural breaks and different types of breaks. If the log per capita income gap exhibits a unit root, or the estimated mean is not in the middle-income band, we reject the country as a middle-income trap candidate.

We find that only seven out of 46 middle-income countries can potentially satisfy our definition of a middle-income trap candidate under one of our various models – in the sense that there is at least one specification where we cannot reject a middle-income trap. These are Cuba, EI Salvador, Lebanon, Peru, Romania, Syria, Thailand, and Turkey.

This result contrasts strongly with the finding of 37 countries in a middle-income trap, if we use the World Bank’s method that simply asks if a country is classified as a middle-income economy at the start and end of the sample period. In particular, many countries that appear to be in a middle-income trap do not satisfy our definition due to the presence of a stochastic trend.

A visual comparison emphasises this distinction. Figure 1 shows Lebanon, Thailand, and Turkey in the left panel, and Brazil, Indonesia, and Columbia in the right. These trends look very similar. But, of these six countries, only the countries in the left panel satisfy our definition. For Brazil and Columbia we cannot reject a unit root in any model.

Figure 1.

Conversely Indonesia’s per capita income gap does not have a unit root. But, as indicated by the fitted line, we find that post-1997 crisis trend growth, relative to the US, has been positive. The picture is similar to Thailand, except that in Thailand’s case we find that the per capita income gap after the crisis is level-stationary.

It is also interesting to note that Lebanon and Thailand were excluded from a World Bank list of middle-income trap countries, since they were not defined as middle-income countries at the start and end of the sample period. Thailand, however, grew into the middle from below the middle-income band, and Lebanon stagnated from above the middle-income band. So our list of candidates is not just a subset of the World Bank’s larger list. Likewise, since it includes countries like Turkey that have grown slowly throughout the sample period, it differs from the candidates of Eichengreen et al.

Traps versus policy mistakes and bad luck

Thus only very few countries’ per capita income time-paths are consistent with the pattern that we would expect to see if there was an equilibrating process that constrains them to the current relative income level. On that basis, our results suggest that if middle-income traps do exist, they are not very common.

One reason for this is that our definition is more stringent than the idea of growth slowdowns and hence, appropriately, our tests require that countries are close to a steady-state path. Thus we exclude transitional countries, some of which may be growing very slowly. This is an important caveat when comparing our study with others in the literature that focuses on slowdowns.

Nevertheless, as noted above, a key reason for our low number of middle-income trap countries is that many suspect countries have a stochastic growth path. One possibility is that the paths of these countries simply reflect bad luck, poor policy choices, or some combination of these. The two cases differ because, while stationarity suggests the presence of equilibrium forces, a stochastic trend suggests there is no reason why the trend will persist – it may just represent a string of bad luck.

Thus the slow growth of most middle-income countries is an important concern. We find that there are significant differences in the per capita trends among these countries. Only a few are consistent with an equilibrating process that we would expect to see if the country was in a middle-income trap. This is far from proving whether middle-income traps exist or not, or from understanding their causes if they do exist.  Nevertheless identifying the different types of slow growth is an important step toward understanding the sources of growth in middle-income countries, and why so many countries seem to be trapped.


Bulman, D, M Eden and H Nguyen (2014) “Transitioning from low-income growth to high-income growth: Is there a middle-income trap?”, World Bank, Policy Research Working Paper, 7104.

Eichengreen, B, D Park and K Shin (2013) “Growth slowdowns redux: New evidence on the middle-income trap”, NBER, Working Paper w18673.

Eichengreen, B, D Park and K Shin (2013) “Growth slowdowns redux: New evidence on the middle-income trap”, VoxEU.org, 11 January.

Gill, I S and H J Kharas (2007) An East Asian renaissance: Ideas for economic

growth, World Bank, Washington DC.

Pritchett, L and L H Summers (2014) “Asiaphoria meets regression to the mean”, NBER, Working Paper 20573.

Pritchett, L and L H Summers (2014) “Growth slowdowns: Middle-income trap vs. regression to the mean”, VoxEU.org, 11 December.

World Bank (2012) “China 2030: Building a modern, harmonious, and creative high-income society”, Technical Report.

Ye, L and P E Robertson (2015) “On the existence of a middle-income trap”, Economic Record, forthcoming.


1 See also Bulman et al. (2014) and Ye and Robertson (2015) for additional tests and references. 

11,128 Reads