Convergence and Development Traps

Transcription

1 Convergence and Development Traps Jean-Claude Berthélemy TEAM, University Paris 1 Panthéon Sorbonne & CNRS berthele@univ-paris1.fr First draft, 7 January 2005 Paper prepared for the World Bank ABCDE-Senegal Conference, Dakar, January 27, 2005

2 Introduction The idea of underdevelopment traps, fundamentally linked to the notion of multiple equilibria, is not new. It emerged at the very early stages of the development economics literature, and is associated in particular with seminal contributions by Young 1928, Rosenstein-Rodan 1943 and Nurkse It has been revisited by growth analysts since the mid-1980s, following empirical contributions by Abramovitz 1986 and Baumol 1986, who have generally associated multiple equilibria with the notion of convergence clubs. This paper first reviews the available empirical research results that provide some evidence of the existence of convergence clubs. Several complementary pieces of evidence are available: the apparition of twin peaks in the international distribution of incomes; the observation of a quadratic relation between initial income and future growth, observed both on a cross-section and on a time-series basis; and the dependence of structural parameters of conditional convergence equations on the initial level of some of the conditioning variables. A consequence of multiple equilibria is that a poor country cannot grow out of poverty unless some policy initiative is taken to change initial conditions in such a way that this country could jump from its low, stable, initial equilibrium to another higher, equally stable, equilibrium. Hence, policies, and not only initial conditions, matter very much in the discussion of convergence clubs. It has taken some time before such new findings on convergence clubs be translated into practical policy recommendations. Such translation has taken recently the form of several big push proposals, e.g. by Sachs and others, 2004, in their work for the United Nations Millennium Project, by Collier 2004, in the context of the discussions initiated by the Blair Commission for Africa, and by Radermacher 2004, in the context of the Global Marshall Plan Initiative supported by the Club of Rome. This idea is also very much behind the British proposal of a new International Finance Facility, which advocates spending upfront much larger aid budgets than the current flows, to be financed by long term borrowings by donor countries, so that the 1

3 outcome of the same fiscal effort could be concentrated on a relatively short period of time, giving hopefully the necessary impetus to lift poor countries out of their underdevelopment trap. These ideas are intellectually appealing. The question, however, is whether they can lead to policy recommendations that make sense. This enquiry should start with an identification of factors that are responsible for the actual underdevelopment poverty trap in which poor countries are possibly locked. A first group of mechanisms may be related to factor accumulation in a broad sense: demography, savings behaviors, and human capital accumulation associated with education. A second group concerns more qualitative aspects of the economic growth process, such as financial development or diversification of the economy. Finally, a third group has to do with political institutions, e.g. corruption and conflicts, which may trap a nation in a destructive dynamics where poverty and poor institutions reinforce each other. Several of these mechanisms may be at work in different phases of the economic development process, so that we may have in fact multiple multiple equilibria. In presence of multiple equilibria, two broad complementary strategies may be conceived to lift a poor country out of its underdevelopment trap. The pure big push strategy consists in artificially increasing for a while the available income of such countries through large transfers of assistance, under the assumption that such transfers of resources would be sufficient to initiate a self-reinforcing economic growth process. This paper suggests that this strategy may very often prove to be ineffective. A second, possibly more effective, strategy consists in promoting economic reforms in such countries, which would increase for a while their rate of economic growth. This paper shows analytically that this strategy has better chances to be effective. This analytical discussion provides also some prediction on what would be the dynamics of the time series of a country that would jump upward from a stable equilibrium to another. In a nutshell, this jump should lead to observe multiple growth peaks, or in other words a growth cycle pattern. 2

4 Then, the most critical question is whether one may detect in the recent economic history any evidence of a successful escape from the underdevelopment trap, consistent with the previous analytical prediction. To answer this question, I use in this paper the Maddison s dataset, which provides, for a relatively large number of countries, complete series of GDP per capita measured in Purchasing Power Parity (PPP), stretching from 1950 to 2001 and sometimes to 2002 or I transform these series with a Hodrick-Prescott filter, in order to eliminate their short-run cyclical component. The observation of such transformed series suggests in a majority of cases the occurrence of growth peaks, or growth acceleration episodes. Such growth accelerations, however, cannot simply be interpreted as jumps from a stable low equilibrium to another stable higher equilibrium: there is no reason to believe that they are not simply the result of a standard convergence process, from a relatively low starting point, below a stable equilibrium, to this stable equilibrium. Nevertheless, in a few instances, the observed time series have some resemblance with the dynamics that would characterize a jump from a stable equilibrium to another; i.e. they are characterized by multiple growth peaks. The relevant policy question is then to ask why some countries, starting at similar levels of development, have apparently jumped out of their underdevelopment trap, and others have not. Obviously, this question is particularly relevant to discuss the relative economic performances over the past fifty years of African countries and other least developed economies on one hand and South-East Asian countries and other emerging economies on the other hand. Answering this question in a systematic way is not easy, given the dearth of comparative economic data for the period when such economies diverged, in the early 1960s. One plausible explanation is however suggested in this paper, which is related to education policies: while many South-East Asian economies have implemented early ambitious education policies, initially aiming at providing universal primary education, most African economies have lagged behind in this respect. This is not due principally to low education expenditure in the latter, but 3

5 rather to very different education strategies, leading to significantly poorer results as far as primary education is concerned. The paper is organized as follows: it starts with a review of stylized facts backing the convergence club hypothesis; secondly, it reviews possible explanations for multiple equilibria; thirdly, it provides an analytical discussion of conditions under which a country can jump out of its underdevelopment trap; fourthly it studies the dynamics of the per capita GDP of developing countries, in search of some evidence of an escape out of the underdevelopment trap; fifthly it discusses the different possible explanations of such successful take-offs; finally it provides some policy conclusions. Stylized Facts on Convergence Clubs In the empirical literature, the multiple equilibria assumption is very much related to the notion of convergence clubs, which suggests that, although no absolute convergence of economies toward a similar level of development is observable, some local convergence properties can be observed. This notion of convergence clubs has been extensively discussed by growth analysts since the mid-1980s, following empirical contributions by Abramovitz 1986 and Baumol Baumol defined convergence clubs principally by policy regimes. Specifically, he identified three converge clubs: OECD developed market economies, centrally planned economies, and a third club, less precisely defined, grouping middle income countries. Conversely, later contributions insisted more, in the tradition initiated by the early development economics literature, on a characterization of convergence clubs by initial conditions. A very simple but empirically powerful approach linking the definition of convergence clubs to initial conditions has been proposed by Chatterji This approach consists in observing, on a cross-section comparative basis, that the rate of growth of GDP per capita depends in a quadratic way, instead of linearly, on its initial level. Typically, in the growth equations estimated by Chatterji, the initial GDP par capita has a positive parameter and its 4

6 square a negative one, which implies that, at low levels of development, economic growth depends positively on the initial income, while the standard absolute convergence relation (a negative relation between the initial income and successive growth) holds at higher levels of development. The paper produced by Hausmann, Pritchett and Rodrik 2004, which describes growth acceleration processes, uncovers a similar non-monotonic concave relationship between the initial level of output per capita and economic growth, using a time-series approach rather than cross-section comparisons. Such results are also very much germane to those proposed, in comparative growth framework, by Thorbecke and Wan 2004, on East Asian emerging economies, or Berthélemy and Söderling 2001, on African countries. Another piece of evidence of the existence of convergence clubs has been provided by Quah 1997, who showed that although the international partial distribution function of incomes was unimodal in 1960, it was bimodal some forty years later. Therefore, countries have diverged, in at least two different groups, which may be viewed as convergence clubs. The notion of conditional convergence introduced by Barro 1991 runs counter the assumption of convergence clubs. According to Barro and his numerous followers, the observed divergence of economies could be explained by the fact that the convergence process depends on a number of independent conditioning variables. This would imply that, although economies converge to different steady states, their growth processes can be represented using the same model: instead of proper multiple equilibria, one would observe multiple variants of the same equilibrium, parameterized by the conditioning variables. However, this conclusion has been challenged by Durlauf and Johnson 1995, who have shown that a standard growth regression à la Barro had significantly different parameters in countries at low level of development and countries at higher level of development. This suggests that, notwithstanding its relevance, the notion of conditional convergence does not preclude the existence of convergence clubs. Such convergence clubs are, again, associated with different initial conditions, leading to different 5

7 growth regimes, and not merely to different growth rates. Durlauf and Johnson have successfully identified two potential threshold points, defined either by the initial level of education or by the initial level of income per capita. Each of these thresholds separates countries in two groups, defining convergence clubs. Hansen 2000 has provided a way to estimate a confidence interval for such thresholds. Berthélemy and Varoudakis 1996 have suggested that at least another factor could be responsible for multiple equilibria: the initial financial development. The Multiple Sources of Multiple Equilibria The possible existence of multiple equilibria in the growth process of an economy has been recognized very early in the theory of economic growth. The standard argument is that there are cumulative processes, leading to an economic decline when the economy is initially below a certain threshold of economic development, while economic progress is possible when this threshold has been passed. I provide in this section a summary of the principal arguments of this kind that have been considered in the previous literature. To simplify the analysis, I consider these various arguments in isolation from each other. In each of these arguments, the variable driving the development level of the economy plays the role of a state variable, z, and the law of motion of this variable has a particular shape leading to cumulative processes. This state variable can be conceived as a factor of production, such as physical or human capital, as a structural economic characteristic, such as financial depth or diversification, or as an institutional feature, such as the proclivity of the economy towards corruption or civil strife. (1) z=h(z & ) Let us define the law of motion of z as: If h(z) is a monotonically decreasing function, there is only one dynamically stable steady state, which defines a common development level towards which all economies should converge. If however it is not monotonic, there are possibly several steady states, as shown in Figure 1. 6

8 [Figure 1 here] In Figure 1, the bolded curve describes a situation where there are two stable steady states (z 2 and z 4 ). Note that in such a framework unstable equilibria (here z 1 and z 3 ) are inevitably alternating with stable equilibria. It should be noted also that, in order to obtain multiple equilibria properties, we need to assume that the h(z) curve crosses several times the horizontal axis, and not only that it is non-monotonic. Otherwise, the considered model would predict growth cycles, instead of multiple equilibria. This kind of outcome is depicted by the dashed curve in Figure 1. Let us now consider the different theoretical insights suggesting that such multiple equilibria may exist. The oldest ones are related with the analysis of the dynamics of accumulation of factors of production. Nelson 1956 has discussed a formal framework in which capital accumulation could be characterized by a cumulative process, due to the absence of savings capacity when incomes are very low. As a consequence of this assumption, at low levels of capital stock, savings and investment are not large enough to cover capital depletion and demographic growth, so that the growth rate of the capital/labor ratio of the economy declines when its initial level is below a certain threshold, while it is increasing immediately above this threshold. This argument is germane to the idea that there is an incompressible minimum level of consumption. In countries approaching this extreme absolute poverty level, the demographic growth rate may also be affected. This argument, initially introduced by Leibenstein 1954, has been also discussed by Solow 1956, again to suggest the possible existence of several steady state equilibria in his growth model. Another possibility of multiple equilibria associated with the process of capital accumulation may be related to a non-convexity of the production function linking output to capital. More precisely, in such analytical framework, one could observe increasing returns to capital at low levels of the capital stock, and therefore also a positive relation between the output 7

9 to capital ratio and hence the growth rate of the capital stock, assuming a constant marginal savings rate and the capital stock. This assumption, which has some similarity with the increasing returns to scale assumption made by Rosenstein-Rodan 1943, was also recognized by Solow 1956 as possibly creating multiple equilibria in his growth model. More recently, the possibility of multiple equilibria has been reconsidered in the course of discussions on the role of human capital, and particularly of education, in the development process. The theoretical model of Azariadis and Drazen 1990 clearly showed that a low level of educational development could lock an economy in a situation of underdevelopment. The paucity of human resources initially available considerably reduces the effectiveness of the education system and the return on education, and consequently obstructs the process of human capital accumulation, since the private return on human capital falls so low that parents do not invest in the education of their children. In this analysis, the education sector has a property similar to that attributed to the research and development sector in standard endogenous growth models, namely a dynamic externality. When the stock of knowledge available within the population is insufficient, the gains from this externality cannot materialize, and as a result growth is hampered unless the State implements a strongly proactive education policy. Following this argument, one could well observe dynamically stable situations where education is poorly developed, and where investing in education is not profitable. In developed economies, economic growth may be also associated to a large extent with productivity gains, related in particular to the results of the research and development activity, and not only with factor accumulation. This may again cerate non-standard growth patterns given the dynamic externalities characterizing the R&D sector. In a developing country, some of these productivity gains may be imported from more advance economies. But structural changes in the domestic economy may lead to productivity gains as well, with non trivial consequences on the dynamics of the economy. This line of argument has been explored by Berthélemy and Varoudakis 1996 regarding the financial deepening process. In a poor country, the initially weak 8

10 state of the financial system and the low income level of the population may persist and reinforce each other as a result of a cumulative process: low incomes imply that the amount of savings to be intermediated is small, which leads to high unit costs and weak competition in the financial sector; the result is a sluggish and inefficient capital accumulation process owing to both the insufficient size of the financial sector and its imperfect competition that prevents economic growth and lock the economy in a low-equilibrium state. Another structural change in the economy that may lead to multiple equilibria is related to the diversification process. Several contributions have suggested that the level of diversification of an economy could create a positive impulse on its growth process. Feenstra and others 1999, in a contribution to the analysis of the role of diversification in economic growth in South Korea and Taiwan (China), have suggested that output diversification could play in the growth process a role very similar to the input diversification considered in the seminal endogenous growth model of Romer Diversification may also reduce the vulnerability of an economy to external shocks; again, this may create favorable conditions to economic growth, if only because the economy can then invest in higher risk higher payoff projects. Several empirical papers have suggested that this positive link between diversification and growth was observable in a number of situations, beyond South Korea and Taiwan (China). 1 In parallel, there are some reasons to believe that economic diversification is influenced by the level of development attained by the economy. Imbs and Wacziarg 2003 have shown empirically that the diversification of an economy could be related to its development level, measured by GDP per capita, through an inverted-u shaped relation. Berthélemy 2004 has suggested that one possible explanation of this diversification dynamics could be related to the development of intra-industry trade: economic diversification could not be profitable in a standard neoclassical model where an economy should specialize on basis of its comparative advantage, but the development of intra-industry trade has shown that other sources of gain from 1 Berthélemy 2004 provides a brief review. 9

11 trade can be found. The theory of intra-industry trade that underlines these new sources of gain from trade can also be used to explain economic diversification. Given that intra-industry trade is the highest when trade partners have similar levels of development, this may explain why Imbs and Wacziarg observed a decline of diversification at very high levels of development, and hence their inverted U-Shaped relation between development and diversification. Whatever the theoretical reason for this inverted-u shaped relation, it may lead again to multiple equilibria, when combined with the arguments saying that a more diversified economy has a higher growth potential. Finally, the institutional framework may also be characterized by dynamic cumulative processes, which may lead an economy towards a poverty trap under some circumstances. This possibility is discussed with some detail by Collier Following his analysis, two kinds of mechanism are particularly relevant to explain observed political and economical cumulative crisis situations in developing countries, particularly in sub-saharan Africa: corruption and civil strife. Vicious circles of corruption have been discussed in a game-theoretical framework by Tirole In a context of high corruption, individuals have no incentive to invest in reputation, and therefore stay corrupt, while in low corruption equilibrium, a bad reputation has a high cost, and being corrupt is not profitable. A related phenomenon of pervasive rent-seeking, leading again to vicious circles, has been also discussed earlier by Krueger 1993, in the framework of political economy models. Political economy models can be used also, as suggested by Collier 2004, to explain situations in which civil strife and poverty reinforce each other, leading again to the possible existence of multiple equilibria. To summarize the foregoing discussion, there is potentially a large number of cumulative processes at work, in which the growth rate of a state variable characterizing the economy would decrease with its initial level, when this level is below a threshold, and would increase with it, at least for a while, when it is higher. Each of these mechanisms can lead either to the existence of multiple equilibria, or to growth cycles. Although I have considered them separately for the sake 10

12 of simplicity, these various mechanisms are mutually interdependent and should be combined to describe potentially multiple multiple equilibria and/or complex growth cycle paths. The Dynamics of the Jump Out of an Underdevelopment Trap Let us assume that the mechanisms possibly responsible for the existence of multiple equilibria that we have discussed in the previous section are theoretically relevant. The next question is whether we have any chance to observe in the real world a jump from one equilibrium to another. To discuss this, I start with a highly stylized analytical framework, in which I suppose that one or several cumulative processes are possibly at work. This multiple multiple equilibria structure cannot be fully described in a single-dimension framework. In theoretical terms, the economy can be described by a vector Z of state variables, which may be conceived for instance as several factors of production. The dynamics of this system is described by a differential equation system: (2) Z=H(Z & ) To simplify the framework, I ignore exogenous technical progress, so that the H function is time-invariant. Under the multiple multiple equilibria assumption, the condition H(Z)=0, which describes steady states, has n solutions, named Z 1 to Z n. (3) y=f(z) Assuming that GDP per capita is a function of Z: it follows immediately that there are at least n solutions to the equation (4) y& = 0 This multiplicity of solutions can be described in a graph similar to that of Figure 1, but where the z state variable is simply replaced by y. This graph hides however part of the complexity of the model. A combination of (1) and (2) leads to: = = (4) y& fz Z& f i Zi H( Z ) i i 11

13 Usually, this cannot be collapsed into an equation like: (5) y=g(y & ) because the growth rate of y does not depend only on y, but on the whole structure of Z. This is a standard aggregation problem. To assume a growth equation such as equation (5), it is necessary to ignore changes in structure of the Z vector, which is accordingly a strong assumption. For a given country, assuming that there are no exogenous shocks, only a portion of Figure 1 will be observed, describing a convergence path toward a stable equilibrium. For instance if the initial income per capita is between y 1 and y 2 it will converge toward y 2, and if it is between y 3 and y 4, it will converge toward y 4. Only exogenous shocks can lead to a durable shift from an initial convergence path towards a given stable equilibrium to a convergence path towards a different stable equilibrium, i.e. in Figure 1 from convergence towards y 2 to convergence towards y 4. Here we should focus our discussion on temporary shocks. A permanent shock, defined as an irreversible change in the curve depicted in Figure 1, would obviously lead to a change in the long term equilibrium. But the relevant question here is whether after a temporary shock a country could move from one long term equilibrium to another. Such exogenous shocks can be defined, in the framework depicted by Figure 1, as a combination of two kinds of shocks: A shock equivalent to a temporary transfer, modifying the initial income, or a temporary productivity shock, modifying its growth rate. A temporary positive transfer of income would have the same effect as a temporary leftward translation of the curve depicting g(y), with a reverse shift of equal magnitude to the right when the transfer is ended. Let us assume that at the time of the shock the considered economy is initially in the neighborhood of its steady state, at a point such as a 0 in Figure 2. Then, if the transfer τ is such as: (6) τ y 3 -y 2 12

14 this shock will have an adverse impact on the growth performances of the economy, which will have initially a negative growth rate, as shown in Figure 2: the economy moves to point a 1. This initial negative growth rate is simply due to the definition of y 2 as a stable equilibrium, implying that immediately to the right of y 2 the growth rate of the economy is negative. Later on, the growth rate of the economy increases, while staying negative for a while. If the transfer is discontinued when the economy has reached a 2, then the growth rate becomes again positive. In the end, the economy converges again towards y 2. In summary, although it temporarily increases the available income (defined by y+τ), a positive transfer has initially a negative impact on growth performances, and after the end of this temporary shock, the income of the economy converges again towards its initial long term equilibrium y 2 : this shock has only led to a circular movement of the economy around its initial equilibrium. [Figure 2 and Figure 3 here] However, if the transfer is sufficiently large, i.e. if: (7) τ>y 3 -y 2 then the growth rate will initially remain positive, and even will possibly increase, as illustrated by the shift from a 0 to a 1, and then to a 2, in Figure 3. However, if the transfer is only temporary, after a while it will be reversed. As depicted in Figure 3, there is still a positive probability that the income of the economy converge toward y 4. More precisely, a necessary condition to obtain eventually a convergence toward y 4 would be: (8) y 4 -y 3 >τ>y 3 -y 2 To obtain such a result, it would be necessary to assume that the periodicity of the cycle characterizing the function g(y) increase with y. If this periodicity is nearly constant, then the previous condition cannot be met, and the economy converges again toward y 2, as depicted in Figure 3. 13

15 To summarize the previous discussion, only a large transfer would possibly drive the economy from a path of convergence toward a low equilibrium to convergence toward a higher equilibrium. And even if the transfer is large enough to prevent initially that the growth rate become negative, there is no guarantee that after termination of the transfer the economy will continue converging toward the higher equilibrium. A temporary productivity shock would have potentially quite different consequences on the economy, as depicted in Figure 4. This kind of shock could be a total factor productivity gain, or could result from a growth-enhancing change in the structure of factors. If a positive productivity shock temporarily lifts the growth rate of the economy to high enough levels, so that the U-shaped portion of the g(y) curve remain above the horizontal axis, the economy will converge toward the higher equilibrium depicted by y 4 in Figure 4. To obtain such a result a necessary and sufficient condition is that when the productivity shock ends, the income of the economy has already reached a level above y 3. [Figure 4 here] At this stage, this discussion leads to three useful conclusions. First of all, an attempt to lift an economy out of a low equilibrium trap through a transfer policy is doomed to failure if the transfer that is granted is small even though this transfer may be given for a very long period of time. This seems to justify big push approaches, which call for large transfers. However, the probability that such a policy succeed does not depends only on the size of the initial transfer, but also on conditions on the shape of g(y), which are not trivial. To be fair with the big push proposals, it is true that a transfer can change the structure of factors Z, given that usually the factor accumulation that it permits is not spread evenly over all factors. This means that the g(y) curve may be also shifted upward thanks to the transfer policy, for instance if it is suitably designed to promote the accumulation of factors that are particularly productive because of their scarcity. But this argument points to the absolute necessity of avoiding that the discussions on such plans focus on the amount of transfer that is requested. 14

16 Second, an attempt to lift the economy out of its low equilibrium through reforms leading to a positive productivity shock, be it through a change in the factor mix or through a total factor productivity improvement, has better chances of succeeding, which suggests that discussing how such a structural change can be obtained should form the core of discussions on Marshall Plan proposals. Third, Figure 4 provides a stylized description of what would be the growth path of an economy that would jump from a low equilibrium to a higher equilibrium. Such a move would imply a non-monotonic evolution of the growth rate of the economy, with peaks and troughs before the economy stabilizes on its new convergence path. In other words, the dynamics of the economy would be characterized by growth cycles featuring multiple peaks. Stylized Facts on Multiple Growth Peaks Up to now, most of the empirical literature on convergence clubs literature has concentrated on cross-country growth comparisons, aimed at identifying convergence clubs. This is useful, and this is a necessary step, but more is needed to draw policy relevant conclusions. The principal question that I want to address in this section is whether there is any available evidence of a jump of one or several economies from a low level equilibrium to a higher level equilibrium. In absence of such empirical evidence, policy advices that would advocate a sort of Marshall Plan strategy would lack empirical foundations. As shown in the previous section, if there are such phenomena, we should observe a rather unusual dynamics in the concerned economies. As illustrated in Figure 4, their growth patterns should be characterized by several successive growth peaks, and not just by a single growth acceleration episode. I provide here an analysis of the observed growth dynamics of a number of developing countries, using annual time series built by Maddison 2003, some of which have been updated in the database recently released by the Groningen Growth and Development Centre and the 15

17 Conference Board (henceforth GGDCCB). As compared with the Heston and Summers database that is more frequently utilized, this dataset provides series from 1950 to 2001 (and 2002 or 2003 for the GGDCCB update), while Heston and Summers series start only later in the 1950s or in 1960 in many instances, and end at best in Given that I am interested in observing jumps from an equilibrium to another, I need to compare growth performances of countries that were initially at similar levels of development, but that have diverged later on. For that matter, using as long time series as possible is necessary. Another reason to choose Maddison s data is that they are better documented, and based on a critical comparison of all possible data sources, including Heston and Summers data when no better source is available. All in all, I have here a dataset of 99 developing countries with complete PPP GDP per capita series available for the whole period or beyond: 49 in Africa, 29 in Asia and the Middle East and 21 in Latin America. I have excluded Eastern Europe and Central Asia transition economies, the evolution of which has totally changed only recently, after the end of communism, as well as Cuba, and a few African countries with very poor quality data. 2 In order to eliminate the short-run cyclical component of these time series, I have transformed them with an appropriate Hodrick-Prescott filter. 3 Observing the relation between the transformed GDP per capita series and their growth rates provides here the empirical basis for drawing the equivalent of the g(y) curves, and possibly for uncovering multiple growth peak patterns that would be associated with equilibrium jumps. Note that, given that the g(y) function 2 I do not report here results for countries that were member of the OECD prior to 1994, although their growth dynamics has in many instances properties that are similar to those reported for developing countries. 3 Usually, such transformation is applied to quarterly series, instead of annual series. Baxter and King 1999 and Ravn and Uhling 2002 have shown that in the context of annual series the smoothing parameter used in the Hodrick-Prescott filter must be adapted accordingly. As Following their approach, I have used a value of 10 for the smoothing parameter (to be compared with 600 recommended by Hodrick and Prescott for quarterly series). The series transformed with the Hodrick-Prescott filter is the natural logarithm of GDP per capita. 16

18 is presumably only an imperfect aggregation function, I prefer drawing a different function for each country, instead of panel data approach. In the case of a single growth peak, which may characterize the standard dynamic path of convergence to a stable steady state equilibrium, the observed g(y) curve has an inverted-u shape. Accordingly, I observe this shape in a number of cases, related in some instances with significant economic progress since An example, for the case of Paraguay, is provided in Figure 5. [Figure 5 here] In several other instances, the observed g(y) curve depicts a cyclical growth pattern instead of a single growth acceleration. An example, for the case of Malaysia, is provided in Figure 5. With a few exceptions, situations in which the economy has not achieved on average any progress since 1950 are associated with the absence of any significant growth peak. In most of such cases the economy is cyclically stagnating around its initial level, as in the case of Senegal reported in Figure 5. Using a parametric method to detect single and multiple growth peaks would be presumably inappropriate, given the peculiar shape of a g(y) curve featuring multiple peaks. I propose here a simple non-parametric method, principally based on the observation of the curves drawn for the different countries. To be more precise, I define here a growth peak as an episode where: - growth is accelerating for at least 7 consecutive years; - growth is positive throughout the same period; - growth peaks at a pace of at least 3.5 percentage points. Given that I have eliminated the short-run cyclical component of the times series, they display few short term fluctuations, which leads to me to detect many cases of continuous growth acceleration phases for long periods of time. There are however a few instances where such 17

19 fluctuations have not been all eliminated, leading to frequents ups and downs instead of smooth acceleration and deceleration patterns. Such inconvenience is minor when as a result the count of peaks drops from one to zero, given that I am only interested in detecting multiple growth peak patterns. This is a drawback however in the cases of Hong Kong, Taiwan (China), Lesotho and the Seychelles, where an inspection of growth graphs suggests multiple growth peaks, which do not pass my test because of their relatively short duration (Figure 6). In the particular case of Hong Kong and Taiwan (China), there are so many peaks in the period that each of them could be only of a rather short duration. Fore these four countries, I have eventually concluded to the existence of multiple peaks, ignoring the intermediate small fluctuations between the start and the end of the growth peak. [Figure 6 here] A synthesis of the analysis of the growth dynamics of countries under review is provided in Tables 1 to 3, respectively for Africa, Asia and the Middle East, and Latin America. [Table 1 to 3 about here] In the first column, I report for each country the average growth rate calculated over the whole period for which data are available. In the second and third columns, I report the average per capita GDP for the 1950s and the 1990s decades respectively. In the fourth column, I report the number of peaks uncovered over the whole period of observation. In order to show that my results are not too sensitive to the choice of selection parameters, I report in the last column results of a classification based on a 4.0 percent, instead of 3.5 percent, growth peak threshold. Such results are quite similar to the previous ones. All in all, almost half of the countries (54 out of 99 countries) have experienced at least one growth peak since This observation is consistent with the findings of Hausmann, Pritchett and Rodrik 2004, who have detected many growth acceleration episodes in Heston and Summers time-series. However, as explained before, observing one single growth peak for a country is not enough to conclude that this country has possibly jumped from a convergence path 18

20 leading to a given equilibrium to a convergence path leading to a different, higher, equilibrium. As a matter of fact, many countries that have experienced single growth peaks have obtained on average since 1950 only modest growth performances. 70 percent of these countries (26 out of 37) have grown at a pace lower than the growth rate of the United States (2.1 percent a year), implying that their backwardness vis-à-vis the developed economies has deepened over time. Multiple peaks are detected for only 17 countries (15 when assuming a stricter selection parameter). As a result of their prolonged growth cycle patterns, these countries have usually performed much better than others. All of them have grown faster than the United States. Symmetrically, occurrences of high growth performances noticed over the past five decades may be associated with the observation of multiple growth peaks, insofar as they are much less frequent in the single or no-growth peak cases. However, there are some cases of countries having grown faster than the Unites States but without multiple growth peaks, which deserve careful analysis. There are firstly some countries that have been able to manage properly the oil booms or mineral booms: Oman and Saudi Arabia (and to some extent Egypt, Syria and Yemen) in the Middle East, Swaziland in Africa, and Trinidad and Tobago (and to some extent Mexico) in Latin America. Such countries have avoided a reversal of the growth process originating in the mineral/oil boom, principally because they have managed cautiously their windfall gains, and also in some cases because they have somewhat diversified their economy out of the mineral/oil sector. Secondly, there are relatively recent transition experiences: China, and to some extent India and Vietnam, where growth performances have been enhanced by the transition reforms, but where we have not a long enough observation period to observe a proper growth cycle pattern. Cape Verde, which implemented a central-planning economic policy after its Independence in 1975 until the end of the 1980s can be also considered as a transition case. 19

21 Thirdly there is Israel, which performed also rather well on average, but its performances have been too much affected, positively and negatively, by the various consequences of its conflicts with neighboring countries, to be interpretable in simple economic terms. Finally there is Costa Rica, which has grown slightly faster than the United States, and belongs to the single growth peak category, but which is in fact a borderline case: it experienced a second growth peak in the 1990s, culminating however at only 3.3 percent a year, instead of the 3.5 percent threshold that I have chosen. An objection to my interpretation could be that for some countries multiple growth peaks have been observed because these countries have been subject to multiple positive shocks. This might be true in some cases. But the main objective of my test was to eliminate some accelerating countries of the list of countries that may have experienced a jump from a low equilibrium to a higher equilibrium. In this sense, my test is rather powerful, since it eliminates many candidates (37 out of 54). Moreover, the fact that I observe only a limited number of multiple growth peak cases suggests that such countries have indeed experienced a quite peculiar dynamics. The multiplicity of shocks is common to all countries, although for most countries it has not led to multiple growth peaks but merely to alternating phases of progress and decline, leading in the long run to rather modest economic performances. To strengthen this interpretation, however, it is necessary to discuss the factors underlying the dynamics of multiple growth peaks countries, compared to other countries with similar initial development level. Only a comparison of factors possibly affecting growth potentials in these two groups of countries may provide more substantive arguments to back the interpretation of the occurrences of multiple growth peaks as jumps out of an underdevelopment trap. The next section is devoted to such an exercise. A Proposed Interpretation of the Divergence of Initially Equally Poor Countries 20

22 I focus here my attention on a subset of countries that suffered equally low levels of development in the 1950s, but which diverged: some of them have achieved very impressive progresses, others have undergone an equally impressive economic failure. In order to avoid comparing countries with different initial development levels, I consider here only countries where GDP per capita was on average below US$ 1,500 (in 1990 US$ PPP terms) in the 1950s. Out of the 17 multiple growth peak cases, this choice eliminates Brazil, Hong Kong, Malaysia, Mauritius, the Seychelles and Singapore, leaving 11 available observations of success stories. The role of Hong Kong and Singapore as business centers has certainly granted these two economies specific favorable conditions that made them quite special cases. Malaysia and Brazil started their development before WWII, although Malaysia was severely hit in the 1930s by the collapse of the rubber market, and Brazil was somewhat hit by the Great Depression and the consecutive worldwide protectionist policies. These economies were certainly in 1950 in a much better position to develop than poor African or Asian economies. Mauritius and the Seychelles, although less developed than these countries, have enjoyed early relatively favorable natural conditions. All in all, it is safe to exclude these economies from a comparison of success stories with poor and poorly performing countries. The list of countries with an initial per capita GDP below US$ 1,500, and which did not experience later on multiple growth peaks, is a relatively long one: it contains 40 African countries, 9 Asian countries, 2 in the Middle East and 2 in Latin America. Not all these countries have data available for a study of their growth potentials, but even after eliminating countries with poor data availability, this leaves plenty enough information to perform a comparison with the 11 selected multiple growth peak countries. 4 4 In the first group of 11 countries, data availability is better, although, given that most of my data are extracted from the World Bank s World Development Indicators online database, some data are missing for Taiwan. 21

23 Given that the initial growth peaks were generally observed, in multiple growth peak countries, in the 1960s or in the early 1970s, I discuss here data characterizing the growth potential of the two groups of countries at the beginning of the 1960s decade. I focus my analysis on measurable economic variables previously identified as factors potentially responsible for the occurrence of a low equilibrium trap. In the 1960s, there were of course institutional differences among poor countries, for instance between Asia and Africa, but the institutional framework of countries which took off in Asia was not considered at that time as particularly favorable to their development. The best known compendium of evidence of institutional drawbacks in Asia in the 1960s is Myrdal s trilogy on Asian Drama (Myrdal 1968). Myrdal s thesis was that the institutional characteristics of these countries constituted severe obstacles to their development. Accordingly, Myrdal wrote these three volumes on basis of his ten year experience in India, but he provided many empirical arguments in support of a generalization of his conclusions to Pakistan, Sri Lanka, Myanmar, Malaysia, Thailand, Indonesia and the Philippines. He spent for instance a full chapter to discuss the corruption issue in Asia and its adverse consequences, including it terms of civil strife: The significance of the corruption in Asia is highlighted by the fact that wherever a political regime has crumbled [ ] a major and often decisive cause has been the prevalence of official misconduct among politicians and administrators (Myrdal, p.937). Myrdal s prediction concerning Asia may have been mistaken in a number of ways but, as a matter of fact, his contribution has provided clear evidence that institutional issues such as corruption and non-democratic and non-transparent political governance, potentially leading to a poverty trap, were not confined to Africa. Quite the contrary, I would argue that many African countries enjoyed in the 1950s and the early 1960s a relatively stable political environment, and benefited from several policy initiatives aimed at promoting their economic development, be it in Francophone or in Anglophone Africa. Some evidence of the developmental attitude in African economies in the decade before Independence can be found in United Nations In former 22

24 French colonies, the France s Constitution of 1946 and its new political orientation toward colonial territories gave an impetus to their development; it led for instance to the creation of the Fonds d Investissement pour le Développement Economique et Social (FIDES), which played a major role by financing economic infrastructure building (Berthélemy 1980). Cote d Ivoire, for example, was on a quite fast growth track in the 1950s, in particular following the opening of the port of Abidjan. A similar trend was observable in British colonies, after the Colonial Development and Welfare Act was passed, in Although more historical material would be useful to fully support this view, looking for quantitative factors rather than for purely institutional factors responsible for the low equilibrium traps in which many poor countries, particularly in Africa, remained locked in the 1950s and the 1960s, seems therefore meaningful. Data on investment (Table 4) suggest that our two groups of countries were quite similar in the 1960s. Countries in the first group, with multiple growth peaks, had not impressive investment ratios until the end of the 1960s, and were comparable on this account to the other poor countries. The external financing of investment, approximated here by the difference between gross fixed capital formation and gross domestic savings, has been on average more favorable in the former group of countries than in the latter. This difference is, however, due only to South Korea, which received significant external assistance from the United States until the early 1970s, and Botswana, which had low savings but significant investments in the mining industry coming from South Africa. At any rate, there is no evidence in the available data that the high performances of the countries that took off in the 1960s was due initially to a jump out of a low savings trap: even though in such countries very high savings and investment rates occurred later on, this was not the case in the 1960s. [Table 4 about here] Demographic data (Table 5) suggest as well very similar initial conditions in the two groups of countries. In both groups, the population growth rate was high, around 2.5 percent, and the age dependency ratio was the same in both groups, around 88 percent. Therefore, one cannot 23