WPS6594 Policy Research Working Paper 6594 Middle-Income Traps A Conceptual and Empirical Survey Fernando Gabriel Im David Rosenblatt The World Bank Operations and Strategy Unit Development Economics September 2013 Policy Research Working Paper 6594 Abstract In recent years, the term “middle-income trap� has categories, the paper examines historical transition entered common parlance in the development policy phases in the inter-country distribution of income based community. The term itself often has not been precisely on previous work in the literature. Transition matrix defined in the incipient literature. This paper discusses analysis provides little support for the idea of a middle- in more detail definitional issues on the so-called income trap. Analysis of cross-country patterns of growth middle-income trap. The paper presents evidence in provides additional support for the conclusions in the terms of both absolute and relative thresholds. To get a paper, which closes with a general discussion of potential better understanding of whether the performance of the policy implications. middle-income trap has been different from other income This paper is a product of the Operations and Strategy Unit, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted at fim@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Middle-Income Traps: A Conceptual and Empirical Survey Fernando Gabriel Im and David Rosenblatt1 Abstract In recent years, the term “middle-income trap� has entered common parlance in the development policy community. The term itself often has not been precisely defined in the incipient literature. This paper discusses in more detail definitional issues on the so-called middle-income trap. The paper presents evidence in terms of both absolute and relative thresholds. To get a better understanding of whether the performance of the middle-income trap has been different from other income categories, the paper examines historical transition phases in the inter-country distribution of income based on previous work in the literature. Transition matrix analysis provides little support for the idea of a middle-income trap. Analysis of cross-country patterns of growth provides additional support for the conclusions in the paper, which closes with a general discussion of potential policy implications. Key words: Economic growth, Convergence, Middle Income Countries JEL Codes: D30, E00, F02, F60, N10, O11, O40 Economic Policy Sector Board 1 World Bank, 1818 H Street, Washington DC 20433. fim@worldbank.org and drosenblatt@worldbank.org. 1 I. Introduction In recent years, the term middle income or “MIC� trap has entered common parlance in the development policy community – particularly, in East Asia where concerns about slower growth following the 1997 regional financial crisis prompted concerns of a protracted period of subpar performance. The term itself often has not been precisely defined in the incipient literature. In some cases, the phenomenon is described in terms of relative “catch-up� with the United States or some other rich country reference (Woo, 2011; Lin and Rosenblatt, 2012). In others, it is based on stagnation or painfully slow growth in absolute income levels. For example, Felipe et al (2012) establish a definition based on the number of years a country takes to move from one income category to another, based on absolute thresholds for low, lower middle, upper middle and high income countries. The concept of a low-equilibrium growth “trap� has been explored in theory and empirically, but with a focus on low-income country “poverty traps� (Azariadis and Drazen, 1990; Azariadis and Stachurski, 2005; Kraay and Raddatz, 2007). A variety of models have been developed that focus on aid volatility (Agenor and Aizenman, 2010), natural resource exploitation (Antoci et al, 2011), or institutions (Capra et al, 2009; Gradstein, 2008). Recently, Agenor and Canuto (2012) developed a model with a knowledge externality in the “design� sector that produces multiple equilibria, including the possibility of a low productivity growth MIC trap. In terms of development strategy or the microeconomic determinants of growth, a number of authors have focused on the peculiar position of MICs within the global supply chains. The basic idea is that incomes (and wages) in MICs have increased enough to require graduation from low-skilled labor intensive activities, but MICs have not yet developed national innovation systems -- or perhaps not even accumulated enough physical and human capital – to compete with high-income countries in more sophisticated products (Gill and Kharas, 2007; Jankowska et al, 2012; Yusuf and Nabeshima, 2009; and Woo, 2009). This line of inquiry leads directly to policy discussions of what needs to be done (Kharas and Kohli, 2011; Shijin et al, 2012; Xiaohe, 2012; Flaeen et al, 2013) to transition successfully from middle income status to high income status. Aiyar et al (2013) define the middle income trap as a special case of growth slowdowns and explore some of the determinants behind these slowdowns. What characterizes a middle income country (MIC)? Quite literally, MICs are defined as the middle of the distribution of countries ranked by per capita income. One could also imagine using some other metric, or metrics, of economic development to define the “middle class.� Scanning a variety of dimensions, a “typical� MIC would have income per capita of about $4,000, an adult literacy rate of about 70 to 93 percent, infant mortality of about 19 to 50 per 1,000 live births, and life expectancy is around 65-72 years.2 In terms of economic structure, the typical MIC possesses a wide variety of productive sectors–ranging from still large primary sectors to industries that may be highly developed—or even at the technological frontier globally. A typical MIC still has substantial shares of its population toiling at relatively low 2 These data are from World Development Indicators (WDI). Median income per capita (“ Atlas� measure of market exchange rates) was approximately $4,300 in 2010, as reported in the WDI. 2 productivity occupations, with limited access to capital, and with earnings that are relatively low on a global scale. On the other hand, one could argue that there is also a great dispersion across many of these socioeconomic characteristics among MICs. In brief, MICs are countries that made substantial progress in social and economic outcomes, but still lag significantly behind the rich countries in most social and economic indicators. Most of the social indicators mentioned above are positively correlated with countries’ income per capita. The simplest way to establish thresholds for MICs would be on that basis. The World Bank classifies a country as a MIC if its income per capita (Gross National Income, in accounting terms) is greater than $1,005 and less than $12,275.3 This group of countries represented about 55 percent of all the economies for which the WDI has data for the year 2010. Broadly speaking, it does represent the middle class and the distribution of countries across these income categories looks evenly distributed (Figure 1)—at least for the categories above low- income status. In this figure, the MIC class is disaggregated into “lower middle-income� (LMIC—up to $3,975) and “upper middle-income� (UMIC). The average incomes vary exponentially; however, the log levels follow a fairly straight line across income categories (line in Figure 1).4 Figure 1: Distribution of countries by income per capita, 2010 35 Log of Group's 12 30 Average GNI per 10 capita (RHS) 25 % of Countries 8 20 6 15 4 10 5 2 0 0 LICs LMICs UMICs HICs Source: World Development Indicators. Does this alleged inability of countries to progress from MIC to high-income status (HIC) imply that there is a “trap�? There have only been about 200 years of modern economic history: what has been the evolution of MICs and what would be a reasonable timeframe for progress to high-income status? This paper surveys the empirical evidence for various definitions of a MIC trap. The closest paper in the literature that we have found is Felipe et al (2012); however, we take a different approach to look at the empirics of whether the MIC to HIC transition is different 3 This is based in “Atlas� dollars, where the “Atlas� measurement smoothes out short -term changes in countries’ exchange rates. In years, where the currency has been fairly stable, the “Atlas dollar� GNI per capita is very close to a GNI per capita based on market exchange rates. More information is available at http://data.worldbank.org/about/country-classifications. 4 Felipe et al, 2012, use a methodology to extrapolate these thresholds to Geary-Khamis PPP dollar thresholds, so as to be able to track the evolution of countries across categories using the longer term Maddison data set. 3 from other transitions in economic progress. First, we will discuss in more detail the definitional issues already alluded to in this introduction. Then we examine historical transition phases in the inter-country distribution of income as well as cross-country patterns of growth to get a better understanding of whether MIC performance is different from other income categories. Finally, we close with a general discussion of potential policy implications. II. How Do We Define a Middle-Income Trap? A number of recent papers (Cai, 2012; Felipe et al, 2012; and Kharas and Kholi, 2011; Lin and Treichel, 2012; and Woo, 2009) as well as seminar presentations (for example, Woo, 2011) discuss the potential existence of a middle-income trap—particularly in the context of a specific country or region. Two definitional issues arise in this incipient literature: how one defines the thresholds for middle-income status; and secondly, how one defines a “trap.� In both cases, there are obvious parallels to the literature on poverty measurement. Poverty can be measured in absolute or relative terms—a fixed household income per capita threshold (in PPP dollars or national currency) is established for the former, while a fixed proportion of mean or median household income per capita is used for the latter. Secondly, there is an extensive literature on poverty “traps�, where a trap implies some form of self-reinforcing mechanism driven by market failures or lack of institutional development that inhibits progress towards either an absolute or relative threshold. Whether one takes a relative or absolute approach to thresholds has very strong implications for descriptive statistics. For example, from a relative perspective, the Latin America and the Caribbean (LAC) region is often cited as the classic case of a MIC trap phenomenon: a review of either the regional aggregate or individual countries shows that income per capita relative to the United States did not progress during the 20th century. Figure 2(a) uses Maddison data to track the progress of the LAC aggregate’s ratio as well as Colombia – the latter simply as an example. On the other hand, if one uses an absolute definition, then one might reach the conclusion that today’s high-income countries themselves were stuck in a “MIC trap� for much of the twentieth century. Figure 2(b) uses 2008 income per capita (in constant PPP adjusted dollars) for Colombia (still somewhat below the world average) and Poland (a recent “graduate� from MIC status, by World Bank standards) and compares the G7 economies to Colombia and Poland over the last century. One sees that it was only in the late 1950s that the G7 economies surpassed the recent income per capita of Colombia, and it was only in the 1970s that they all surpassed Poland. This figure raises the question then: Were Colombia and most of Latin America stuck in a MIC “trap� for the entire twentieth century? Or was Europe stuck in a MIC “trap� for most of the twentieth century? Or are both statements relevant? Again, this reminds one of the literature on poverty measurement, where there is a clear welfare economics criterion for defining poverty based on an absolute threshold (Ravallion, 2008); however, there are other social cohesion views 4 for taking a relativist approach.5 It may well be that the average Colombian or Pole has a higher economic standard of living than the average Western European or American in 1930; however, they still aspire be as well off as the average Western European or American today. Figure 2 (a) and (b): Two Views on a “Trap�—Relative and Absolute MIC Thresholds (a) LAC and Colombia: Ratio of GDP per capita relative (b) G7 and Modern MIC comparators: Absolute levels to the United States of income per capita (constant, PPP) 16000 0.4 US 14000 0.35 Canada 12000 0.3 Poland (2008) United 10000 Kingdom 0.25 France 8000 0.2 Colombia (2008) 6000 Germany 0.15 LAC Aggregate 4000 Italy 0.1 Colombia 2000 Japan 0.05 0 0 1900 1904 1908 1912 1916 1920 1924 1928 1932 1936 1940 1944 1948 1952 1956 1960 1964 1968 1900 1905 1910 1915 1920 1925 1930 1935 1940 1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 Source: Authors’ elaboration based on data from Maddison (2010). The relative or absolute approach also has an impact on the type of analysis that one might undertake to understand the dynamics of a presumed MIC trap. The relative approach immediately takes one down the path of two relevant threads of the economics literature: (a) absolute convergence (or the lack thereof); and (b) inter-country distribution of income. On the former, there is a vast theoretical literature on neo-classical growth models,6 and a vast empirical literature on testing convergence or lack thereof (Pritchett, 1997). On the second thread, the literature on the world distribution of income has noted that, in recent decades, widening global gaps between rich and poor individuals have been driven more by increases in the difference between countries’ incomes per capita than by differences in income across households within countries (Bourguignon and Morrison, 2002; Milanovic, 2005). The absolute definition might lead one to focus more on why some countries enter actual stagnation rather than failed “catching up.� Here issues of macroeconomic stability may have played a leading role—as noted, in particular, in the literature on the Latin American experience with periodic debt crises of the last quarter century (Calvo, 1998; Calvo and Reinhart, 2002; Reinhart et al, 2003). The stagnation in per capita GDP levels also relates to the literature on 5 For a new “weakly relative� measure that makes an explicit accounting for social cohesio n, see Ravallion and Chen (2011). For the implications for global poverty measurement, see Chen and Ravallion (2012). 6 Parente and Prescott (2005), and Banerjee and Duflo (2005) present surveys from different perspectives. 5 low-growth “poverty traps� that generally focuses on low income countries, 7 as mentioned above. III. Relative “Trap�: Some Unpleasant MIC History The preceding section (and Figure 1a) already alluded to the obvious failures of countries or entire regions to ever successfully catch up with the rich countries’ income per capita—even when adjusted by purchasing power parity. There have also been some successes, but these have been painfully rare, as documented extensively elsewhere (Felipe et al, 2012; World Bank and Development Research Center (PRC), 2012; Gill and Kharas, 2007; Lin and Rosenblatt, 2012). Only a limited number of countries have made substantial progress in relative terms, and those countries are concentrated in a handful of East Asian economies, the European periphery, and special cases of substantial (per capita) natural resource discoveries. 8 Systematic growth slowdowns among MICs are another way to view the inability to converge (Eichengreen et al (2011) and Aiyar et al (2013)). By definition, if a country grows faster (in per capita terms) than the rich countries, it will eventually catch up with the high-income countries’ GDP per capita. A relevant benchmark is the long-run per capita growth of high-income countries over the last century, which is on the order of 1.5-2.0 percent per annum. Over the last 50 years, it was 2 percent – using WDI data. If one looks at the average per capita growth rates for 141 developing countries9 over the last 50 years (1961-2011), 80 grew by at least 1.5 percent – of which, 64 grew by at least 2 percent (WDI data). While these countries are, in fact, converging, this is painfully slow convergence. Thirty-one countries grew by 3 percent or more (in per capita terms), while only 9 reached 5 percent. More generally, the time it will take a developing country to reach the GDP per capita of a high income reference is given by the following equation: ( ) ( ) where R is the initial ratio of the high GDP per capita reference to the MIC’s GDP per capita, is the MIC’s compound rate of growth of GDP per capita, and is the high income country’s compound rate of growth of GDP per capita. A MIC on the border between LMIC and UMIC has a GDP per capita that is on the order of 1/10 the United States’ level or about 1/8 of the high income OECD average (in PPP dollars). The relevant “R� ratio is on the order of 8 to 10 then. One can read from Table 1 below that if this MIC grows by only 3 to 4 percent in per capita terms, it will take a century or two to catch up with the rich countries – assuming that the rich countries grow at 1.8 percent in per capita 7 There is also a parallel microeconomic literature on poverty traps at the household level. See, for example, Antman and McKenzie, 2007. 8 The classic example of the latter is Equatorial Guinea. 9 This is using current year classification of developing economies. This excludes the “graduates� from MIC status then. 6 terms. The MIC will eventually win the race, but it will be a very long race. One should note that the official World Bank threshold for “high-income country� is substantially lower than US or OECD averages. If one were to compare a prototypical MIC to that threshold, then the relevant R ratio would be much lower: around 3. And the ratio would be 12 or less, even for the poorest LMICs. On the other hand, recent MIC graduates tend to grow faster than high income countries with a longer tenure at that status, so the relevant might be higher for constructing a table for catch-up to borderline HICs. Table 1: Number of Years for Convergence to Rich Countries GDP per capita, as a function of MIC Growth rate and Initial per capita GDP Ratio (HIC threshold/country’s level) ( = 1.8%) R 2% 3% 4% 5% 6% 7% 8% 9% 10% 25 1640 275 151 104 80 65 54 47 42 20 1526 256 140 97 74 60 51 44 39 15 1380 231 127 87 67 54 46 40 35 10 1173 196 108 74 57 46 39 34 30 5 820 137 75 52 40 32 27 24 21 4 706 118 65 45 34 28 23 20 18 3 560 94 51 35 27 22 19 16 14 2 353 59 32 22 17 14 12 10 9 1.5 207 35 19 13 10 8 7 6 5 Source: Authors’ calculations based on Equation (1). Notes: Assuming 1.8 percent growth in real GDP per capita in the rich country reference. R is the ratio of rich country GDP per capita to MIC GDP per capita. The is the rate of growth of GDP per capita in the MIC. Dividing line is to show minimum growth rates to converge to future rich country reference level in under 50 years. One can look at country examples to get a better feeling for what would be involved in catching up over the next 50 years. 10 The table below presents calculations of the required growth rate for catch-up for a collection of current middle-income countries, along with the actual growth rate in the last column.11 Both the United States and an aggregate for high-income OECD countries are used as a reference point. Note that the GDP per capita of these high income references are assumed to grow at 1.8 percent, resulting in GDP per capita of about $103,000 and $82,000 for the US and OECD respectively 50 years from now (measured in constant 2005 PPP dollars). Comparing the middle columns with the last column, we see that most countries have been substantially off-track in recent decades in terms of catching up to either a US standard or a high-income OECD standard. 10 The Commission on Growth and Development (2008) did a similar exercise (page 113 in the statistical annex), but with different assumptions on growth trajectories. 11 Actual is based on 1980 and 2011 levels in constant PPP adjusted dollars. The PPP adjusted series only goes back to 1980 in the World Development Indicators database. 7 Table 2: Required (per capita) Growth Rate for “Catch-up� in 50 Years 2011 GDP Per Required Required Actual Capita (PPP Growth Growth Rate Growth Rate adjusted 2005 Rate To To OECD (1980 to dollars) US (HIC) 2011) Argentina 15,501 3.9 3.4 1.4 Brazil 10,278 4.7 4.2 1.0 Chile 15,272 3.9 3.4 3.3 China 7,404 5.4 4.9 8.9 India 3,203 7.2 6.7 4.3 Indonesia 4,094 6.7 6.2 3.7 Malaysia 13,672 4.1 3.7 3.4 Mexico 12,776 4.3 3.8 0.7 Nigeria 2,221 8.0 7.5 1.0 Russian Federation 14,808 4.0 3.5 N.A. South Africa 9,678 4.9 4.4 0.3 Thailand 7,633 5.4 4.9 4.1 Turkey 13,466 4.2 3.7 2.7 Memo Items: US 42,486 OECD 33,726 Source: Authors’ calculations based on data from the World Development Indicators . A growth rate of 1.8 percent (per capita) is assumed for the high-income country reference points. IV. Absolute Thresholds: Some (Relatively) Pleasant MIC Arithmetic Obviously, if one establishes an absolute threshold – as is done by the World Bank—then any positive growth rate of income per capita will eventually result in successful passage to HIC status. Two problems have arisen. First, some developing countries and regions have actually experienced declines in income per capita for periods spanning a few decades (e.g., Latin America during the debt crisis or Sub-Saharan Africa during the first decades following independence). Secondly, per capita growth of only 1 or 2 percent would condemn some lower middle income countries to “developing� status for centuries. That said, for many upper MICs, sustained growth of 2 to 4 percent per capita would result in high-income status in a generation or two. Following on the calculations in the previous section, the simple arithmetic is even simpler, if one takes the HIC standard at a fixed level of GDP per capita (in constant dollars). The is then zero, and the time to achieve HIC status is much faster, as shown in Table 3. In addition, if one uses the World Bank definition of HIC, then the initial R is in a much tighter range with a maximum of about 12 (for poorest LMICs), and a maximum of 3 for all the upper middle-income countries, as noted earlier. 8 Table 3: Years towards reaching absolute high-income status, as a function of initial GDP per capita and MIC growth rate (per capita) R 1% 2% 3% 4% 5% 6% 7% 8% 9% 25 323 163 109 82 66 55 48 42 37 20 301 151 101 76 61 51 44 39 35 15 272 137 92 69 56 46 40 35 31 10 231 116 78 59 47 40 34 30 27 5 162 81 54 41 33 28 24 21 19 4 139 70 47 35 28 24 20 18 16 3 110 55 37 28 23 19 16 14 13 2 70 35 23 18 14 12 10 9 8 1.5 41 20 14 10 8 7 6 5 5 Source: Authors’ calculations. Bold italics are added to note that Upper MICs have a ratio R that is about 3 or less – if using World Bank definitions for these thresholds. Note: The dividing line shows minimum growth rates to reach current HIC threshold in under 50 years. Another issue is the precise definition of the HIC threshold. There is no precise “science� to all this; however, the most thorough review of country classification that we know of is Nielsen (2011). Nielsen (2011) compares IMF, OECD and World Bank classifications and then proposes an alternative taxonomy that would first determine the number of categories based on the distribution of development outcomes and then set thresholds at the average of distribution.12 For example, in the 3 tier system, the break between developing countries and rich countries would be the average outcome, and the sub-category between lower and medium developing countries would be at the average of that group. The World Bank’s classification has operational implications in that the LIC-MIC threshold determines access to concessional IDA financing. The World Bank also uses a “graduation� threshold for initiating a dialogue to “graduate� the country from access to all World Bank financing, even the non-concessional IBRD. This threshold is roughly half the HIC threshold. Heckelman et al (2011) find that this threshold has worked reasonably well in terms of its objective: countries have graduated when they have reached the level of institutional development and creditworthiness to have adequate access to private financial markets for the country’s development needs. Given that there is not a consensus on threshold definition, we will focus on general rules of thumb. One more ambitious threshold – relative to the World Bank definition—would be today’s GDP per capita of the OECD average or the United States or the richest country in the 12 It should be noted that some non-income dimensions have improved dramatically. In addition, many are inherently different from the level of income per capita, given upper or lower bounds. For example, net secondary enrollment and literacy rates are bounded by 100 percent, and child mortality rates by zero below. In addition, there are access-to-service issues that exist today that were non-existent when some of today’s HICs were MICs (for example, internet access). There is some discussion and data on these issues provided in Appendix 1. 9 world. Another ambitious threshold would be the OECD or US level of (real) income per capita in 1980 – assuming that even by today’s standards, these countries were already “rich� in 1980. In terms of table 2, these higher thresholds imply moving up to higher rows in the table. In terms of specific country examples, the following figures display that a number of MICs could still reach the HIC status in 2035 – roughly one generation from now—if they can sustain per capita economic growth rates on the order of 2.5 to 4.5 percent. The horizontal lines used for comparison are the 1980 GDP per capita of the United States and (high-income) OECD aggregate (constant 2005 PPP dollars). The values (from World Development Indicators) are 25,510 and 19,386 for the US and OECD respectively. The countries were chosen for illustrative purposes only; however, they do represent a wide range of MICs. Note that in the Figure 3c, with 4.5 percent growth, a number of the more advanced MICs would also breach the 2011 levels of US and OECD GDP per capita ($42,436 and $32,726, respectively, in constant 2005 PPP dollars). Figures 3(a)-(c): HICs in a generation with sustained moderate growth? (a) With per capita Growth of 2.5 percent 30000 25000 USA-1980 20000 OECD -1980 15000 10000 2011 5000 2035 (proj) 0 (b) With per capita Growth of 3.5 percent 40000 35000 30000 25000 USA-1980 20000 OECD-1980 15000 10000 2011 5000 2035 (proj) 0 10 (c) with per capita growth of 4.5 percent 50000 45000 40000 35000 30000 USA-1980 25000 20000 OECD-1980 15000 10000 2011 5000 2035 (proj) 0 Source: Authors’ calculations; 2011 data are from World Development Indicators. All measured in constant 2005 PPP dollars. V. Are MICs Really Different? Some Old Approaches Applied to a “New� Concept The above section provided some definitions and places some parameter values on MICs’ growth rates to achieve success in some reasonable (but admittedly loosely defined) timeframe.13 The MIC trap concept inevitably invokes a literature from the 1990s and early 2000s on the patterns of economic growth (for example, Pritchett 1997 and 2000) as well as the patterns of transition across the inter-country distribution of income (Quah, 1993, 1996 and 1997, and Kremer et al., 2001). One approach, inspired by the literature on the inter-country distribution of income, is to look at transitions across income categories (in either relative or absolute terms). Are MIC-to- HIC transitions more infrequent than other transitions?14 Or is the inter-country distribution of income per capita similarly stable for all income categories? Another simple empirical approach to exploring the relevance of the “trap� term is to just look at MICs’ growth patterns. In simple terms, do MICs seem to perform differently from other income classes of countries? a. Transition Matrices and the MIC Trap: Are MICs Less Upwardly Mobile? We follow the transition matrix approach previously laid out in Quah (1993) and Kremer et al. (2001) to describe the dynamics of income per capita distribution across countries. We classify countries into groups by relative income per capita taking the United States as a benchmark. 15 Rather than estimating one-year transition probabilities as in Quah or five-year 13 Felipe et al (2012) define a trap threshold based on the median number of years it takes countries to cross lower MIC and upper MIC absolute thresholds. 14 The authors would like to thank Aart Kraay for suggesting this approach. 15 Our results remain the same even if we take the top 5 richest countries rather than just the United States. This is apparent if we simply look at the growth differential between the initial high income countries and the United States. 11 transition probabilities as in Kremer et al., we focus on ten-year intervals.16 The longer time span seems more appropriate to analyze the so-called MIC trap since transitions (or lack of them) from middle income to high income status do not occur overnight. Moreover, our focus is mainly on the transitions from middle-income to other status and less on the long-run distribution of world income. Nevertheless, we do compute the ergodic distribution and the half-life implied by the respective matrices of transition probabilities. We exclude from our analysis oil-exporting as well as resource rich countries since these countries are outliers as resource discoveries are a very special form of “catch-up.�17 We include many East European and Central Asian countries in the analysis, although for many of them GDP per capita series were only available since 1990 in our sample.18 We use PPP income per capita from Maddison (2010) to compute the transition matrices for 125 countries from 1950 to 2008. Suppose that a country's relative income per capita follows a first-order Markov chain with time-invariant transition probabilities denoted by , with i = 1,..., N and j = 1,..., N. Each describes the likelihood that state i will be followed by state j. We denote the set of states as . The probability that equals some particular value j depends only on the most recent value and not on other past realizations of : ( | ) ( | ) In other words, a country's relative income today depends solely on yesterday's relative income and not on any other past realizations. We can represent the transition probabilities in matrix form as: ( ) 16 This also reduces the effect of business cycle fluctuations on countries that are near the relative income thresholds. We recalculate the transition matrices detrending the GDP per capita series using the Hodrick-Prescott filter. Our main results change very little. 17 The following countries are included in the analysis: Afghanistan, Albania, Argentina, Armenia, Australia, Austria, Bangladesh, Belarus, Belgium, Benin, Bolivia, Bosnia, Brazil, Bulgaria, Burkina Faso, Burma, Burundi, Cambodia, Canada, Cape Verde, Central African Republic, Chile, China, Comoro Islands, Costa Rica, Côte d'Ivoire, Croatia, Cuba, Czech Republic, Denmark, Djibouti, Dominican Republic, Egypt, El Salvador, Eritrea and Ethiopia, Estonia, Finland, France, Gambia, Georgia, Germany, Ghana, Greece, Guatemala, Guinea, Guinea Bissau, Haïti, Honduras, Hong Kong SAR, China, Hungary, India, Ireland, Israel, Italy, Jamaica, Japan, Jordan, Kenya, Kyrgyzstan, Laos, Latvia, Lebanon, Lesotho, Liberia, Lithuania, Macedonia, Madagascar, Malawi, Malaysia, Mali, Mauritania, Mauritius, Moldova, Mongolia, Morocco, Mozambique, Namibia, Nepal, Netherlands, New Zealand, Nicaragua, Niger, North Korea, Pakistan, Panama, Paraguay, Peru, Philippines, Poland, Portugal, Puerto Rico, Romania, Rwanda, Sao Tomé & Principe, Senegal, Seychelles, Sierra Leone, Singapore, Slovakia, Slovenia, Somalia, South Africa, South Korea, Spain, Sri Lanka, Swaziland, Sweden, Switzerland, Taiwan, China, Tajikistan, Tanzania, Thailand, Togo, Tunisia, Turkey, Uganda, United Kingdom, Ukraine, Uruguay, United States, Uzbekistan, West Bank and Gaza, Zaire (Congo-Kinshasa), Zambia, and Zimbabwe. 18 Our main messages do not change even if we exclude these countries (see Appendix 3). 12 P is an N x N matrix of transition probabilities or just transition matrix. Note that ∑ . In matrix notation, we have that P x 1 = 1, where 1 is an N x 1 vector of ones. Diagonal elements indicate the probability of countries of remaining in the same relative income per capita group after a ten-year period. Off-diagonal elements indicate the probability of a given country to jump into another relative income per capita category after one period or step. If P is ergodic, we can compute the ergodic or stationary distribution of the Markov chain, which can be interpreted as the unconditional probability of each of . We may also be interested in the speed of convergence towards the steady state. For that, we utilize the concept of half-life of convergence, which refers to the number of periods required to cut the difference between the current distribution and the stationary distribution in half: | | . is the second largest eigenvalue of the transition matrix.19 Two additional concepts may be useful for the analysis: the mean first passage time and the mean first recurrence time. The former -denoted as mpij- indicates the expected number of periods to reach state j for the first time starting at state i. The latter indicates the expected time it takes to return to a given state i for the first time. We denote this mrii. The mean first passage time can be computed using the elements of the transition matrix and solving for a system of linear equations: ∑ Then, using the off-diagonal values of the mean first passage time, we can substitute into the following formula to calculate the mean first recurrence time as follows: ∑ ( ) We compute two different transition matrices using two alternative relative income groups. In both cases, we divide countries into five relative income groups (states 1 to 5). The selection of the relative income intervals, along with the choice of the period intervals, plays a fundamental role in determining the elements of the transition probability matrix. By adopting two different relative income classifications, we try to reduce some of the arbitrariness associated with the choice of the income groups. We assume that the process is time-invariant, as it is often done in the literature. In one set of transition matrices, countries are classified as follows: those with incomes less than 0.15 of the US income; those between 0.15 and 0.30 of the US income; those between 0.30 and 0.45 of US income; those between 0.45 and 0.60 of US income; and those with income higher than 0.60 of US income. In the second relative income classification, we adopt a similar country grouping as in Kremer et al. (2001).20 Countries are sorted into five 19 See, for example, Hamilton (1994) for a general reference on Markov chains. Kremer et al. (2001) also utilizes the half life to analyze the transition path towards the stationary distribution. 20 Quah (1993) divides countries into five categories: countries with less than 1/4 of the world average per capita income, those between 1/4 and 1/2, those between 1/2 world average and world average income, those between 1 and 2, and those with income greater than 2 times the world average income. As pointed out by Kremer et al. (2001), 13 groups: those with incomes less than 1/16 of US income; those between 1/16 and 1/8 of US income; those between 1/8 and 1/4 of US income; those between 1/4 and 1/2 of US income; and those with income higher than 1/2 of US income. Each cell of Panel I of Table 4 and Table 5 indicates the number of transitions from one relative income group to another one. Similarly, each cell of Panel II shows the probability of making such transition. The sum of each row of Panel I amounts to the number of states observed between 1950 and 2008.21 The sum of each row of Panel II should equal unity. Hence, a cell aij in Panel I illustrates the total number of transitions from group i to group j over ten-year intervals for the period in consideration. pij, on the other hand, in Panel II, denotes the probability of transition between relative income groups i and j. These transition matrices can be used to describe the evolution of world income over time. For instance, after n periods, the relative distribution of income at time t + n will be given by , where is a N x 1 vector describing the relative income distribution at time t. Table 4: Ten-year transition matrix 1950-2008 Panel I: Total number of transitions Panel III: Mean first passage time [0-0.15) [0.15-0.3) [0.3-0.45) [0.45-0.6) ≥0.6 [0-0.15) [0.15-0.3) [0.3-0.45) [0.45-0.6) ≥0.6 [0-0.15) 356 20 0 0 0 [0-0.15) 20.9 146.0 63.1 75.0 69.6 [0.15-0.3) 25 88 23 2 0 [0.15-0.3) 374.5 52.3 44.3 56.2 50.8 [0.3-0.45) 1 11 19 14 3 [0.3-0.45) 723.4 127.2 106.5 34.7 28.0 [0.45-0.6) 0 0 4 17 11 [0.45-0.6) 973.4 250.0 250.0 35.6 9.6 ≥0.6 0 0 0 1 87 ≥0.6 1061.4 338.0 338.0 88.0 1.1 Panel II: Transition matrix p(i,j) [0-0.15) [0.15-0.3) [0.3-0.45) [0.45-0.6) ≥0.6 Ergodic distribution for transition matrix [0-0.15) 0.95 0.05 0.00 0.00 0.00 0.05 0.01 0.01 0.03 0.90 [0.15-0.3) 0.18 0.64 0.17 0.01 0.00 Eigenvalues for transition matrix [0.3-0.45) 0.02 0.23 0.40 0.29 0.06 0.20 0.56 0.76 0.98 1.00 [0.45-0.6) 0.00 0.00 0.13 0.53 0.34 Half-life for transition matrix ≥0.6 0.00 0.00 0.00 0.01 0.99 40.0 The total number of states observed is as follows: 376, 138, 48, 32 and 88. Not surprisingly, Table 4 shows that transitions between adjacent cells are not uncommon in a ten- year span. On the other hand, transitions between non-adjacent groups are less frequent. Moreover, results in Table 4 suggest that countries are more likely to stay in their relative income group over a ten-year period: the probability of being in group i and remaining in group i is greater than the probability of being in group i and moving to group j after a decade –i.e. pii > pij for i≠j for each single row. In other words, diagonal elements are larger than non-diagonal elements. Table 4 indicates that low income countries relative to the US are much more likely to remain in that group (p11 equals 0.95). This may suggest the low income countries may be subject to a poverty trap. On the other hand, the probability of being a high income country and remaining as such is also high (p55 equals 0.99), indicating that once you join the high income this can lead to a distribution in which most countries have more than twice the world average income. Countries can be classified relative to a leading country or a group of leading countries to avoid this problem. 21 We also include in the tables the transition for the final years of our sample, i.e. the transitions from 2000 to 2008. 14 country club, you are very likely to remain part of this selective group. Only one demotion episode has been recorded for 1950-2008. How can we evaluate the so-called MIC trap? We can either look at the probability of starting as a middle income country and staying as such, or we can also look at the probability of moving out of that relative income bracket. Alternatively, we can compare the expected mean first passage times, i.e. the expected time to move from one income category to another one (Panel III).22 From our relative income classification, we could define a lower middle income country as those countries with relative incomes between 0.15 and 0.30, “middle-middle� as 0.30 to 0.45, and upper-middle income country as having an income per capita greater than 0.45 but less than 0.6 of the income per capita of the US.23 Now that we have a working definition for a middle income country, what does Table 4 tell us about the existence of the MIC trap? First, the probability of being a middle income country and staying as such is lower than the probability of being a low or high income country and remaining in those relative income bracket groups (p22, p33 and p44 equal 0.64, 0.40 and 0.53, respectively). Moreover, in fourteen out of forty-eight occasions, “middle-middle� income countries were able to move up to the upper middle income bracket and three transitions from “middle-middle� income to high income were recorded over the period under study. “Middle- middle� income countries have also been subject to negative shocks and slowdowns in growth, resulting in "downgrades" in their relative income status. However, these backwards transitions were less frequent than the upwards transitions. The lower-middle income group is more problematic – with more persistence – and has only an equally likely chance of moving up or down the ladder. Table 5: Ten-year transition matrix 1950-2008 Panel I: Total number of transitions Panel III: Mean first passage time [0-1/16) [1/16-1/8) [1/8-1/4) [1/4-1/2) ≥1/2 [0-1/16) [1/16-1/8) [1/8-1/4) [1/4-1/2) ≥1/2 [0-1/16) 178 14 0 0 0 [0-1/16) 6.2 37.5 38.3 68.0 113.2 [1/16-1/8) 31 86 23 0 0 [1/16-1/8) 71.1 13.2 24.6 54.3 99.5 [1/8-1/4) 1 24 94 26 0 [1/8-1/4) 160.9 23.8 23.9 29.7 74.9 [1/4-1/2) 0 3 14 67 15 [1/4-1/2) 244.4 103.5 103.7 22.9 45.2 ≥1/2 0 0 0 1 105 ≥1/2 350.4 209.5 209.7 106.0 1.4 Panel II: Transition matrix p(i,j) [0-1/16) [1/16-1/8) [1/8-1/4) [1/4-1/2) ≥1/2 Ergodic distribution for transition matrix [0-1/16) 0.93 0.07 0.00 0.00 0.00 0.16 0.05 0.04 0.04 0.70 [1/16-1/8) 0.22 0.61 0.16 0.00 0.00 Eigenvalues for transition matrix [1/8-1/4) 0.01 0.17 0.65 0.18 0.00 0.41 0.61 0.85 0.99 1.00 [1/4-1/2) 0.00 0.03 0.14 0.68 0.15 Half-life for transition matrix ≥1/2 0.00 0.00 0.00 0.01 0.99 50.6 22 By definition, the mean first passage time of a diagonal element is zero. For illustrative purposes, the diagonal elements of the matrix in Panel III are comprised of the mean first recurrence time, or in other words, the expected time it takes to move away from a diagonal element and then return to that state for the first time. 23 For example, in 2008 (Maddison data), Dominican Republic (0.14), Peru (0.17), Romania (0.16) and Sri Lanka (0.16) are near the 0.15 threshold. Uruguay (0.32) and Malaysia (0.33) are just above the 0.30 threshold. Chile (0.42) and Latvia (0.48) are near the 0.45 threshold. Estonia (0.64) and South Korea (0.63) are just over the 0.60 threshold. 15 Table 5 tells a very similar story. With this relative income grouping, the distribution of states across brackets seems more balanced: 192, 140, 145, 99, and 106 for income groups 1, 2, 3, 4, and 5 respectively. As with the previous income classification, jumping between non- adjacent cells is a rare occurrence. It is much more likely for a country to stay in their relative income group over a ten-year period. Countries in the low income bracket find it difficult to move out from there (p11 equals 0.93). Similarly, countries that reached high income status tend to stay in that relative income bracket (p55 equals 0.99). This second classification opens up two tiers that are both roughly within the “less-than 0.15� (1/8 =0.125) category of Table 4. One worrisome statistic is that this second lowest category appears to be more downwardly mobile than upwardly mobile over the period. Finally, Panel II of Table 5 shows that there is upwards and downwards mobility for middle income countries. For group 4, out of a total of 99 cases, in seventeen occasions countries have moved to a lower income bracket, whereas in fifteen occasions middle income countries have become high income countries. One common theme across both income classifications is that the convergence process towards the stationary distribution is painfully slow. The asymptotic half-life is estimated at 35.4-40.0 for the ten-year period matrices using the relative income per capita classification in Table 4, and 50.6-53.4 using the relative income classification in Table 5. Since each step or period is measured at ten years, for Tables 4 and 5 this implies that it will roughly take 400 years and 506 years to cut the distance between the current and the ergodic distribution in half. Since in the ergodic distribution, most countries are “HICs,� these half -life calculations are broadly consistent with the simple, intuitive calculations presented in Table1. Now we examine the mean first passage time matrices depicted in Tables 4 and 5. On the off-diagonal cells, element mp45 is the smallest, and the sum of mp45 and mp35 is smaller than the sum of mp24 and mp34: it takes less time to move from upper-middle or middle-middle to HIC status than from lower-middle or middle-middle to upper-middle. This corresponds to the higher combined probability of this occurrence. Using the other relative income classification (Table 5), we observe that in this case element mp45 is larger than mp34, and the sum of mp45 and mp35 is larger than the sum of mp24 and mp34: it takes more time to move from upper-middle or middle-middle to high income than from lower-middle or middle-middle to upper-middle income status. This corresponds to the sum of p24 and p34 being greater than the sum of p45 and p35. However, it should be noted that this is not much larger. The intuition behind the differences between the matrices in Tables 4 and 5 is that the income brackets become wider for upper middle income countries, and that this may result in longer transition times to move from one income category to the other. Overall, the analysis of the transition probability and the mean first passage time matrices suggests very little or no evidence of the existence of a "middle income trap": the upper middle to high income transitions appear to be as likely as lower middle income to upper middle income transitions. In terms of the expected time it takes to move from upper middle income to high income status, we find very little evidence supporting the claim that it takes longer to make this transition than the transition from lower middle income to middle income. 16 b. Robustness Checks We perform several robustness checks. The results for our first relative income grouping remain unchanged, whereas, as we will see, they are less supportive of a MIC trap for the second relative income classification. First, we examine whether transitions (or lack of) from one relative income category to the other are the result of business cycle fluctuations, especially for those cases where the relative level of income is near one of the relative income thresholds. While we previously argued that this is not an important concern given the ten-year period considered for the transition matrices, in Appendix Table A3.1 we apply the Hodrick-Prescott filter to correct for high frequency fluctuations in income per capita.24 Then, we recalculate the transition matrices relying on the detrended series. The transition matrices computed with the filtered series looked very similar to the previous ones, suggesting little change from our previous findings. The analysis of the mean first passage time matrices also yields similar conclusions. Second, we exclude from our sample all countries that have a shorter time series of income per capita –those countries for which the series start in 1990 in our sample. Elements of the mean first passage time matrix for the second relative income grouping mp45 and mp34 are of similar magnitude, and the sum of mp45 and mp35 is about the same as the sum of mp24 and mp34: it takes more or less the same time to move from upper-middle or middle-middle to high income than lower-middle or middle-middle to upper-middle. This logically corresponds to the fact that the sum of p24 and p34 is about the same as the sum of p45 and p35. Finally, we take a longer transition period. The rationale behind doing this is twofold. First, the shorter the time period, the lower the probability to jump from one income class to another. Second, all the periods are of equal length.25 Not surprisingly, for the 14-year period transition matrices, the half-life is shortened to 29.8 and 43.8 periods for the different income categorizations. This may imply less periods or steps, but each period spans a longer time. The mean first passage time matrix (Table A3.3) suggests that it takes more or less the same time to move from upper-middle or middle-middle to high income than lower-middle or middle-middle to upper-middle, and the sum of the corresponding elements (p24 and p34, on the one hand, and p45 and p35, on the other) also points in this direction. Overall, this section reinforces the idea that the transition probabilities and the expected transition time from upper middle income to high income do not differ radically from the transition from lower middle income to upper middle income, rendering little support to the existence of a middle income trap. c. Patterns of Growth: Do MICs Look Different? Relevant to the MIC trap discussion is the literature on sudden shifts in growth patterns – whether they are accelerations (Hausmann et al, 2004) or slowdowns (Eichengreen et al, 2011). In terms of empirical evidence of a “trap,� Eichengreen et al find an empirical regularity in a 24 We apply a smoothing parameter of 6.25, the standard for annual data. 25 One of the reasons behind taking 2000-2008 as the last period for the ten-year period transition analysis was to use all the years available in our time-series. 17 panel of countries: that at about $17,000 (2005 PPP adjusted dollars) fast growing economies’ growth rates tend to slow down by two percentage points. It is not clear that this is evidence of a MIC trap. First of all, the slowdown threshold is at a level that would probably correspond to a high-income country by World Bank standards, depending upon the PPP adjustment factor involved. Secondly, a slowdown by two percentage points in per capita growth – for example, from 9 to 7 percent or 8 to 6 percent certainly maintains a country on a rapid pace towards convergence, as discussed in sections III and IV above. Finally, if we were to define the MIC trap relative to a leading country (or group of leading countries), a more relevant metric would be the country's growth vis-à-vis the leader's. Aiyar et al (2013) find that TFP slowdowns in MICs are more frequent than in HICs and LICs. They explore structural, institutional, economic, and policy determinants of these slowdowns. A more general approach to patterns of growth is laid out in Pritchett (2000), motivated by the fact that a single time trend may not represent an adequate characterization of the evolution of GDP per capita for most countries. Here we reproduce his results with another decade or so of data. The additional decade was one in which developing countries, in general, and middle-income countries, in particular, experienced more favorable growth. Various macroeconomic crisis episodes – the multiple Latin American episodes, the Russian crisis, the Asian crisis—were left behind, and faster and more stable growth was restored across a wide variety of developing countries. Table 6: Patterns of Growth Since 1950, by number of countries by income classification End-year WB Steep Hills Plateaus Mountains Plains (Mild) classification* below: Hills Accelerators LICs (39) 0 1 6 8 10 14 L-MICs (31) 2 6 6 6 4 7 U-MICs (23) 1 6 6 0 1 9 HICs (11) 0 8 2 0 0 1 o/w HIC (21) 7 10 1 0 0 3 “Escapees�** (Germany!) Source: Authors’ calculations based on data from Maddison, World Bank. *In 1950, WB classification did not exist, so we use the end-year. (Admittedly, this opens up a tautology.) **Relative standard for non-HIC status in 1950 was used, given differences in Maddison and World Bank data. Less than 50 percent of US in 1950 (or initial data year for that country) was used as threshold. In this section we report several of Pritchett's (2000) statistics and we also include the average growth differential of a given country relative to the United States, which may prove to be useful in explaining the evolution of a country's relative income status. Pritchett's (2000) characterization of growth patterns into steep hills, hills, plateaus, mountains, plains and (mild) accelerators is very informative. The distribution of these patterns of growth is described in Table 6 below. The methodology involves identifying break points in OLS growth rates across countries and then categorizing the pattern of growth based on the pre-break growth rate and post-break growth rate. For example, plateaus would be a pattern whereby countries grew by greater than 1.5 percent pre-break, but then leveled off into a slower, but still positive growth 18 rate.26 The complete results of the full calculations by country are presented in Appendix 2. The summary of how countries fit into patterns is presented in Table 6. Since the World Bank Classifications did not exist in 1950, Table 6 presents a bit of a tautology: end-period HICs would tend to have performed better than non-HICs during the preceding decades, and end-period LICs would have tended to perform worse. That is how they ended up in these categories. On the other hand, in relative terms, our earlier analysis of transition matrices revealed that there is a lot of persistence across relative categories. Table 6 then shows that the actual patterns of growth across MICs vary substantially. There are a few “super-fast� examples, some fast examples, and many (mild) accelerators that offer some hope for the future for MICs. Unfortunately, there are also many “plateaus�, “mountains� (growth collapses) and “plains� observed. Similarly, there are different paths to success when examining the last row of Table 6. While most are “hills� or even “steep hills� (the super-stars), there are also several examples of more gradual paths to success. Take, for instance, the set of countries that were successful in making the transition to high income countries according to the two income groupings that we have adopted in the previous section. We then compare selected indicators for those countries that still remain MICs (“still� MICs) and fell within one standard deviation interval of the initial relative income levels of those countries that successfully became HICs. For Table 4, we have that the median and mean relative income for those countries that transitioned into HIC status were 0.36 and 0.32. The standard deviation was about 0.15, giving us a lower bound of 0.17 and an upper bound of 0.47.27 For Table 5, the median and mean initial relative income for the “new HICs� was 0.26 and 0.28, with a standard deviation of 0.12, giving us a lower and upper bound of 0.16 and 0.39.28 A very revealing picture emerges from Tables 7 and 8. The initial HICs have more or less grown at a similar pace as the United States –the average growth differential is negligible. In general, they are characterized by relatively low volatility –provided that we believe a single trend growth describes well the growth pattern of a given country—, and with a few exceptions, most of them are characterized by what Pritchett denominates “hills�. The average of the ratio of the final GDP per capita and the minimum GDP per capita for the initial HICs is about 3.5. 26 According to Pritchett's classification, we have: i) steep hills: pre- and post-break OLS growth are higher than 3%, ii) hills: pre- and post-break OLS growth are higher than 1.5%, iii) plateaus: pre-break OLS growth is higher than 1.5%, post-break OLS growth is positive but less than 1.5%, iv) mountains: pre-break OLS growth is higher than 1.5%, post-break OLS growth is negative, v) plains: pre- and post-break OLS growth are less than 1.5%, vi) (mild) accelerators: pre-break OLS growth less than 1.5%, post-break OLS growth is higher than 1.5%. 27 The maximum value for the relative income in 1950 was 0.57 and the minimum was 0.09 for the set of “new HICs�. 28 The maximum value for the relative income in 1950 was 0.44 and the minimum was 0.09 for the set of “new HICs� using the relative income thresholds depicted in Table 5. 19 Table 7: Patterns of Growth Since 1950: Old HICs, new HICs, and “still MICs� Avg. GDP Final to Initial GDP Final GDP Growth Country OLS growth Volatility growth minimum to US GDP to US GDP pattern LP differential GDP Initial HICs Australia 0.022 0.033 0.001 3.414 0.775 0.812 Hills Canada 0.022 0.060 0.001 3.465 0.763 0.810 Hills Denmark 0.023 0.064 0.001 3.549 0.726 0.790 Hills Netherlands 0.024 0.068 0.004 4.118 0.627 0.792 Hills Sweden 0.021 0.082 0.002 3.606 0.708 0.783 Hills Switzerland 0.015 0.082 -0.003 2.770 0.948 0.805 Plateaus United Kingdom 0.021 0.024 0.001 3.421 0.726 0.762 Hills Mean: Initial HICs 0.021 0.059 0.001 3.478 0.753 0.793 New HICs Austria 0.030 0.110 0.012 6.511 0.388 0.774 Hills Belgium 0.026 0.075 0.005 4.331 0.571 0.759 Hills Estonia 0.050 0.129 0.020 2.477 0.466 0.640 Accelerators Finland 0.029 0.088 0.010 5.723 0.445 0.781 Hills France 0.025 0.100 0.005 4.285 0.542 0.713 Hills Germany 0.025 0.131 0.009 5.360 0.406 0.667 Plateaus Hong Kong 0.049 0.097 0.027 14.293 0.232 1.017 Steep hills Ireland 0.037 0.103 0.016 8.080 0.361 0.895 Steep hills Italy 0.029 0.128 0.010 5.685 0.366 0.639 Hills Japan 0.042 0.239 0.023 11.879 0.201 0.732 Hills Singapore 0.051 0.141 0.025 12.856 0.232 0.902 Steep hills South Korea 0.059 0.105 0.036 24.922 0.089 0.629 Steep hills Spain 0.037 0.122 0.018 9.003 0.229 0.632 Steep hills Taiwan 0.059 0.095 0.035 22.851 0.096 0.671 Steep hills Mean: New HICs 0.039 0.119 0.018 9.875 0.330 0.746 "Still" MICs Bolivia 0.008 0.103 -0.013 1.878 0.201 0.095 Plateaus Bulgaria 0.020 0.227 0.010 5.382 0.173 0.285 Plateaus Brazil 0.023 0.135 0.003 3.846 0.175 0.206 Plateaus Chile 0.020 0.145 0.003 3.593 0.384 0.423 Accelerators Costa Rica 0.021 0.094 0.004 4.117 0.205 0.258 Hills Cuba 0.006 0.147 -0.008 2.037 0.214 0.121 Accelerators Greece 0.034 0.169 0.017 8.544 0.200 0.525 Hills Guatemala 0.013 0.104 -0.007 2.243 0.218 0.143 Plateaus Hungary 0.018 0.134 0.003 3.831 0.259 0.305 Hills Israel 0.031 0.152 0.012 6.367 0.295 0.575 Hills Jordan 0.018 0.161 0.003 3.509 0.174 0.183 Plateaus Lebanon 0.007 0.159 -0.007 2.300 0.254 0.143 Accelerators Mauritius 0.032 0.122 0.011 5.836 0.260 0.466 Hills Namibia 0.009 0.110 -0.007 2.116 0.226 0.147 Plateaus Nicaragua -0.009 0.236 -0.017 1.308 0.169 0.054 Mountains Panama 0.021 0.130 0.002 3.605 0.200 0.214 Hills Paraguay 0.016 0.133 -0.008 2.204 0.166 0.106 Plains Peru 0.007 0.141 -0.005 2.335 0.241 0.173 Hills Poland 0.020 0.129 0.005 4.152 0.256 0.326 Hills Portugal 0.036 0.124 0.014 6.919 0.218 0.463 Hills Puerto Rico 0.034 0.119 0.014 7.031 0.224 0.483 Hills Seychelles 0.024 0.100 0.001 3.195 0.200 0.196 Mountains South Africa 0.007 0.110 -0.010 1.891 0.265 0.154 Plateaus Turkey 0.026 0.060 0.008 4.970 0.170 0.259 Hills Mean: "Still" MICs 0.018 0.135 0.001 3.884 0.223 0.263 Memorandum items: USA OLS growth Volatility Avg. GDP growth differential United States 0.021 0.032 0.000 All countries 0.017 0.138 -0.001 Note: GDP per capita series for Estonia starts in 1990. 20 Table 8: Patterns of Growth Since 1950: Old HICs, new HICs, and “still MICs� Avg. GDP Final to Initial GDP Final GDP Growth Country OLS growth Volatility growth minimum to US GDP to US GDP pattern LP differential GDP Initial HICs Australia 0.022 0.033 0.001 3.414 0.775 0.812 Hills Belgium 0.026 0.075 0.005 4.331 0.571 0.759 Hills Canada 0.022 0.060 0.001 3.465 0.763 0.810 Hills Denmark 0.023 0.064 0.001 3.549 0.726 0.790 Hills France 0.025 0.100 0.005 4.285 0.542 0.713 Hills Netherlands 0.024 0.068 0.004 4.118 0.627 0.792 Hills New Zealand 0.014 0.046 -0.007 2.437 0.884 0.598 Plateaus Sweden 0.021 0.082 0.002 3.606 0.708 0.783 Hills Switzerland 0.015 0.082 -0.003 2.770 0.948 0.805 Plateaus United Kingdom 0.021 0.024 0.001 3.421 0.726 0.762 Hills Mean: Initial HICs 0.021 0.063 0.001 3.540 0.727 0.762 New HICs Austria 0.030 0.110 0.012 6.511 0.388 0.774 Hills Estonia 0.050 0.129 0.020 2.477 0.466 0.640 Accelerators Finland 0.029 0.088 0.010 5.723 0.445 0.781 Hills Germany 0.025 0.131 0.009 5.360 0.406 0.667 Plateaus Greece 0.034 0.169 0.017 8.544 0.200 0.525 Hills Hong Kong 0.049 0.097 0.027 14.293 0.232 1.017 Steep hills Ireland 0.037 0.103 0.016 8.080 0.361 0.895 Steep hills Israel 0.031 0.152 0.012 6.367 0.295 0.575 Hills Italy 0.029 0.128 0.010 5.685 0.366 0.639 Hills Japan 0.042 0.239 0.023 11.879 0.201 0.732 Hills Singapore 0.051 0.141 0.025 12.856 0.232 0.902 Steep hills Slovenia 0.036 0.054 0.013 1.951 0.468 0.583 Steep hills South Korea 0.059 0.105 0.036 24.922 0.089 0.629 Steep hills Spain 0.037 0.122 0.018 9.003 0.229 0.632 Steep hills Taiwan 0.059 0.095 0.035 22.851 0.096 0.671 Steep hills Mean: New HICs 0.040 0.124 0.019 9.767 0.298 0.711 "Still" MICs Bolivia 0.008 0.103 -0.013 1.878 0.201 0.095 Plateaus Bulgaria 0.020 0.227 0.010 5.382 0.173 0.285 Plateaus Brazil 0.023 0.135 0.003 3.846 0.175 0.206 Plateaus Chile 0.020 0.145 0.003 3.593 0.384 0.423 Accelerators Costa Rica 0.021 0.094 0.004 4.117 0.205 0.258 Hills Cuba 0.006 0.147 -0.008 2.037 0.214 0.121 Accelerators Djibuti -0.008 0.148 -0.023 1.149 0.157 0.040 Plains Guatemala 0.013 0.104 -0.007 2.243 0.218 0.143 Plateaus Hungary 0.018 0.134 0.003 3.831 0.259 0.305 Hills Jordan 0.018 0.161 0.003 3.509 0.174 0.183 Plateaus Lebanon 0.007 0.159 -0.007 2.300 0.254 0.143 Accelerators Malaysia 0.039 0.101 0.013 7.284 0.163 0.330 Steep hills Mauritius 0.032 0.122 0.011 5.836 0.260 0.466 Hills Namibia 0.009 0.110 -0.007 2.116 0.226 0.147 Plateaus Nicaragua -0.009 0.236 -0.017 1.308 0.169 0.054 Mountains Panama 0.021 0.130 0.002 3.605 0.200 0.214 Hills Paraguay 0.016 0.133 -0.008 2.204 0.166 0.106 Plains Peru 0.007 0.141 -0.005 2.335 0.241 0.173 Hills Poland 0.020 0.129 0.005 4.152 0.256 0.326 Hills Portugal 0.036 0.124 0.014 6.919 0.218 0.463 Hills Puerto Rico 0.034 0.119 0.014 7.031 0.224 0.483 Hills Seychelles 0.024 0.100 0.001 3.195 0.200 0.196 Mountains Singapore 0.051 0.141 0.025 12.856 0.232 0.902 Steep hills South Africa 0.007 0.110 -0.010 1.891 0.265 0.154 Plateaus Turkey 0.026 0.060 0.008 4.970 0.170 0.259 Hills Mean: "Still" MICs 0.018 0.132 0.001 3.983 0.216 0.259 Memorandum items: USA OLS growth Volatility Avg. GDP growth differential USA 0.021 0.032 0.000 All countries 0.017 0.138 -0.001 Note: GDP per capita series for Estonia and Slovenia start in 1990. 21 The group of new HICs is comprised of hills, steep hills, accelerators, and one plateau (Germany, which is not surprising given that the break year is 1990). While volatility is higher, these countries have grown much faster than the United States: the growth differential average is 1.8-1.9%. The average of the ratio of the final GDP per capita and the minimum GDP per capita for the new HICs is about 9.8-9.9, almost three times larger than for the initial HICs. Thus, it is not surprising why these countries have joined the HIC group. Finally, we take a look at those countries that remain middle income, or “still� MICs. Volatility is the highest among the three groups depicted here. Their average growth differs very little from the leading country or the initial HICs. The average ratio of final to minimum GDP is between 3.9 and 4, not much different from the ratio of the initial HICs. Finally, we see a wide range of growth patterns, including hills, plateaus, plains, accelerators, and mountains. We do not observe, however, steep hills, a very common categorization among new HICs or “escapees�. Table 9: Time required to cross the lower and upper threshold of upper-middle countries Year for Year for GDP to US GDP to US Average growth Average growth differential for the Average growth differential until lower upper Number of Country GDP - Lower GDP - Upper differential for 1950- transition period from MIC to HIC reaching the upper UMIC bondary of boundary of years threshold threshold 2008 status threshold UMIC UMIC Countries that have crossed both lower and upper thresholds of the UMIC: 0.45 and 0.60 of US GDP per capita between 1950 and 2008 Austria 1955 0.46 1969 0.60 14 0.012 0.022 0.024 Finland 1951 0.45 1970 0.64 19 0.010 0.019 0.019 Germany 1953 0.46 1958 0.63 5 0.009 0.062 0.058 Hong Kong 1976 0.47 1982 0.62 6 0.027 0.055 0.032 Italy 1957 0.47 1968 0.61 11 0.010 0.028 0.030 Ireland 1980 0.46 1996 0.61 16 0.016 0.020 0.012 Japan 1966 0.46 1970 0.65 4 0.023 0.080 0.061 Singapore 1980 0.49 1990 0.61 10 0.025 0.030 0.026 South Korea 1994 0.45 2007 0.61 13 0.036 0.026 0.036 Spain 1973 0.46 2006 0.61 33 0.018 0.009 0.018 Taiwan 1991 0.46 2004 0.61 13 0.035 0.027 0.036 Mean 13.1 0.020 0.034 0.032 Countries that have crossed both lower and upper thresholds of the UMIC: 0.25 and 0.50 of US GDP per capita between 1950 and 2008 Greece 1957 0.26 2007 0.51 50 0.017 0.015 0.017 Hong Kong 1956 0.25 1978 0.50 22 0.027 0.034 0.029 Japan 1955 0.25 1968 0.54 13 0.023 0.058 0.057 Singapore 1969 0.26 1981 0.50 12 0.025 0.060 0.026 South Korea 1983 0.26 1997 0.50 14 0.036 0.049 0.039 Spain 1954 0.26 1975 0.51 21 0.018 0.036 0.034 Taiwan 1978 0.26 1993 0.51 15 0.035 0.048 0.040 Mean 21.0 0.026 0.043 0.035 In Table 9, we list the countries that have crossed both the lower and upper thresholds of what we classify as upper middle income countries between 1950 and 2008.29 On average, it took about 13.1 years (or 21 years, depending on the arbitrary relative income threshold we choose) to make the transition. A first look at the last three columns suggests that all these countries grew much faster than the United States. While this is not a rigorous test, Table 9 shows that the average growth differential until reaching the upper UMIC threshold is slightly 29 Estonia, for example, was left out because only a few years were available and the transition period from upper MIC to HIC spans most of the years available in the sample (1990-2006), thus, making the comparison of growth differentials across periods less meaningful. 22 lower than the average growth differential for the transition period, providing little support to this idea of a growth slowdown once a country reaches upper middle income status, at least for this set of countries. We now proceed to compare whether some of these countries have slowed down once they have become upper middle income countries. As we pointed out earlier, several authors have suggested that middle income countries may be in a crossroad between low-skilled labor intensive activities linked to low income countries and more sophisticated products that are associated with high income countries which, in turn, may translate in the need to move from a development model based on low skills-low wages towards a productive structure based on innovation and high skills. This transition process may be neither smooth nor straightforward. On the other hand, these countries are also relatively closer to their respective steady-states, which may also suggest a moderation in growth. We therefore examine this growth slowdown hypothesis by looking at upper middle income countries that failed to graduate (the cases of Greece, Israel, Portugal, and Puerto Rico) and countries that succeeded in becoming high income countries (Ireland, Korea, Singapore, Spain and Taiwan, China) based on the 0.45-0.60 UMIC classification.30 For the former group, a simple one-tailed means test for each country suggests that there has been a slowdown in growth -measure as growth differential relative to the United States- after crossing the upper middle income threshold. For the latter, however, only for Spain we find that there has been a statistically significant moderation in growth. We perform a similar exercise but using the 0.25-0.50 relative income definition. We look at the cases of new HICs: Korea, Singapore, and Taiwan, China. A means test for each of these countries suggests that only for the case of Singapore we observe a difference in growth that is statistically significant -in this case an acceleration rather than a slowdown. Among countries that have crossed the 0.25 relative income threshold but failed to become HICs (Brazil, Costa Rica, Jamaica, Panama, Thailand, Malaysia, Jordan, Namibia and Seychelles) 31 , we observe a statistically significant slowdown only in Brazil and Jamaica. In both cases, once these countries crossed the 0.25 threshold, average growth differential has become negative. Finally, we take a quick look at the cases of Thailand and Malaysia, the latter cited as an example of the so-called "middle-income trap". Both countries crossed the 0.25 relative income threshold in the early 1990s. However, the Asian crisis of 1997 adversely impacted the progression towards HIC status. In 2008, GDP per capita relative to the US stood at 0.33 and 30 The choice of countries follows two criteria: at least ten yearly observations before crossing the 0.45 threshold and ten yearly observations after crossing the 0.45 threshold for UMICs or ten yearly observations in the UMIC income bracket for countries that jumped into the HIC group (above 0.60). This leaves Japan and Hong Kong SAR, China, out, for example. 31 The choice of countries corresponds to the following criteria: at least ten yearly observations before crossing the 0.25 threshold and ten yearly observations after crossing the 0.25 threshold for UMICs or ten yearly observations in the UMIC income bracket for countries that jumped into the HIC group (above 0.50). 23 0.28 for Malaysia and Thailand, respectively, whereas average growth differential after crossing the UMIC lower threshold was 2% and 1.2% respectively. If we assume that these countries were to keep this pace, it will take about 20 years for Malaysia and about 50 years for Thailand to cross the 0.5 relative income threshold. It should be noted that the average relative growth rates for each of these countries has been dampened by the output contraction during the 1997 Asian crisis. From 2000 onwards, the growth differential with respect to the United States was 2% for Malaysia and 2.7% for Thailand and about 2.5% for both in the last five years of the sample. At the latter pace, it would take about 17 and 24 years respectively to cross the 0.5 relative income per capita threshold. Take, for instance, the relative income per capita of Malaysia and Thailand in 1996 (0.31 and 0.27). Had the Asian Crisis not taken place, Malaysia would have become a high income country by 2016 and Thailand by 2021. VI. Policy Discussion The simplest version of the neoclassical growth model suggests that, if technologies are the same across countries, then higher returns to capital in low capital per worker countries would ensure convergence towards a common level of income per capita and a common growth rate—in other words, unconditional convergence. This has not been the reality – whether one considers all the countries of the world or subsets like MICs. In recent decades, numerous theories have focused on a variety of factors, “endogenous� growth models that explain differences in the rate of technological change removing the assumption of constant returns to scale, and policy, institutional and historical explanations that may result in multiple equilibriums or simply differences across countries (Banerjee and Duflo, 2005; Azariadis and Stachurski, 2005; Parente and Prescott, 2005). The empirical literature has focused on conditional convergence and which factors, or controls, are the most appropriate for explaining the lack of unconditional convergence, ranging from the rate of capital accumulation and population growth/human capital (Mankiw et al, 1992), the role of public infrastructure (Calderón and Servén, 2004) to policies, including the investment climate, and institutions (Dollar and Kraay, 2003; Loayza and Servén, 2009). Jones and Romer (2010) have developed what they call the “New Kaldor Facts� that emerge from the empirical literature. Without entering into all six, two of them are: (a) there is a large degree of variation in modern growth rates and that variation in growth increases with the distance from the technology frontier; and (b) there are large differences in total factor productivity across countries--differences in measured inputs explain less than half of the differences in GDP per capita. This discussion raises doubts about the relevance of using a relative measure of MIC success or failure or “trap.� It may be reasonable to imagine a future in which all countries are “rich� from some absolute perspective; however, given differences in preferences, endowments, geography, initial conditions, historical antecedents and even random shocks, it is hard to imagine a world in which all countries have a nearly identical income per capita. Even if all policies were “perfect,� it is not clear that one would expect rapid or complete convergence. In 24 addition, the results above indicate that transition states have taken a long time to converge. In the 200 years of modern economic history, one might have hoped that more countries could have caught up to the world leaders, but this is a relatively short time span for convergence to occur. It may be a misnomer to consider growth “slowdowns� to be a “trap,� and in some cases, they have represented just a slower convergence path. Our results also show that MICs do not really look that different in terms of transitions across the inter-country distribution of income. Their growth patterns also do not conform to one clear pattern that can be easily characterized as a “trap.� On the other hand, the “MIC Trap� concept is useful for guiding policy discussions. First, it recognizes the particular challenges faced by countries at that stage of development. Secondly, it calls attention to the limited number of MICs that have been fully successful in attaining a truly developed country status – even if absolute incomes have risen and even if non- income dimensions have improved substantially in many countries. At the same time, a certain amount of realism might be added to the discussion. The identification of the small group of fast MIC “escapees� can lead to a form of “outlier worship.� The attempt to grow at 7 or 10 percent could lead to unsustainable policies that eventually create the “trap�-like pattern of dismal growth that MICs are trying to avoid in the first place. Gradualism may be more sustainable and less risky – especially for upper middle income countries. If conditional convergence is the reality, then policies might focus on overcoming the “initial conditions� that inhibit growth – institutions and other factors. 32 Aiyer et al (2013) provide some evidence of potential determinants of MIC slowdowns that are relevant to the debate. In addition, some of the non-income factors that are often used to measure development progress may themselves be the control factors limiting a country’s convergence. P olicies that affect the distribution and increase the inclusiveness of growth (for example, investing in the human capital of the poor) may actually accelerate catch-up, as has been noted in the vast theoretical and empirical literature on economic growth. 32 Felipe et al (2012) conduct an empirical evaluation of these factors in determining exit from their definition of a MIC trap. 25 Appendix 1: Other dimensions of MIC status in historical perspective As noted in the introduction, many other socio economic indicators are highly correlated with income per capita; however, relative performance is often on a different scale. For example, access to safe drinking water is either 100 percent or less. In other words, for some access to service measures of development or human capital attainment, one does not have the case of infinitely “moving goalposts,� as discussed above in term of relative MIC status, as measured by income per capita. In addition, technological change implies that there are services today that many MIC residents enjoy that did not even exist in the late 19th or early 20th century, when today’s HICs were, in fact, MICs by some absolute standard. Table A1.1 below presents some basic data on the status of middle income countries in a number of non-income dimensions. We see that Upper MICs are reaching over 90 percent access to improved water source and electricity. Life expectancy in Upper MICs is above the US level in 1920 or 1960 and is almost 93 percent the level of the US life expectancy. Even Lower MICs have a life expectancy that is 83 percent of the US level. The infant mortality rate in both Upper and Lower MICs is far lower than it was in the US in 1920, but they still lag pretty badly behind the current US level – especially in Lower MICs which suffer from infant mortality that is over 8 times as high as in the US today. Education variables are lagging as well. In general terms, however, while the history of “catch-up� in per capita incomes is not very favorable, the history on a number of non-income social dimensions is substantially better – at least from the perspective of this brief sketch of some relevant social outcomes. Table A1.1: Non-income Dimensions of MICs and Comparison to the United States Upper MIC Lower MIC US in 1920 US in 1960 US in 2010 (2009) (2009) Access to electricity (% of population) 97.3 67.7 Life expectancy at birth, total (years) 72.4 65.2 54.1 67.4 78.2 Mortality rate, infant 19.3 50.7 85.8 26 6.5 (per 1,000 live births) School enrollment, secondary 84.4 59.0 64.3** 84.4** 96.0 (% gross) School enrollment, secondary (% net) 75.9 52.4* 89.5 Physicians (per 1,000 people) 1.7 0.8 1.37 1.48 2.4 Improved water source 91.9 86.5 (% of population with access) Literacy Rate (percent) 93.5 70.9 94 97.8*** Sources: World Development Indicators and Bureau of the Census (1975) Notes: *Data for 2008 instead of 2009 **Slightly different definition: primary and secondary, % of population 5 to 20 years of age ***1959 instead of 1960. US data for 2010 is from WDI. US data for 1920 and 1960 is from Bureau of the Census (1975). 26 Appendix 2: Statistical Annex on Growth Patterns and Country Classifications Table A2.1: List of countries List of countries Afghanistan Guinea Bissau Nepal Albania Greece New Zealand Argentina Guatemala Pakistan Armenia Hong Kong Panama Australia Honduras Peru Austria Croatia Philippines Burundi Haïti Poland Belgium Hungary Puerto Rico Benin India North Korea Burkina Faso Ireland Portugal Bangladesh Israel Paraguay Bulgaria Italy Romania Bosnia Jamaica Rwanda Belarus Jordan Senegal Bolivia Japan Singapore Brazil Kenya Sierra Leone Central African Republic Kyrgyzstan El Salvador Canada Cambodia Somalia Switzerland South Korea Sao Tomé and Principe Chile Laos Slovakia China Lebanon Slovenia Côte d'Ivoire Liberia Sweden Comoro Islands Sri Lanka Swaziland Cape Verde Lesotho Seychelles Costa Rica Lithuania Togo Cuba Latvia Thailand Czech Rep. Morocco Tajikistan Germany Moldova Tunisia Djibouti Madagascar Turkey Denmark Macedonia Taiwan Dominican Republic Mali Tanzania Egypt Burma Uganda Eritrea and Ethiopia Mongolia Ukraine Spain Mozambique Uruguay Estonia Mauritania United States Finland Mauritius Uzbekistan France Malawi West Bank and Gaza United Kingdom Malaysia South Africa Georgia Namibia Zaire (Congo-Kinshasa) Ghana Niger Zambia Guinea Nicaragua Zimbabwe Gambia Netherlands 27 Table A2.2: Growth Patterns and Country Classifications Steep hills: Pre-break OLS growth > 3 percent, Post-break OLS growth 3 > percent - 10 countries GDP to US GDP to US Pre-break Post-break GDP - GDP to US GDP - Pre-break Post- avg. growth avg. growth Initial Initial GDP - Year of Break OLS break OLS differential differential Country Income classification year year Final year Final year break year growth growth with US with US Hong Kong High income: nonOECD 1950 0.232 2008 1.017 1986 0.657 0.053 0.031 0.029 0.023 Ireland High income: OECD 1950 0.361 2008 0.895 1982 0.481 0.032 0.053 0.008 0.026 Malaysia Upper middle income 1950 0.163 2008 0.330 1990 0.221 0.035 0.034 0.007 0.026 Singapore High income: nonOECD 1950 0.232 2008 0.902 1987 0.539 0.053 0.038 0.023 0.027 Slovenia High income: nonOECD 1990 0.468 2008 0.583 2004 0.498 0.030 0.053 0.004 0.036 South Korea High income: OECD 1950 0.089 2008 0.629 1968 0.122 0.035 0.060 0.017 0.044 Spain High income: OECD 1950 0.229 2008 0.632 1961 0.301 0.034 0.032 0.018 0.018 Taiwan High income: nonOECD 1950 0.096 2008 0.671 1986 0.352 0.060 0.044 0.037 0.033 Thailand Lower middle income 1950 0.085 2008 0.281 1988 0.170 0.040 0.034 0.017 0.030 Tunisia Lower middle income 1950 0.117 2008 0.196 1986 0.143 0.033 0.032 0.009 0.011 Hills: Pre-break OLS growth > 1.5 percent, Post-break OLS growth > 1.5 percent - 31 countries GDP to US GDP to US Pre-break Post-break GDP - GDP to US GDP - Pre-break Post- avg. growth avg. growth Initial Initial GDP - Year of Break OLS break OLS differential differential Country Income classification year year Final year Final year break year growth growth with US with US Albania Lower middle income 1950 0.105 2008 0.133 1991 0.078 0.024 0.054 0.001 0.017 Australia High income: OECD 1950 0.775 2008 0.812 1980 0.776 0.025 0.022 -0.001 0.002 Austria High income: OECD 1950 0.388 2008 0.774 1972 0.676 0.047 0.021 0.027 0.004 Belgium High income: OECD 1950 0.571 2008 0.759 1969 0.660 0.030 0.020 0.006 0.005 Burma Low income 1950 0.041 2008 0.100 1987 0.040 0.021 0.071 0.002 0.041 Canada High income: OECD 1950 0.763 2008 0.810 1990 0.813 0.026 0.021 0.002 -0.001 China Lower middle income 1950 0.047 2008 0.216 1976 0.050 0.020 0.064 0.007 0.045 Costa Rica Upper middle income 1950 0.205 2008 0.258 1981 0.248 0.032 0.023 0.009 0.000 Denmark High income: OECD 1950 0.726 2008 0.790 1964 0.827 0.029 0.019 0.007 0.000 Egypt Lower middle income 1950 0.095 2008 0.119 1975 0.087 0.019 0.020 -0.008 0.013 Finland High income: OECD 1950 0.445 2008 0.781 1969 0.585 0.037 0.023 0.012 0.009 France High income: OECD 1950 0.542 2008 0.713 1969 0.717 0.038 0.017 0.014 0.001 Greece High income: OECD 1950 0.200 2008 0.525 1963 0.314 0.051 0.025 0.033 0.013 Hungary High income: OECD 1950 0.259 2008 0.305 1990 0.278 0.027 0.033 0.004 0.002 Israel High income: nonOECD 1950 0.295 2008 0.575 1971 0.569 0.052 0.018 0.032 0.002 Italy High income: OECD 1950 0.366 2008 0.639 1961 0.559 0.051 0.023 0.036 0.004 Japan High income: OECD 1950 0.201 2008 0.732 1966 0.460 0.074 0.026 0.055 0.012 Lesotho Lower middle income 1950 0.037 2008 0.063 1976 0.048 0.030 0.024 0.009 0.012 Mauritius Upper middle income 1950 0.260 2008 0.466 1968 0.196 0.020 0.041 -0.008 0.019 Morocco Lower middle income 1950 0.152 2008 0.111 1992 0.109 0.018 0.020 -0.006 -0.003 Netherlands High income: OECD 1950 0.627 2008 0.792 1964 0.739 0.031 0.020 0.011 0.002 Pakistan Lower middle income 1950 0.067 2008 0.072 1997 0.068 0.026 0.024 0.002 0.000 Panama Upper middle income 1950 0.200 2008 0.214 1988 0.198 0.032 0.017 0.006 -0.006 Peru Upper middle income 1950 0.241 2008 0.173 1988 0.167 0.015 0.023 -0.006 -0.003 Poland Upper middle income 1950 0.256 2008 0.326 1989 0.246 0.026 0.039 0.000 0.013 Portugal High income: OECD 1950 0.218 2008 0.463 1970 0.364 0.047 0.026 0.022 0.009 Puerto Rico High income: nonOECD 1950 0.224 2008 0.483 1969 0.385 0.054 0.025 0.028 0.007 Slovakia High income: OECD 1990 0.335 2008 0.418 2004 0.324 0.026 0.078 -0.003 0.059 Sweden High income: OECD 1950 0.708 2008 0.783 1967 0.811 0.033 0.016 0.007 0.000 Turkey Upper middle income 1950 0.170 2008 0.259 1966 0.193 0.023 0.024 0.007 0.008 United Kingdom High income: OECD 1950 0.726 2008 0.762 1980 0.696 0.022 0.023 -0.001 0.003 28 Table A2.2: Growth Patterns and Country Classifications (cont.) Plateaus: Pre-break OLS growth > 1.5 percent, 0 < Post-break OLS growth < 1.5 percent - 21 countries GDP to US GDP to US Pre-break Post-break GDP - GDP to US GDP - Pre-break Post- avg. growth avg. growth Initial Initial GDP - Year of Break OLS break OLS differential differential Country Income classification year year Final year Final year break year growth growth with US with US Argentina Upper middle income 1950 0.522 2008 0.353 1981 0.403 0.021 0.012 -0.005 -0.007 Burkina Faso Low income 1950 0.050 2008 0.034 1973 0.044 0.023 0.012 0.000 -0.010 Bulgaria Upper middle income 1950 0.173 2008 0.285 1964 0.286 0.052 0.008 0.040 0.001 Bolivia Lower middle income 1950 0.201 2008 0.095 1982 0.131 0.015 0.011 -0.013 -0.013 Brazil Upper middle income 1950 0.175 2008 0.206 1972 0.222 0.029 0.011 0.010 -0.001 El Salvador Lower middle income 1950 0.156 2008 0.094 1979 0.139 0.021 0.012 -0.002 -0.015 Gambia Low income 1950 0.064 2008 0.033 1983 0.049 0.021 0.006 0.000 -0.021 Germany High income: OECD 1950 0.406 2008 0.667 1990 0.687 0.034 0.014 0.015 -0.004 Guinea Low income 1950 0.032 2008 0.020 1981 0.029 0.020 0.007 -0.002 -0.014 Guatemala Lower middle income 1950 0.218 2008 0.143 1982 0.202 0.025 0.011 -0.002 -0.014 Jamaica Upper middle income 1950 0.139 2008 0.118 1974 0.237 0.045 0.004 0.027 -0.021 Jordan Lower middle income 1950 0.174 2008 0.183 1976 0.182 0.019 0.006 -0.001 0.006 Kenya Low income 1950 0.068 2008 0.035 1977 0.056 0.016 0.000 -0.007 -0.014 Mauritania Low income 1950 0.049 2008 0.042 1964 0.065 0.031 0.005 0.007 -0.004 Mozambique Low income 1950 0.119 2008 0.069 1974 0.102 0.020 0.012 -0.002 -0.012 Namibia Upper middle income 1950 0.226 2008 0.147 1962 0.241 0.017 0.004 0.000 -0.009 New Zealand High income: OECD 1950 0.884 2008 0.598 1977 0.682 0.019 0.014 -0.006 -0.007 Philippines Lower middle income 1950 0.112 2008 0.094 1984 0.108 0.023 0.014 0.004 -0.012 South Africa Upper middle income 1950 0.265 2008 0.154 1962 0.267 0.018 0.002 0.002 -0.012 Swaziland Lower middle income 1950 0.075 2008 0.101 1970 0.135 0.052 0.006 0.020 0.000 Switzerland High income: OECD 1950 0.948 2008 0.805 1969 1.056 0.030 0.009 0.005 -0.006 Mountains: Pre-break OLS growth > 1.5 percent, Post-break OLS growth < 0 percent - 14 countries GDP to US GDP to US Pre-break Post-break GDP - GDP to US GDP - Pre-break Post- avg. growth avg. growth Initial Initial GDP - Year of Break OLS break OLS differential differential Country Income classification year year Final year Final year break year growth growth with US with US Burundi Low income 1950 0.038 2008 0.015 1985 0.032 0.017 -0.022 -0.005 -0.028 Comoro Islands Low income 1950 0.059 2008 0.018 1975 0.049 0.030 -0.007 0.006 -0.036 Côte d'Ivoire Lower middle income 1950 0.109 2008 0.035 1968 0.116 0.029 -0.016 0.001 -0.027 Guinea Bissau Low income 1950 0.030 2008 0.020 1964 0.050 0.054 -0.006 0.037 -0.017 Mongolia Lower middle income 1950 0.046 2008 0.032 1990 0.057 0.031 -0.006 0.007 -0.033 Nicaragua Lower middle income 1950 0.169 2008 0.054 1979 0.114 0.024 -0.010 0.000 -0.033 Niger Low income 1950 0.065 2008 0.017 1960 0.067 0.019 -0.015 0.000 -0.027 North Korea Low income 1950 0.089 2008 0.036 1971 0.165 0.039 -0.032 0.022 -0.031 Romania Lower middle income 1950 0.124 2008 0.157 1971 0.210 0.045 -0.001 0.022 -0.004 Seychelles Lower middle income 1950 0.200 2008 0.196 1996 0.242 0.026 -0.012 0.004 -0.010 Togo Low income 1950 0.060 2008 0.019 1964 0.066 0.021 -0.016 0.002 -0.024 West Bank & Gaza Lower middle income 1950 0.100 2008 0.070 1997 0.185 0.031 -0.096 0.013 -0.073 Zaire (Congo-Kinshasa) Low income 1950 0.060 2008 0.008 1962 0.064 0.016 -0.034 -0.002 -0.040 Zambia Low income 1950 0.069 2008 0.027 1959 0.081 0.023 -0.010 0.008 -0.019 29 Table A2.2: Growth Patterns and Country Classifications (cont.) Plains: Pre-break OLS growth < 1.5 percent, Post-break OLS growth < 1.5 percent - 15 countries GDP to US GDP to US Pre-break Post-break GDP - GDP to US GDP - Pre-break Post- avg. growth avg. growth Initial Initial GDP - Year of Break OLS break OLS differential differential Country Income classification year year Final year Final year break year growth growth with US with US Benin Low income 1950 0.113 2008 0.045 1979 0.057 0.000 0.006 -0.026 -0.007 Central African Republic Low income 1950 0.081 2008 0.017 1979 0.041 -0.001 -0.013 -0.021 -0.032 Djibouti Lower middle income 1950 0.157 2008 0.040 1970 0.138 0.008 -0.020 -0.017 -0.026 Ghana Low income 1950 0.117 2008 0.053 1975 0.077 0.010 0.009 -0.011 -0.014 Haïti Low income 1950 0.110 2008 0.022 1976 0.065 -0.005 -0.021 -0.021 -0.032 Honduras Lower middle income 1950 0.137 2008 0.075 1974 0.106 0.009 0.003 -0.016 -0.007 Liberia Low income 1950 0.110 2008 0.026 1990 0.046 -0.003 -0.033 -0.026 -0.018 Madagascar Low income 1950 0.099 2008 0.023 1981 0.050 0.003 -0.011 -0.019 -0.031 Malawi Low income 1950 0.034 2008 0.024 1971 0.033 0.015 0.004 -0.006 -0.005 Paraguay Lower middle income 1950 0.166 2008 0.106 1977 0.143 0.014 0.001 -0.008 -0.007 Sao Tomé & Principe Lower middle income 1950 0.086 2008 0.048 1961 0.082 0.002 0.001 -0.010 -0.008 Somalia Low income 1950 0.111 2008 0.031 1991 0.044 -0.003 -0.001 -0.018 -0.022 Tanzania Low income 1950 0.044 2008 0.024 1962 0.039 0.001 0.002 -0.012 -0.010 Uruguay Upper middle income 1950 0.487 2008 0.317 1967 0.329 -0.002 0.014 -0.020 -0.001 Zimbabwe Low income 1950 0.073 2008 0.025 1997 0.054 0.014 -0.053 -0.005 -0.064 Accelerators: Pre-break OLS growth < 1.5 percent, Post-break OLS growth > 1.5 percent - 34 countries GDP to US GDP to US Pre-break Post-break GDP - GDP to US GDP - Pre-break Post- avg. growth avg. growth Initial Initial GDP - Year of Break OLS break OLS differential differential Country Income classification year year Final year Final year break year growth growth with US with US Afghanistan Low income 1950 0.067 2008 0.028 1992 0.024 0.001 0.037 -0.022 0.008 Armenia Lower middle income 1990 0.261 2008 0.373 1999 0.155 -0.032 0.118 -0.048 0.094 Bangladesh Low income 1950 0.056 2008 0.037 1971 0.038 0.008 0.020 -0.015 -0.003 Belarus Upper middle income 1990 0.310 2008 0.404 1994 0.215 -0.070 0.069 -0.070 0.036 Bosnia Upper middle income 1990 0.161 2008 0.233 1996 0.134 -0.132 0.046 -0.106 0.090 Cambodia Low income 1950 0.050 2008 0.080 1998 0.037 0.014 0.100 -0.003 0.073 Cape Verde Lower middle income 1950 0.047 2008 0.088 1980 0.045 0.009 0.038 -0.012 0.036 Chile Upper middle income 1950 0.384 2008 0.423 1982 0.275 0.012 0.042 -0.005 0.012 Croatia High income: nonOECD 1990 0.317 2008 0.286 1999 0.226 -0.002 0.041 -0.032 0.020 Cuba Upper middle income 1950 0.214 2008 0.121 1991 0.113 0.011 0.034 -0.012 0.000 Czech Republic High income: OECD 1990 0.383 2008 0.413 2002 0.335 0.013 0.053 -0.013 0.033 Dominican Republic Upper middle income 1950 0.107 2008 0.143 1970 0.104 0.014 0.022 -0.005 0.011 Eritrea & Ethiopia Low income 1950 0.041 2008 0.028 1990 0.025 0.013 0.023 -0.012 0.005 Estonia High income: nonOECD 1990 0.466 2008 0.640 1993 0.341 -0.116 0.067 -0.111 0.037 Georgia Lower middle income 1990 0.328 2008 0.192 2003 0.139 -0.032 0.085 -0.060 0.073 India Lower middle income 1950 0.065 2008 0.095 1979 0.048 0.014 0.039 -0.007 0.021 Kyrgyzstan Low income 1990 0.155 2008 0.091 1994 0.078 -0.143 0.030 -0.137 -0.004 Laos Low income 1950 0.064 2008 0.054 1997 0.042 0.012 0.038 -0.009 0.019 Latvia Upper middle income 1990 0.427 2008 0.475 1993 0.229 -0.252 0.075 -0.219 0.042 Lebanon Upper middle income 1950 0.254 2008 0.143 1987 0.100 0.012 0.036 -0.014 0.005 Lithuania Upper middle income 1990 0.373 2008 0.364 1993 0.229 -0.150 0.057 -0.138 0.020 Macedonia Upper middle income 1990 0.171 2008 0.130 1993 0.132 -0.074 0.018 -0.073 -0.007 Mali Low income 1950 0.048 2008 0.037 1993 0.031 0.012 0.032 -0.008 0.007 Moldova Lower middle income 1990 0.266 2008 0.114 1994 0.101 -0.202 0.031 -0.168 -0.013 Nepal Low income 1950 0.052 2008 0.036 1985 0.035 0.008 0.018 -0.011 0.001 Rwanda Low income 1950 0.057 2008 0.033 1993 0.032 0.015 0.032 -0.009 0.008 Senegal Low income 1950 0.132 2008 0.047 1991 0.053 -0.002 0.015 -0.022 -0.008 Sierra Leone Low income 1950 0.069 2008 0.022 1995 0.031 0.008 0.020 -0.015 -0.027 Sri Lanka Lower middle income 1950 0.131 2008 0.157 1971 0.095 0.005 0.033 -0.014 0.012 Tajikistan Low income 1990 0.130 2008 0.049 1996 0.033 -0.235 0.058 -0.206 0.018 Uganda Low income 1950 0.072 2008 0.032 1978 0.038 0.010 0.017 -0.019 -0.008 Ukraine Lower middle income 1990 0.260 2008 0.160 1994 0.135 -0.117 0.046 -0.117 -0.003 United States High income: OECD 1950 1.000 2008 1.000 1964 1.000 0.015 0.020 0.000 0.000 Uzbekistan Low income 1990 0.184 2008 0.169 2002 0.128 -0.015 0.063 -0.034 0.045 30 Appendix 3: Transition Matrices Table A3.1: Transition matrices including East European countries (Hodrick-Prescott filtered time-series) Panel I: Total number of transitions Panel III: Mean first passage time [0-0.15) [0.15-0.3) [0.3-0.45) [0.45-0.6) ≥0.6 [0-0.15) [0.15-0.3) [0.3-0.45) [0.45-0.6) ≥0.6 [0-0.15) 357 21 0 0 0 [0-0.15) 37.8 188.2 68.2 68.3 61.0 [0.15-0.3) 25 89 23 2 0 [0.15-0.3) 662.1 67.8 50.2 50.3 43.0 [0.3-0.45) 0 10 19 15 3 [0.3-0.45) 1292.0 170.2 144.7 29.1 20.5 [0.45-0.6) 0 0 3 16 11 [0.45-0.6) 1624.6 332.7 332.7 36.2 6.5 ≥0.6 0 0 0 1 87 ≥0.6 1712.6 420.7 420.7 88.0 1.1 Panel II: Transition matrix p(i,j) [0-0.15) [0.15-0.3) [0.3-0.45) [0.45-0.6) ≥0.6 Ergodic distribution for transition matrix [0-0.15) 0.94 0.06 0.00 0.00 0.00 0.03 0.01 0.01 0.03 0.93 [0.15-0.3) 0.18 0.64 0.17 0.01 0.00 Eigenvalues for transition matrix [0.3-0.45) 0.00 0.21 0.40 0.32 0.06 0.22 0.56 0.75 0.98 1.00 [0.45-0.6) 0.00 0.00 0.10 0.53 0.37 Half-life for transition matrix ≥0.6 0.00 0.00 0.00 0.01 0.99 35.4 Panel I: Total number of transitions Panel III: Mean first passage time [0-1/16) [1/16-1/8) [1/8-1/4) [1/4-1/2) ≥1/2 [0-1/16) [1/16-1/8) [1/8-1/4) [1/4-1/2) ≥1/2 [0-1/16) 178 12 0 0 0 [0-1/16) 6.3 43.9 42.7 74.7 116.5 [1/16-1/8) 30 89 23 0 0 [1/16-1/8) 84.2 14.8 26.8 58.9 100.7 [1/8-1/4) 1 24 96 26 0 [1/8-1/4) 187.9 28.1 27.8 32.1 73.9 [1/4-1/2) 0 3 11 69 15 [1/4-1/2) 285.2 125.5 125.2 22.5 41.8 ≥1/2 0 0 0 1 104 ≥1/2 390.2 230.5 230.2 105.0 1.4 Panel II: Transition matrix p(i,j) [0-1/16) [1/16-1/8) [1/8-1/4) [1/4-1/2) ≥1/2 Ergodic distribution for transition matrix [0-1/16) 0.94 0.06 0.00 0.00 0.00 0.16 0.05 0.04 0.04 0.72 [1/16-1/8) 0.21 0.63 0.16 0.00 0.00 Eigenvalues for transition matrix [1/8-1/4) 0.01 0.16 0.65 0.18 0.00 0.44 0.63 0.86 0.99 1.00 [1/4-1/2) 0.00 0.03 0.11 0.70 0.15 Half-life for transition matrix ≥1/2 0.00 0.00 0.00 0.01 0.99 53.4 31 Table A3.2: Transition matrices excluding East European countries Panel I: Total number of transitions Panel III: Mean first passage time [0-0.15) [0.15-0.3) [0.3-0.45) [0.45-0.6) ≥0.6 [0-0.15) [0.15-0.3) [0.3-0.45) [0.45-0.6) ≥0.6 [0-0.15) 351 17 0 0 0 [0-0.15) 44.6 189.3 66.1 71.8 66.7 [0.15-0.3) 20 84 19 1 0 [0.15-0.3) 944.9 59.9 44.5 50.1 45.1 [0.3-0.45) 0 6 17 14 2 [0.3-0.45) 1866.9 167.6 148.7 23.4 17.8 [0.45-0.6) 0 0 3 15 11 [0.45-0.6) 2199.3 332.3 332.3 36.8 5.9 ≥0.6 0 0 0 1 87 ≥0.6 2287.3 420.3 420.3 88.0 1.1 Panel II: Transition matrix p(i,j) [0-0.15) [0.15-0.3) [0.3-0.45) [0.45-0.6) ≥0.6 Ergodic distribution for transition matrix [0-0.15) 0.95 0.05 0.00 0.00 0.00 0.02 0.01 0.01 0.03 0.94 [0.15-0.3) 0.16 0.68 0.15 0.01 0.00 Eigenvalues for transition matrix [0.3-0.45) 0.00 0.15 0.44 0.36 0.05 0.24 0.58 0.76 0.98 1.00 [0.45-0.6) 0.00 0.00 0.10 0.52 0.38 Half-life for transition matrix ≥0.6 0.00 0.00 0.00 0.01 0.99 39.3 Panel I: Total number of transitions Panel III: Mean first passage time [0-1/16) [1/16-1/8) [1/8-1/4) [1/4-1/2) ≥1/2 [0-1/16) [1/16-1/8) [1/8-1/4) [1/4-1/2) ≥1/2 [0-1/16) 177 14 0 0 0 [0-1/16) 8.5 43.0 43.0 77.1 109.6 [1/16-1/8) 31 84 19 0 0 [1/16-1/8) 102.0 15.1 29.3 63.4 96.0 [1/8-1/4) 0 21 92 22 0 [1/8-1/4) 261.4 29.4 29.4 34.1 66.7 [1/4-1/2) 0 0 10 59 13 [1/4-1/2) 407.4 146.0 146.0 22.0 32.6 ≥1/2 0 0 0 1 105 ≥1/2 513.4 252.0 252.0 106.0 1.3 Panel II: Transition matrix p(i,j) [0-1/16) [1/16-1/8) [1/8-1/4) [1/4-1/2) ≥1/2 Ergodic distribution for transition matrix [0-1/16) 0.93 0.07 0.00 0.00 0.00 0.12 0.04 0.03 0.05 0.77 [1/16-1/8) 0.23 0.63 0.14 0.00 0.00 Eigenvalues for transition matrix [1/8-1/4) 0.00 0.16 0.68 0.16 0.00 0.45 0.65 0.86 0.99 1.00 [1/4-1/2) 0.00 0.00 0.12 0.72 0.16 Half-life for transition matrix ≥1/2 0.00 0.00 0.00 0.01 0.99 53.4 32 Table A3.3: Transition matrices including East European countries (14-year transitions) Panel I: Total number of transitions Panel III: Mean first passage time [0-0.15) [0.15-0.3) [0.3-0.45) [0.45-0.6) ≥0.6 [0-0.15) [0.15-0.3) [0.3-0.45) [0.45-0.6) ≥0.6 [0-0.15) 234 12 1 0 0 [0-0.15) 35.6 116.2 53.8 59.3 51.0 [0.15-0.3) 20 55 18 2 0 [0.15-0.3) 611.1 49.0 37.7 41.8 33.6 [0.3-0.45) 0 5 12 12 4 [0.3-0.45) 1200.7 105.3 95.0 21.8 12.3 [0.45-0.6) 0 0 3 8 9 [0.45-0.6) 1369.3 168.7 168.7 28.6 4.7 ≥0.6 0 0 0 1 53 ≥0.6 1423.3 222.7 222.7 54.0 1.1 Panel II: Transition matrix p(i,j) [0-0.15) [0.15-0.3) [0.3-0.45) [0.45-0.6) ≥0.6 Ergodic distribution for transition matrix [0-0.15) 0.95 0.05 0.00 0.00 0.00 0.03 0.01 0.01 0.04 0.92 [0.15-0.3) 0.21 0.58 0.19 0.02 0.00 Eigenvalues for transition matrix [0.3-0.45) 0.00 0.15 0.36 0.36 0.12 0.11 0.48 0.70 0.98 1.00 [0.45-0.6) 0.00 0.00 0.15 0.40 0.45 Half-life for transition matrix ≥0.6 0.00 0.00 0.00 0.02 0.98 29.8 Panel I: Total number of transitions Panel III: Mean first passage time [0-1/16) [1/16-1/8) [1/8-1/4) [1/4-1/2) ≥1/2 [0-1/16) [1/16-1/8) [1/8-1/4) [1/4-1/2) ≥1/2 [0-1/16) 113 10 0 0 0 [0-1/16) 7.0 43.9 38.1 68.0 91.7 [1/16-1/8) 28 50 17 0 0 [1/16-1/8) 74.0 19.6 25.8 55.7 79.4 [1/8-1/4) 1 20 58 22 0 [1/8-1/4) 190.3 31.6 30.4 29.8 53.6 [1/4-1/2) 0 0 9 41 14 [1/4-1/2) 300.1 109.8 109.8 19.6 23.7 ≥1/2 0 0 0 1 65 ≥1/2 366.1 175.8 175.8 66.0 1.4 Panel II: Transition matrix p(i,j) [0-1/16) [1/16-1/8) [1/8-1/4) [1/4-1/2) ≥1/2 Ergodic distribution for transition matrix [0-1/16) 0.92 0.08 0.00 0.00 0.00 0.14 0.04 0.03 0.05 0.74 [1/16-1/8) 0.29 0.53 0.18 0.00 0.00 Eigenvalues for transition matrix [1/8-1/4) 0.01 0.20 0.57 0.22 0.00 0.30 0.55 0.81 0.98 1.00 [1/4-1/2) 0.00 0.00 0.14 0.64 0.22 Half-life for transition matrix ≥1/2 0.00 0.00 0.00 0.02 0.98 43.8 33 REFERENCES Agénor, P-R and J. Aizenman, 2010, “Aid volatility and poverty traps,� Journal of Development Economics, 91, 1-7. Agénor, P-R and O. Canuto, 2012, “Middle-Income Growth Traps,� World Bank Policy Research Working Paper 6210, Washington D.C.: World Bank. Aiyar, S., R. Duval, D. Puy, Y. Wu and L. Zhang, 2013, “Growth Slowdowns and the Middle- Income Trap,� IMF Working Paper WP/13/71. Washington D.C.: International Monetary Fund. Antman, F., and D. McKenzie, 2007, “Poverty traps and Nonlinear Income Dynamics with Measurement Error and Individual Heterogeneity,� Journal of Development Studies, 43- (6), 1057-1083. Antoci, A., M. Galeotti, and P. Russu, 2011, “Poverty trap and global indeterminancy in a growth model with ope-access natural resources,� Journal of Economic Theory, Vol. 146, 569-591. Azariadis, C. and A. Drazen, 1990, “Threshold externalities in economic development,� Quarterly Journal of Economics, 105, 501-26. Azariadis, C. and J. Stachurski, “Poverty Traps,� Chapter 5 in Aghion P. and S.N. Durlauf, eds., Handbook of Economic Growth, Volume 1A, Princeton: Princeton University Press. Banerjee, A. and E. Duflo, 2005, “Growth Theory through the Lens of Development Economics,� Chapter 7 in Aghion P. and S.N. Durlauf, eds., Handbook of Economic Growth, Volume 1A, Princeton: Princeton University Press. Bourguignon, F. and Morrison, C., 2002, “Inequality among world citizens: 1890-1992, American Economic Review 92(4), 727-744, September. Bureau of the Census, 1975, Historical Statistics of the United States: Colonial Times to 1970-- Bicentennial Edition, Part 1, Washington D.C.: U.S. Government Printing Office. Calderón, C. and L. Servén, 2004, “The effects of infrastructure development on growth and income distribution,� World Bank Policy Research Working Paper 3400. 34 Calvo, G., 1998, “Capital Flows and Capital-Market Crises: The Simple Economics of Sudden Stops,� Journal of Applied Economics, Vol.1, No.1, 35-54. November. Calvo, G. and C. Reinhart, 2002, “Fear of Floating,� Quarterly Journal of Economics, 107(2), 379-408. Capra, C. M, T. Tanaka, C.F. Camerer, L. Feiler, V. Sovero and C.N. Noussair, 2009, “The Impact of Simple Instiutions in Experimental Economies with Poverty Traps,� The Economic Journal, 119 (July), 977-1009. Chen, S. and M. Ravallion, 2012, “More Relatively-Poor People in a Less Absolutely-Poor World,� World Bank Policy Research Working Paper 6114, July. Commission on Growth and Development, 2008, The Growth Report: Strategies for Sustained Growth and Inclusive Development, published by the International Bank for Reconstruction and Development, Washington D.C.. Dollar, D. and A. Kraay, 2003, “Institutions, trade, and growth,� Journal of Monetary Economics, 50, 133-162. Eichengreen, B., D. Park, and K. Shin, 2011, “When Fast Growing Economies Slow Down: International Evidence and Implications for China,� NBER Working Paper 16919, March. Felipe, J., A. Abdon and U. Kumar, 2012, “Tracking the Middle-income Trap: What is It, Who Is in It, and Why?�, Levy Economics Institute of Bard College, Working Paper No. 715. Flaaen, A., E. Ghani and S. Mishra, 2013, “How to Avoid Middle Income Traps? Evidence from Malaysia,� World Bank Policy Research Working Paper Number 6427. Gill, I. and H. Kharas, 2007, An East Asian Renaissance: Ideas for Economic Growth, Washington D.C.: World Bank. Gradstein, M., 2008, “Institutional Traps and Economic Growth,� International Economic Review, 49(3), 1043-1066. Hamilton, J., 1994, Time Series Analysis, Princeton University Press. Hausmann, R., L. Pritchett, and D. Rodrik, 2004, “Growth Accelerations,� NBER Working Paper 10566. June. 35 Heckelman, J.C., S. Knack and F.H. Rogers, 2011, “Crossing the Threshold: An Analysis of IBRD Graduation Policy,� World Bank Policy Research Working Paper 5531, January. Jankowska, A., A. Nagengast and J.R. Perea, 2012, “The Product Space and the Middle-Income Trap: Comparing Asian and Latin American Experiences,� OECD Development Centre Working Paper No. 311, Paris: OECD. Jones, C. I., and P.M. Romer, 2010, “The New Kaldor Facts: Ideas, Institutions, Population, and Human Capital,� American Economic Journal: Macroeconomics 2010, 2:1, 224-245. Kharas, H. and H. Kohli, 2011, “What is the Middle Income Trap, Why do Countries Fall into It, and How Can It be Avoided?,� Global Journal of Emerging Market Economies, 3 (3), 281-289. Kraay, A. and C. Raddatz, 2007, “Poverty traps, aid, and growth,� Journal of Development Economics 82, 315-347. Kremer, M., A. Anatski and J. Stock, 2001, “Searching for Prosperity,� Carnegie-Rochester Conference Series on Public Policy, Vol. 55, pp. 275-303. Lin, J. Y. and D. Rosenblatt, 2012, “Shifting patterns of economic growth and rethinking development,� Journal of Economic Policy Reform, forthcoming in Volume 15(3), http://dx.doi.org/10.1080/17487870.2012.700056. Loayza, N. and L. Servén, eds., 2010, Business Regulation and Economic Performance, Washington D.C.: The World Bank. Maddison, A. , 2010, “Historical Statistics for the World Economy: 1-2009 AD,� Updated databases that originally was published in The World Economy: Historical Statistics, OECD Development Centre, Paris 2003. Available at Groningen Growth and Development Centre, www.ggdc.net. Mankiw, N. G., D. Romer and D.N. Weil, 1992, “A contribution to the empirics of economic growth,� The Quarterly Journal of Economics, 107(2), 407-437. Milanovic, B., 2005, Worlds Apart: Measuring International and Global Inequality, Princeton: Princeton University Press. 36 Nielsen, Lynge, 2011, “Classification of countries based on their level of development: How it is done and how it could be done,� IMF Working Paper WP/11/31, February. Parente S.L., and E.C. Prescott, 2005, “A Unified Theory of the Evolution of International Income Levels,� Chapter 21 in Aghion P. and S.N. Durlauf, eds., Handbook of Economic Growth, Volume 1A, Princeton: Princeton University Press. Pritchett, L., 1997, “Divergence, big time,� The Journal of Economic Perspectives 11(3), 3-17, Summer. Pritchett, L., 2000, “Understanding Patterns of Economics Growth: Searching for Hills among Plateaus, Mountains, and Plains,� World Bank Economic Review, 14(2), 221-50. Quah, D., 1993, “Empirical Cross-Section Dynamics in Economic Growth,� European Economic Review, Vol. 37, pp. 426-34. -------, 1996, “Convergence as distribution dynamics (with or without growth),� Centre for Economic Performance Discussion Paper No. 317, LSE. -------, 1997, “Empirics for Growth and Distribution: Stratification, Polarization, and Convergence Clubs,� Journal of Economic Growth, Vol. 2, pp. 27-59. Ravallion, M., 2008, “Poverty Lines,� in S. Durlauf and L. Blume, eds., The New Palgrave Dictionary of Economics, London: Palgrave Macmillan. Ravallion M. and S. Chen, 2011, “Weakly Relative Poverty,� Review of Economics and Statistics, 93 (4): 1251-1261. Reinhart, C., K. S. Rogoff and M. A. Savastano, 2003, “Debt Intolerance,� NBER Working Paper 9908, Cambridge: NBER, August. Shijin, L., Z. Junkuo, H. Yongxhi, and L. Peilin, 2012, “’Middle-Income Trap’ and ‘High- Income Wall’: Challenges and Opportunities to China,� China Economist, 7(1), 4-28. Woo, W.T., 2009, “Getting Malaysia Out of the Middle-Income Trap,� Mimeo, Davis: University of California Davis, August 13, 2009. Woo, W.T., 2011, “Understanding the Middle-Income Trap in Economic Development: The Case of Malaysia,� Presentation, The 2011 World Economy Asia lecture, University of Nottingham Globalization and Economic Policy Conference, January 13. 37 World Bank, 2012, World Development Indicators, World Bank: Washington D.C. World Bank and Development Research Center of the State Council (PRC), 2012, China 2030: Building a Modern, Harmonious and Creative High-Income Society, Washington D.C.: The World Bank. Xiaohe, M., 2012, “How China Can Avoid the ‘Middle-Income Trap’,� China Economist, 7(1), 98-105. Yusuf, S. and K. Nabeshima, 2009, “Can Malaysia Escape the Middle-Income Trap? A Strategy for Penang,� World Bank Policy Research Working Paper 4971. 38