Introduction

Despite the considerable amount of research devoted to economic growth and develop­ment, economists have not yet discovered how to make poor countries rich. As a result, poverty remains the common experience of billions.

The literature that we survey here contains the beginnings of an answer to this ques­tion. First, it is true that technology is the primary determinant of a country’s income. However, the most productive techniques will not always be adopted: There are self­reinforcing mechanisms, or “traps”, that act as barriers to adoption. Traps arise both from market failure and also from “institution failure”; that is, from traps within the set of institutions that govern economic interaction. Institutions - in which we include the state, legal systems, social norms, conventions and so on - are determined endogenously within the system, and may be the direct cause of poverty traps; or they may interact with market failure, leading to the perpetuation of an inefficient status quo.

There is no consensus on the view that we put forward. Some economists argue that the primary suspect for the unfortunate growth record of the least developed countries should be bad domestic policy. Sound governance and free market forces are held to be not only necessary but also sufficient to revive the poor economies, and to catalyze their convergence. Because good policy is available to all, there are no poverty traps.

The idea that good policy and the invisible hand are sufficient for growth is at least vacuously true, in the sense that an all-seeing and benevolent social planner who com­pletes the set of markets can succeed where developing country governments have failed.

In essence, this is the same story that the competitive neoclassical benchmark econ­omy tells us: Markets are complete, entry and exit is free, transaction costs are negligi­ble, and technology is convex at an efficient scale relative to the size of the market. As a result, the private and social returns to production and investment are equal. A complete set of “virtual prices” ensures that all projects with positive net social benefit are un­dertaken. Diminishing returns to the set of reproducible factor inputs implies that when capital is scarce the returns to investment will be high. The dynamic implications of this benchmark were summarized by Solow (1956), Cass (1965), and Koopmans (1965). Even for countries with different endowments, the main conclusion is convergence.

There are good reasons to expect this benchmark will have relevance in practice. The profit motive is a powerful force. Inefficient practices and incorrect beliefs will be pun­ished by lost income. Further, at least one impetus shaping the institutional environment in which the market functions is the desire to mitigate or correct perceived social prob­lems; and one of the most fundamental of all social problems is scarcity. Over time institutions have often adapted so as to relieve scarcity by addressing sources of market inefficiency.^[210]

In any case, the intuition gained from studying the neoclassical model has been highly influential in the formulation of development policy.

Yet the evidence on whether or not non-distortionary policies and diminishing returns to capital will soon carry the poor to opulence is mixed. Even relatively well governed countries have experienced little or no growth. For example, Mali rates as “free” in recent rankings by Freedom House. Although not untroubled by corruption, it scores well in measures of governance relative to real resources [Radlet (2004), Sachs et al. (2004)]. Yet Mali is still desperately poor. According to a 2001 UNDP report, 70% of the population lives on less than $1 per day. The infant mortality rate is 230 per 1000 births, and household final consumption expenditure is down 5% from 1980.

Mali is not an isolated case. In fact for all of Africa Sachs et al. (2004) argue that

With highly visible examples of profoundly poor governance, for example in Zim­babwe, and widespread war and violence, as in Angola, Democratic Republic of Congo, Liberia, Sierra Leone and Sudan, the impression of a continent-wide gov­ernance crisis is understandable. Yet it is wrong. Many parts of Africa are well governed, and yet remain mired in poverty. Governance is a problem, but Africa’s development challenges are much deeper.

There is a further, more subtle, problem with the “no poverty traps” argument. While the sufficiency of good policy and good governance for growth is still being debated, what can be said with certainty is that they are both elusive. The institutions that deter­mine governance and other aspects of market interaction are difficult to reform. Almost everyone agrees that corruption is bad for growth, and yet corruption remains perva­sive.

To put the problem more succinctly, the institutional framework in which market in­teraction takes place is not implemented “from above” [Hoff (2000)]. Rather it is deter­mined within the system. This includes the formal, legalistic aspects of the framework, but is particularly true for the informal aspects, such as social norms and conventions.

The above considerations lead us back to poverty traps. First, numerous deviations from the neoclassical benchmark generate market failure. Because of these failures, good technologies are not always adopted, and productive investments are not always undertaken. Inefficient equilibria exist. Second, institutions are not always simple choice variables for benevolent national planners. Bounded rationality, imperfect information, and costly transactions make institutions and other “rules of the game” critical to eco­nomic performance; and the equilibria for institutions may be inefficient.

Moreover, these inefficient equilibria have a bad habit of reinforcing themselves. Cor­rupt institutions can generate incentives which reward more corruption. Workers with imperfectly observed skills in an unskilled population may be treated as low skilled by firms, and hence have little incentive to invest large sums in education. Low demand discourages investment in increasing returns technology, which reduces productivity and reinforces low demand. That these inefficient outcomes are self-reinforcing is im­portant - were they not then presumably agents would soon make their way to a better equilibrium.

Potential departures from the competitive neoclassical benchmark which cause mar­ket failure are easy to imagine. One is increasing returns to scale, both internal and external.

The presence of fixed costs for a given technology is more troubling for the neo­classical benchmark in poor countries because there market scale is relatively small. If markets are small, then the neoclassical assumption that technologies are convex at an efficient scale may be violated. The same point is true for market scale and specializa­tion, in the sense that for poor countries a given increase in market scale may lead to considerably more opportunity to employ indirect production.^[211]

Another source of increasing returns follows from the fact that modern production techniques are knowledge-intensive. As Romer (1990) has emphasized, the creation of knowledge is associated with increasing returns for several reasons. First, knowledge is non-rival and only partially excludable. Romer’s key insight is that in the presence of productive non-rival inputs, the entire replication-based logical argument for constant returns to scale immediately breaks down. Thus, knowledge creation leads to positive technical externalities and increasing returns. Second, new knowledge tends in the ag­gregate to complement existing knowledge.

If scale economies, positive spillovers and other forms of increasing returns are important, then long run outcomes may not coincide with the predictions of the neoclas­sical benchmark. The essence of the problem is that when returns are increasing a rise in output lowers unit cost, either for the firm itself or for other firms in the industry. This sets in motion a chain of positive self-reinforcement. Lower unit cost encourages pro­duction, which further lowers unit cost, and so on.

Another deviation from the competitive neoclassical benchmark that we discuss at length is failure in credit and insurance markets. Markets for loans and insurance suf­fer more acutely than most from imperfections associated with a lack of complete and symmetric information, and with all the problems inherent in anonymous trading over time. Borrowers may default or try not to pay back loans. The insured may become lax in protecting their own possessions.

One result of these difficulties is that lenders usually require collateral from their borrowers. Collateral is one thing that the poor always lack. As a result, the poor are credit constrained. This can lead to an inefficient outcome which is self-reinforcing: Collateral is needed to borrow funds. Funds are needed to take advantage of economic opportunities - particularly those involving fixed costs. The ability to take advantage of opportunities determines income; and through income is determined the individual’s wealth, and hence their ability to provide collateral. Thus the poor lack access to credit markets, which is in turn the cause of their own poverty.

An important aspect of this story for us is that many modern sector occupations and production techniques have indivisibilities which are not present in subsistence farming, handicraft production or other traditional sector activities. Examples include projects requiring fixed costs, or those needing large investments in human capital such as edu­cation and training. The common thread is that through credit constraints the uptake of new technologies is inhibited.

With regards to insurance, it has been noted that - combined with limited access to credit - a lack of insurance is more problematic for the poor than the rich, because the poor cannot self-insure by using their own wealth. As a result, a poor person wishing to have a smooth consumption path may be forced to choose activities with low variance in returns, possibly at the cost of lower mean. Over time, lower mean income leads to more poverty.

Credit and insurance markets are not the only area of the economy where limited information matters. Nor is lack of information the only constraint on economic in­teraction: The world we seek to explain is populated with economic actors who are boundedly rational, not rational. The fact that people are neither all-knowing nor have unlimited mental capability is important to us for several other reasons.

One is that transactions become costly; and this problem is exacerbated as societies become larger and transactions more impersonal. Interaction with large societies re­quires more information about more people, which in turn requires more calculation and processing [North (1993, 1995)]. Second, if we concede that agents are bound­edly rational then we must distinguish between the objective world and each agent’s subjective interpretation of the world. These interpretations are formed on the basis of individual and local experience, of individual inference and deduction, and of the in- tergenerational transmission of knowledge, values and customs. The product of these inputs is a mental model or belief system which drives, shapes and governs individual action [Simon (1986), North (1993)].

These two implications of bounded rationality are important. The first (costly trans­actions) because when transactions are costly institutions matter. The second (local mental models and subjective beliefs) because these features of different countries and economies shape their institutions.

In this survey we emphasize two related aspects of institutions and their connection to poverty traps. The first is that institutions determine how well inefficiencies arising within the market are resolved. A typical example would be the efforts of economic and political institutions to solve coordination failure in a given activity resulting from some form of complementary externalities. The second is that institutions themselves can have inefficient equilibria. Moreover, institutions are path dependent. In the words of Paul A. David, they are the “carriers of history” [David (1994)].

Why are institutions characterized by multiple equilibria and path dependence? Al­though human history often shows a pattern of negotiation towards efficient institutions which mitigate the cost of transactions and overcome market failure, it is also true that institutions are created and perpetuated by those with political power. As North (1993, p. 3) has emphasized, “institutions are not necessarily or even usually created to be so­cially efficient; rather they, or at least the formal rules, are created to serve the interests of those with the bargaining power to create new rules”.

Moreover, the institutional framework is path dependent because those who currently hold power almost always have a stake in its perpetuation. Consider for example the current situation in Burundi, which has been mired in civil war since its first democrati­cally elected president was assassinated in 1993. The economic consequences have not been efficient. Market-based economic activity has collapsed along with income. Life expectancy has fallen from 54 years in 1992 to 41 in 2000. Householdfinalconsumption expenditure is down 35% from 1980. Nevertheless, the military elite have much to gain from continuation of the war. The law of the gun benefits those with most guns. Cur­few and identity checks provide opportunities for extortion. Military leaders continue to subvert a peace process that would lead to reform of the army.

Path dependence is strengthened by positive feedback mechanisms which reinforce existing institutions. For example, the importance of strong property rights for growth has been extensively documented. Yet Acemoglu, Johnson and Robinson (2005, this volume) document how in Europe during the Middle Ages monarchs consistently failed to ensure property rights for the general population. Instead they used arbitrary expro­priation to increase their wealth and the wealth of their allies. Increased wealth closed the circle of causation by reinforcing their own power. Engerman and Sokoloff (2005) discuss how initial inequality in some of Europe’s colonial possessions led to policies which hindered broad participation in market opportunities and strengthened the posi­tion of a small elite. Such policies tended to reinforce existing inequality (while acting as a break on economic growth).

Path dependence is also inherent in the way that informal norms form the founda­tions of community adherence to legal stipulations. While the legal framework can be changed almost instantaneously, social norms, conventions and other informal insti­tutions are invariably persistent (otherwise they could hardly be conventions). Often legislation is just the first step a ruling body must take when seeking to alter the de facto rules of the game.^[212]

Finally, bounded rationality can be a source of self-reinforcing inefficient outcomes independent of institutions. For example, even in an otherwise perfect market a lack of global knowledge can cause agents to choose an inefficient technology, which is then reinforced by herd effects.^[213] When there are market frictions or nonconvexities such outcomes may be exacerbated. For example, if technology is nonconvex then initial poor choices by boundedly rational agents can be locked in Arthur (1994).

In summary, the set of all self-reinforcing mechanisms which can potentially cause poverty is large. Even worse, the different mechanisms can interact, and reinforce one another. Increasing returns may cause investment complementarities and hence co­ordination failure, which is then perpetuated by pessimistic beliefs and conservative institutions. Rent-seeking and corruption may discourage investment in new technol­ogy, which lowers expected wages for skilled workers, decreasing education effort and hence the pool of skilled workers needed by firms investing in technology. The disaf­fected workers may turn to rent-seeking. Positive feedbacks reinforce other feedbacks. In these kinds of environments the relevance of the neoclassical benchmark seems ten­uous at best.

Our survey of poverty traps proceeds as follows. Section 2 reviews key development facts. Section 3 considers several basic models associated with persistent poverty, and their implications for dynamics and the data. Section 4 looks at the empirics of poverty traps. Our survey of microfoundations is in Sections 5-8. Section 9 concludes.

There are already a number of surveys on poverty traps, including two by the first author [Azariadis (1996, 2005)]. The surveys by Hoff (2000) and Matsuyama (1995, 1997) are excellent, as is Easterly (2001). See in addition the edited volumes by Bowles, Durlauf and Hoff (2005) and Mookherjee and Ray (2001). Parente and Prescott (2005) also focus on barriers to technology adoption as an explanation of cross-country varia­tion in income levels. In their analysis institutions are treated as exogenous.

2.