Perhaps the most compelling question in all of economics is the breadth of global poverty. That people living in some nation-states are more prosperous than those in others has preoccupied economists since Adam Smith. After more than half of a century in which development economics has qualified as a formal division of economic theory the question is compelling as ever. The hundreds of millions of people who live beneath the already miserly World Bank standard of poverty – one U.S. dollar a day – testify to the urgency of trying to understand why transformational economic growth does and does not happen.1
Among the most compelling controversies with respect to promoting growth is the extent to which good economic policy can help. The 1990s were perhaps the high-water mark of the belief that policy was decisive, with many economists and political leaders coalescing around the Washington Consensus – the idea that market-oriented economic policies such as openness to foreign trade and investment, lean fiscal policies, and minimal government restrictions on pricing and resource movement promote growth. In more recent years, after the financial crises in developing countries of the last ten years, there has been some rethinking of that consensus. But while there is voluminous research on such particular questions as the best exchange-rate systems to prevent financial crises or whether to pursue expansionary and monetary policy after the occur, little is known in the broader sense about how economic policy can affect economic growth. The question is hardly idle, in that there is a growing theoretical and empirical literature that posits an extraordinarily high degree of non-economic determinism governing which nations prosper and which do not, in whose presence orthodox liberal policy is a poor response. It is the goal of this paper to use try to measure the extent to which, given other causes, economic policy can in fact affect economic growth, and if so how. It is in the spirit of Naude (2004), who attempts to isolate the ceteris-paribus effects of particular types of country features and policy on growth in Africa, and of Easterly (1993), who found that growth was considerably more unstable than country characteristics, including economic policy. The method also allows explanation of the extent to which recent years have been a time of reform and, in line with these papers but with a different method, of the effects of reform when it actually occurs.
Policy and its alternatives as contributors to growth One can think of modern development economics as a triangle of theories seeking to explain the prevalence and occasional overcoming of poverty. At the vertices of that triangle are the schools of thought emphasizing economic policy, institutional quality, and “endowments,” particularly biological and geographic ones. Much postwar thinking about development economics descends from the neoclassical growth model of Solow (1956). There is a well-behaved production function. Its technological parameters and the population growth rate are exogenous, and “growth” occurs through the accumulation of physical capital until a steady state is reached. This model motivated perhaps the most influential empirical paper, that of Barro (1991). His cross-country regressions confirmed two implications of the neoclassical model: that growth depends on physical capital accumulation (i.e. investment) and that per capita income at least conditional converged to the steady-state level, in that the rate of growth was negatively related to current per capita gross domestic product. And he modeled economic policy by transforming Solow’s steady-state per capita income into potential steady-state income: the maximum that could be had given the underlying production technology. This occurred because high levels of government spending or government-induced price distortions caused the resource base to be used suboptimally from the perspective of maximizing per capita income (although they might in principle achieve other desirable goals). This channel through which policy affects growth might then be called the Barro channel. The rise of the Washington Consensus represented a temporary triumph in the marketplace ideas of this channel.
At roughly the same time on the theoretical side, nonconvexities were introduced into the policy vertex, particularly via the productivity-enhancing role of knowledge and human capital (Lucas, 1988; Romer, 1986). In this literature societies that invest in activities that yield knowledge grow more rapidly than those that do not, other things equal. There is thus no particular reason to expect convergence in global standards of living, particularly if wealthier countries spend more on such activities. In addition to the Barro channel (which the knowledge models do not exclude), economic policy can have an independent effect, via the knowledge channel, by increasing or decreasing the ability to generate or make use of knowledge and the positive production externalities it generates.
The second vertex emphasizes institutional quality. The long-run view is most famously found in North (1990). In this school of thought institutions develop endogenously, but some of them turn out to be more growth-friendly than others. Institutional change, while difficult, is critical to growth. A very influential subset of institutional analysis focuses on corruption and rent-seeking. Here the emphasis is not on what to do but what to avoid doing. Excessive government entanglement with the economy breeds not just resource misallocation through the Barro channel but also increased effort devoted to redistributive rather than productive activity. Among the key theoretical papers are Tullock (1967), Krueger (1974) and Bhagwati (1982). The seminal empirical paper indicating that corruption, a close companion of the rent-seeking identified in this literature, is hostile to growth is Mauro (1995).
The final vertex emphasizes endowments. In this view, countries face certain geographical, biological and other constraints, which can decisively influence potential growth. For example, being landlocked can isolate the country from global trading networks, and laboring under a large malaria burden can destroy human capital. Nature deals the fundamental hand that countries must play. Sachs and Malaney (2002) argue that malaria in particular has a substantial negative effect on growth. Bleakley (2003) uses both macro- and microdata for the southern United States and finds that elimination of malaria in geographic areas yields higher education levels and that lack of exposure to it is associated with higher income. Other work by Gallup, Sachs and Mellinger (1999) argues for the larger importance of geography – distance from water, climate, and the prevalence of tropical diseases – in determining prosperity. Perhaps the longest-term view is that of Diamond (1997), who provides a model incorporating the geographic accidents of domesticated-animal distribution (and the immunity the presence of many animals who can be domesticate promotes), the ease with which inventions can spread to similar climates (a function of the extent to which migration can occur along an east-west rather than a north-south path), and other endowments far removed from economic policy as an explanation of why Europeans colonized the rest of the world rather than the reverse
The empirical evidence on these hypotheses is mixed. In dissent against the Gallup and Sachs (1998) view, Easterly and Levine (2003b) find empirically that whatever geographic effects exist work through institutions. This may occur because Europeans settled in areas with climates similar to their own, and in doing so brought their institutions with them (Hall and Jones, 1999). It may also occur because the nature of European settlement differed, depending on whether or not the local geography was favorable to extraction or settlement, with the latter environment more conducive to the imposition of favorable institutions (Acemoglu, Johnson and Robinson, 2002), or even because access to sea lanes promotes better institutions (Acemoglu, Johnson and Robinson, 2005. In this school of thought the rules of the game trump where you are as an explanation for modernization, although where you are may determine the rules you adopt.
But these all-or-nothing characterizations of the problem ignore the possibility of an interior solution. It would be surprising if the “reason” for poverty were at any vertex. Perhaps it is true that endowments and policy contribute to the level of economic performance. The analysis in this paper is not carried out with the intention of debunking one or the other as an influence on economic growth, but to measure how much policy can contribute, given other constraints. One argument that serves as a foil is that of Easterly (1993), who finds that “luck,” in the form of terms-of-trade shocks and world technological progress, eliminate most if not all of the detectable influence of policy.
Data and Basic Method To determine the effectiveness of policy reform it is necessary to distinguish between policy achievements, i.e. the extent to which policy has actually mirrored what the Washington Consensus recommended, and policy effects, i.e. the relation between the goals of the Consensus and economic growth. In this section the latter task is attempted, to se the stage for investigation of the former. The measurements of policy effects will use three cross-country growth regressions. The tactic is to classify several right-hand variables as policy-related, and to standardize for other (especially endowment) factors that also influence growth. I employ three data sources. One is the well-known Barro/Lee data set of economic data for five-year intervals from 1960-4 to 1990-4. The second is a set of geographic data compiled by Gallup, Mellinger and Sachs (1999).2 The third involves the presence of either internal or external military conflict, and is taken from the Correlates of War dataset, which covers thousands of such conflicts since the early 1800s. These data are descended from work by Singer and Small (1972).
a15 AVGSCHOOL (1) The framework is straightforward and not the only defensible empirical procedure, but it is widely used and serves to set the frame of reference for evaluating what policy can and cannot do. GROWTH is growth in real per capita GDP over the relevant interval. PUREGC is a measure of government consumption as a percentage of GDP, PINSTAB is political instability (a measure of the sum of assassinations and coups in the country), and PRIGHTS and CIVLIBS are the Freedom House measures of political rights (the ability to participate in the political system) and civil liberties (measuring, roughly, freedom of political action). These latter two variables are on a 1-7 inverse scale, so that a higher number indicates less freedom. TERMS is the changes in the country’s terms of trade, INV is investment as a percentage of GDP, INFLATION is its inflation rate, and OPEN is the Sachs/Warner (1995) measure of an economy’s openness to global economic forces. They all come from the Barro/Lee data. BMP, the black-market premium on the country’s currency does, is used as a measure of price and other government-imposed distortions in the economy. If one accepts the rent-seeking argument that corruption is a function of the number of things to be corrupt about, i.e. the amount of government special privileges and interventions in free exchange, it can proxy for the amount of at least the potential for corruption, as well as the inefficiency deriving from such interventions for more conventional static-inefficiency reasons.
From the geography data set, AIRDIST is the distance in kilometers to the closest major port. LANDLOCK is a dummy variable taking the value one if the country is landlocked, and TROPICAR is the percentage of the country’s land area located in the tropics. WAR measures various combinations of the number of what the Correlates of War dataset characterizes as external wars with other nation-states, external wars with non-state forces and internal wars among different military factions, some formally affiliated with the government and some not, occurring in the country. I choose not to distinguish between various types of warfare. Finally, PCGDP and AVGSCHOOL measure GDP and average schooling among the country’s residents at the start of the relevant regression interval.
Growth over the entire 1965-95 interval The results of the first regression specification, Model 1, are in Table 1. PCGROWTH is annual average growth in per capita GDP from 1965 to 1995. PUREGC, INF, BMP, INV, TOT and OPENSW are the averages of the Barro/Lee figures for these variables from 1965-9 to 1990-4. POLRIGHTS and CIVLIBS are analogous averages from the 1970-4 intervals (when the Freedom House ratings began) to 1990-4. WAR is the sum of dummy variables over the entire interval for each type of war. Its maximum theoretical value would be 18, if a country suffered from each of the three types of war during each of the six intervals. (The actual maximum value, seven, was shared by Cambodia and the Philippines.) AIRDIST, LANDLOCK and TROPICAR are simply as defined above. PCGDP is the 1960 value of per capita GDP in 1985 dollars, calculated using the Laspeyres index method. TYR65 is average years of schooling in 1965. It is thus assumed in Model 1 that the amount of human capital is something that sets potential output as an initial condition, unlike physical capital, which is assumed to be added as a factor as in the neoclassical growth model.
Consistent with that model, investment has a positive and significant effect on growth, as does initial average schooling. Initial per capita GDP also has a negative effect on growth, suggestive of neoclassical convergence. Greater political-participation rights positively affect growth, while, surprisingly, greater civil-liberties protection has a negative effect. With respect to the policy variables, three of the four have statistically significant effects, all in the directions found by Barro (1991). Government consumption beyond defense and education and the size of the black-market premium have a negative effect and openness has a positive effect. Only inflation among the policy variables is not significant at at least the ten-percent level. Among the geographic variables, only LANDLOCK is significant, with a negative sign consistent with the endowments literature. WAR is insignificant. (Several other specifications of the amount of war were tried, and in each case the results were the same.)
Growth by Five-year Interval An alternative specification is to interpret the dataset as a panel. Table 2 reports OLS and random-effects estimation for (1), which are Models 2 and 3. In this case the dependent variable is average growth in per capita income over a five-year period. AVGSCHOOL and PCGDP are values at the beginning of the interval. PINSTAB is the total value over the interval. PUREGC, INFLATION and INV are averages over the interval, as is DEM, which is the Barro/Lee 0-1 continuous index of “democracy.” It replaces the Freedom House measures because of their absence in the 1965-9 interval. WAR is a binary variable taking the value one if any type of war occurs in the interval.
There is not much in the results to distinguish Models 2 and 3. In the OLS estimation government consumption, initial real GDP, the black-market premium, terms of trade change, investment, openness, landlocked status, the percentage of land that is tropical, inflation and the war dummy are significant at at least the ten-percent level. In the random-effects estimation the differences are that inflation and landlocked status are not significant, while political instability and years of schooling are.
Measuring the effect of policy The goal of is to measure the effect of policy on economic growth after taking account of other variables which might also affect growth rates. If there are n policy measures, then one measure of the effect of policy is
, (2) where ai is the regression coefficient for that policy variable and bi is the value it takes. This provides an estimate of the net growth-friendliness of country policies.
One key task is to define what constitutes “policy.” I will identify four variables as potentially policy-related: PUREGC, OPENSW, INFLATION and BMP. They are elements of policy in the sense that their magnitude is under substantial if not total control of the political authorities. Another question is the role of schooling. In most societies, schooling is substantially a public function, and in the regressions here (as in much previous work) higher levels of schooling are associated with higher growth rates. However, “substantially” is not the same thing as “primary,” and in many societies the extent to which schooling achievement is the result of policy will vary with the schooling level (e.g., primary vs. tertiary). Primary and secondary education are often substantially publicly provided, while the extent to which tertiary education is a public function varies considerably across countries. It is clearly not possible to attribute all of the gains to schooling to government policy, nor is it possible to ignore the role of the latter. Schooling is a variable affected by the state, but its provision is not generally thought of as “policy” in the Washington Consensus sense. Note also that war and its absence, both civil and interstate, is the result of government policy broadly defined. But since it is not generally the result of economic policy per se, and since the data do not allow the attribution of a particular conflict to a particular decision by a particular government, it is not included as a policy variable. The effect of the political system – democracy, the extent of political-rights and civil-liberties protection – on growth is more direct, but that too is largely beyond the realm of economic policy.
Table 3 reports the effects of policy for all three models. In the first method, the figures represent the effects of policy measured over a thirty-year interval, where the ai are the average values over each of the six five-year intervals in the data set for the policy variables, and the bi are the estimated coefficients from Table 1. In the second and third methods the ai are calculated for each interval, with the coefficients from the second regression multiplied by the same data values as in the first, and what is reported is the average value between 1965-9 and 1990-4 for these variables. All ai are thus averages of five-year averages; the only difference is in the coefficients bi. Note that because many of the non-policy variables have a statistically significant effect on growth (and because the openness index is binary, and arbitrarily defined so that a one value indicates openness), the measures should be thought of as marginal effects, to be added on to whatever hand geography, terms of trade shocks, etc. have dealt.
The three results suggest that bad economic policy can subtract a fair amount from potential economic growth. The variation between the most and least growth-friendly countries with respect to policy is least in Model 1, and increases in Models 2 and then 3. Note also that only in Model 3 is inflation significant and thus included in the calculation of P. Overall roughly one country in six in the sample reports policies that handicap growth by at least two percent in per capita terms. Given that 2.8 percent growth is by itself the growth rate required to double the standard of living in 25 years, or roughly one generation, it is very believable that misbegotten economic policy explains much of what makes poor countries poor. Even accepting the fatalistic view of geographic determinism, there is still a role for policy to play.
Other implications The approach, in addition to providing an estimate of what policy can and cannot achieve, has several other uses. Among them are the ability to objectively identify and characterize economic reform, some implications for the effect of openness policies, and the unique position of Africa with respect to economic policy.
The objective reality of economic reform There is a growing literature to match the growing controversy over how impoverished countries with years of weak economic growth should try to raise their standard of living, and how wealthy countries can contribute. There is controversy in the literature (Burnside and Dollar, 2000, on the optimistic hand; Easterly and Levine, 2003a, on the other) over whether foreign aid in conjunction with good policy can promote growth.
One of the difficulties in resolving this and other questions about economic policy is finding a measure of it. This is particularly relevant to the controversy over the merits of radical versus gradual reform. For example, Arrow (2000) indicates that there are reasons to be concerned about both gradual reform carried out over several years and radical reform carried out across many policy dimensions in a very short period of time. Gradual reform is not credible, but radical has the potential to be so disruptive as to discredit reform or incur social instability. But how can radical and gradual reform be empirically distinguished? The technique here provides a means to do that. P is simply a measure of the net growth-friendliness of economic policy. A change from one interval to the next indicates reform. A sufficiently large change in the value of P is then considered to be radical reform. Policy could similarly become considerably less growth-friendly. The five-year nature of the data limits the precision of dating the onset of reform and raises the possibility that dramatic reform may overlap two intervals, but in general the procedure allows identification of truly substantial economic reform.
Table 4 contains all cases in which the value of P changes by at least 0.02 (i.e., the net effect on per capita GDP growth changes by at least two percentage points) from one interval to the next, using Model 2. There are twelve instances of each type. In the case of pro-growth policy changes, many of the episodes coincide with what are generally thought of as episodes of radical reform – e.g., Chile and Ghana after the Augusto Pinochet and Jerry Rawlings coups in 1973 and 1982, and Israel in the second half of the 1980s. Again the size of the effects is worth noting – in the Chilean case, over ten percentage points over two intervals. This again suggests that good or bad policy can have a substantial effect on growth, even if it is not the only effect. That Ghana could go from a disastrous change for the worse in 1980-4 to one of the biggest changes for the best in 1990-4 is suggestive of both how wildly economic policy can gyrate in developing countries and how autocratic leaders such as Jerry Rawlings can be tolerated despite their repression of political freedoms if economic policy improves enough.
Openness Shifting from a closed to an open economy is a special case in the analysis because of the binary nature of the independent variable. But based on the coefficients for OPENSW in Models 1-3, a complete shift in the openness variable is in the various models associated with a positive effect on growth ranging from roughly 0.9 to 1.6 percentage points. This is consistent with the findings of most but not all of the cross-country empirical growth literature. (The most prominent exception is Rodriguez and Rodrik (2001), who argue that most models claiming to find a positive association between openness and growth suffer from various specification errors.) The binary nature of the variable suggests that most of the observations containing a shift from zero to one represent a one-time, substantial change in trade policies. That such changes with respect to trade in particular are positively associated with growth is modest testimony in favor of radical reform.
Perhaps more interesting is the synergy between openness and geography. An implied subtext of much of the endowments literature is that to be landlocked and distant from major ports is to be put at a substantial disadvantage in terms of the ability to grow rapidly. Indeed in two of the three models LANDLOCK has a significant and negative coefficient. But in fact for such countries openness may be even more important. If LANDLOCK is interacted with OPEN, OPEN retains its significance in Model 1 while the interaction term is significant (p < 0.07) with a positive sign without appreciably changing the other results. In Models 2 and 3 the interaction term is not significant, although LANDLOCK is not significant in either case. (Details available upon request.) This provides admittedly incomplete evidence that it is perhaps for landlocked country that openness is most important. Their inability to directly access ocean trade routes with other countries makes it all the more imperative that such trade routes be open into the country via the land. If one accepts the premise that one of the key features of the last 150 years or so has been a sharp decline in transportation costs, the costs of being a landlocked country may decline as long as borders are kept open to goods, services, migration and investment from countries that are not landlocked. Openness is no guarantee, particularly if there are several national borders between the country and the ocean. The country would then require that there be openness in all the countries between it and the ocean. But it is certainly true that to be landlocked is not to be consigned unavoidably to penury, with Switzerland being the most obvious counterexample. The possible synergy between openness and unfavorable geographic endowments is an important avenue for further research.
Africa The disastrous performance of Africa in the postcolonial era is the subject of an extensive literature all by itself. Perhaps nowhere else does the endowments/institutions/policy controversy come more sharply into focus. One of the primary stylized facts that the endowments hypothesis is most often called upon to explain is the miserable situation of much of sub-Saharan Africa not just with respect to economic growth but corruption, ethnic conflict, warfare and a host of other variables. Diamond (1997) devotes his entire penultimate chapter to a thorough investigation of how Africa was handicapped by a lack of domesticable animals, a north-south geographic orientation that prevented (because of climate differences as populations move north or south) the migration of technological improvements in agriculture and implements, and the small portion of its land suitable for cultivation, and Easterly and Levine (1997) find that a different sort of endowment, ethnic diversity, determines bad policy.
But countries with unlucky geography can in principle still overcome this handicap through better institutions, better policy or both. Numerous countries with current or past malaria problems (e.g., Botswana, Thailand) or otherwise suffering from geographic handicaps (e.g., Chile) have made great economic strides through some combination of good institutions and policy. And so geography is not destiny. What is so striking about Africa is the prevalence of bad policy. The average value of P over 1965-1994 is -.0208 for all sub-Saharan countries (n =24), and -.0077 for all other countries (n = 64), an extraordinary difference. Twelve of the twenty nations with the worst figures for P in Model 1 in Table 3 are sub-Saharan. Africa is a geographic outlier, and may be an institutional outlier (Block, 2001), but it is also a policy outlier. The ability to document this effect strongly suggests that any successful turnarounds in Africa must have a strong policy component. It may be, given the broader results in this paper, that good policy is sufficient in many cases to fix some of what ails Africa, although that claim merits further investigation. Even if bad policy is casually after some other endowment effect, the analysis here allows emphasis on the ultimate problem to be solved.