InternationalDevelopmentAid Xavier Sala-i-Martin Columbia University June 2013
Heated Debate • Aid has some positive effects on growth (Jeffrey Sachs 2004) and needs to be multiplied. • Aid has effect on growth, only under some circumstances (conditional aid) • Conditional on Policies (Craig, Burnside and Dollar (2000), Dalgaard and Tarp (2004)) • Conditional on type: • Infrastructures [Clemens, Radelet, and Bhavnani (2004)] • Education [Michaelova and Weber (2006) and Dreher, Nunnenkamp and Thiele (2007)] • Health [Mishra and Newhouse (2007)] • Aid has NO effect on growth or may even undermine it (Peter Bauer (1972), Bill Easterly (2006))
Causality • Aid could systematically go to countries that are in trouble (like a natural disaster): if natural disasters tend to generate low (or negative) growth, this will tend to generate a negative association between growth and aid. • Aid could systematically go to “reward” countries that did things well in the past. If growth persists, then there will be a positive association even though aid does not really cause positive growth. • In order to solve this problem, econometricians use “instrumental variables”. IV estimates are supposed to see the correlation between exogenous aid and growth
Empirical Evidence • Early studies found positive correlations (Papeneck 1973, Levy 1988). • But then came Peter Boone (1994): Aid and growth are not correlated, period.
Empirical Evidence • Then a very Influential paper was written by Burnside and Dollar (2000): • They find α2 close to zero and α3>0. That is, AID has a positive effect on growth ONLY if the country at the receiving end conduct good policies. • After this paper was published, IFIs and the whole world demanded more international aid and conditionality on good policies.
Empirical Evidence • Problems with the paper: it is NOT robust to the definition of “aid”, “growth”, or “good policy” (Easterly, Levine and Roodman (2003)). • Roodman (2007, “The Anarchy of Numbers”) also adds that it is not robust to time period changes
Empirical Evidence • Definition of Aid: • Burnside and Dollar use “Grant Aid” (excluding subsidized loans and debt rescheduling). • Normal definition (called ODA) includes subsidized loans and debt rescheduling. • The two measures are highly correlated (0.933) • But when Easterly et all use this second measure, α3 becomes insignificantly different from zero.
Empirical Evidence • Definition of Good Policy: • Burnside and Dollar construct a measure which is an average of inflation, fiscal deficit and a measure of openness (originally proposed by Sachs and Warner 1995) • Easterly et al use TRADE/GDP instead of Sachs-Warner qualitative measure, they add “Black market premium” and “financial depth” (ratio of M2/GDP which is a measure of financial development) and… • … the coefficient α3 becomes insignificantly different from zero.
Empirical Evidence • Definition of growth • Burnside and Dollar use 4 year averages • Easterly et al criticize this because it contains business cycle noise. • If use 10-year averages… α3 becomes insignificantly different from zero • Roodman (2007) also shows that the paper is not robust to changes in the sample period of analysis.
Instrumental Variables • Burnside and Dollar (2000) use instruments that are based on “policy quality”. • The problem is that these variables may be correlated with aid (as they good policies attract more aid) but may also affect growth directly (so they are not good instruments) • Rajan and Subramanian (2005) criticize these instruments and use “colonial origin” variables and “language” variables as instruments (France and UK tend to give more aid to their colonies; if the fact that you have had one colonial power rather than another does not affect growth, then these are good instruments) • Problem: Shleifer et al (various papers) argue that the legal origin (partly inherited from colonial powers) DOES have an effect on growth • Rajan and Subramanian main result: holding constant a number of RHS variables, the IV correlation between aid and growth is zero.
Despite the evidence that Aid does not work (or may hurt) • Some people ask for more aid: Sachs, Gordon Brown, etc • How can they argue that more aid is necessary if the evidence is that aid does not work? • POVERTY TRAPS.
Poverty Traps: Theory • Start with Fundamental Equation of “Solow-Swan”: • Δk=sf(k) - (δ+n) k or • Δk/k=sf(k)/k - (δ+n) • If s and n are constant, and f(.) is neoclassical (concave with inada conditions), then UNIQUE AND STABLE STEADY STATE • Poverty Trap Theory: instead of unique and stable steady state, THREE STEADY STATES and Lower and Upper steady states stable and middle one unstable
Poverty Traps: Theory • Savings trap (savings rate is close to zero for poor countries for subsistence reasons and then shuts up as income increases) • Nonconvexity in the production function (there are increasing returns for some range of k)
Savings and Non-Convexities Traps Stable Stable δ+n Unstable s(k)f(k)/k
Poverty Traps: Theory • Demographic trap (impoverished families choose to have lots of children) - www.gapminder.org
Demographic Trap Stable Unstable s(k)f(k)/k Stable δ+n k
Poverty Traps: Implications • Small amount of aid does not work • Hence, the fact that aid has not worked in the past does not prove that it is ineffective. • However, if the total amount of aid is increased enough to put countries over the unstable steady state, countries will converge to the high-income steady state • NOTE: this is different from having two savings lines (if we have two savings lines with two steady states, then NO amount of aid will work!)
Problem 1: Savings Trap • Need to have THREE steady states: • For savings line to cross three times the depreciation line, you need the savings rate have to behave in “s” shape: • First low and constant (the savings line declines so it crosses de depreciation line from above and describes a stable steady state) • Then s should be raising for intermediate levels of k (so that the product s(k)*f(k)/k is upward sloping) • Then it should stay constant at a higher level (so that s(k)*f(k)/k becomes downward sloping again • In sum, it is NOT enough to argue that “poor people save less”. • There is NO evidence that saving rates accelerate sufficiently rapidly to justify the savings poverty trap (Kraay and Raddatz (2005))
Problem 2: Savings Trap • If there is technological progress, the savings trap automatically disappears!
Problem 3: Demographic Trap • If there is technological progress, the savings trap automatically disappears!
Problem 3: Fertility Behavior • True that fertility declines as income increases... but population growth is the sum of fertility, minus mortality, plus net migration • Mortality also declines with capital (and income) • And net migration increases with capital • Hence, need to argue that fertility declines MORE THAN OFFSET mortality declines, migration reversals and the diminishing returns to capital so that the savings and depreciation lines cross three times • This is empirically unlikely
Problem 4: Non-Convexities Trap • Normally, non-convexities can be easily convexified (for example, by using an average of the two technologies) • Thus, not only you need to argue that non-convexities exist, but need to argue that non-convexities cannot be “convexified” by averaging production from below the convex and above area • This is a lot harder
Problem 6: Evidence of Conditional Convergence suggests that “fundamentals” explain low income • Holding constant conditioning variables, the partial correlation between initial income and growth is negative • Again: To have poverty traps, we should have multiple steady states with same savings and depreciation lines (not that there are multiple savings lines). • If there are multiple savings lines, there is no reason to have increased aid
Forecasting the future of the WDI (by country) • Quah’s Methology: Based on historical experience • Пpp=probability of poor in 1960 staying poor in 2000 • П pr=probability of poor becoming rich • П rp=probability of rich becoming poor • П rr=probability of rich staying rich
Forecasting the future of the WDI (by country) • Npoor(2040)=Npoor(2000)* П pp+ Nrich(2000)* П rp • Nrich(2040)=Npoor(2000)* П pr+ Nrich(2000)* П rr • Repeat the procedure infinite many times to get the ergodic (steady-state) distribution • Conclusion: depends on Venezuela and Trinidad-Tobago
Problem • Not very robust (Kremer, Onatski and Stock show that it depends on one or two data points)
Interesting questions: • If it is “corruption” but we increase aid (we double in the next five years, and double it again five years later) because we think “poverty traps”, could we possibly induce more corruption? • Why doesn’t aid work?
The World of International Development Aid • The Many Players: • International Institutions (IMF, WB, United Nations, OECD,.. • Development Ministries of rich countries (USAID, Sweden, etc). • NGOs (non-profit organizations) • Left-Wing radicals (antiglobalization people) • Right-Wing radicals (including some churches) • Great Men and Women: Jeffrey Sachs, Kofi Annan, Desmond Tutu, Rigoberta Menchu, Subcomandante Marcos, the Pope, the Dalai Lama • Great Economists: Angelina Jolie, Bono, Tony Blair, Bob Geldof, Al Gore … • Well Intended people... But good intentions are NOT enough
Mechanisms that Work • Markets • Suppliers need to listen to customers • Responsibility/Accountability if don’t supply what’s wanted • Why? • Customer has something the supplier wants (money) Customers Information Firms Money Products
Mechanisms that Work • Liberal Democracy • Listen to “customers” • Responsibility/Accountability • Why? • Customer has something the supplier wants (votes) Voters Information Politicians Votes Policies
The Aid World Donors A strange sequence of Principal-Agent problems With misaligned incentives WB Bureaucrats African Bureaucrats African Citizens ? X
IDA • This means • We DON’T KNOW what works • ... and we don’t have incentives to LEARN! • We don’t have incentives to SATISFY CUSTOMERS (African citizens). • We have incentives to SATISFY DONORS (rich citizens and rich governments)! • Perverse outcomes!
IDA: Citizens of RICH World (Donors) • Donors have their own preferences (which may not coincide with true needs) • Sharon Stone and Malaria • Prostitution vs ARVs • Donors confuse Inputs and Outputs (because they are satisfied with SPENDING, not getting results) • Ten things you did not know about the World Bank • Donors some times don’t know what they are talking about • Ashraf, Gine, Karlan (2008)
IDA: The Local Intermediaries • Aid may lead to • Corruption (Natural Resource curse) • Marshall Plan was 2.5% of French and German GDP • Average African country receives more than 15% of GDP in Aid. • Misalocation of Talent • Culture of dependency and subsidy: Africa is stripped off its self initiative • When government revenue does not depend of economic success (as it is the case, for example, for countries with government that live of taxation)… government has less incentive to promote growth. • Government waste and patronage • Lack of interest in the right policies
IDA: NGOs • Donors are not accountable (unlike firms or politicians) • Oxfam and Cashew Nuts • Bill Gates and Primary care Doctors • Emmanuel Kuadzi • Donors only do things that are seen as “benevolent” • BUT Maybe the solution is investment, sacrifice, hard work... • Maybe Promotion of BUSINESS is the key!
Should we STOP IDA? • No: the debate should not be on whether to increase the amount of AID but HOW AID should be spent? • Learning at the MICRO level: because we know the GENERAL principles: • Markets are good • Rule of law is good • Education is good • Innovation is good • But how do you implement these in a particular country?
Correlation vs Causation • Example (1): African peasants. • There was a ebola epidemic. • Government sent doctors to the worst-affected areas. • Peasants observed that in areas with lots of doctors, there was lots of ebola. • Peasants concluded doctors were making things worse. • Based on this insight, they murdered the doctors.
Correlation vs. Causation • Example (2): SAT preparation courses in the US. • In 1988, Harvard interviewed its freshmen and found those who took SAT “coaching” courses scored 63 points lower than those who did not. • One dean concluded that the SAT courses were unhelpful and “the coaching industry is playing on parental anxiety.”
Correlation and Causation • Example (3): In Tanzania, a NPO provides extra teachers to the schools that want to participatein their program. The goal is to reduce the number of classes missed by students due to teacher absenteeism (a big problem in Africa). • One year after the intervention, they evaluate the grades of the students in the schools were the program was implemented are higher. • The NPO concludes that the problem is a success
All three examples • There is confusion of correlation and causation • Suffer from Sample Selection Bias: “treated” and “control” (or non-treated) groups were not selected randomly: • Ugandan Ebola: doctors were NOT assigned to random villages but to the worst-off communities; Hence there was a correlation between number of doctors and disease (reverse causation). • Harvard: Students who take prep courses are not random students but students that are more likely to do worse in SATs (that’s why they take the course!). (reverse causation) • NPO: schools that decided to participate were not random but more likely to have a responsible director or teachers (spurious correlation: good schools’ desire to improve teaching causes both the good grades and desire to participate in the program)
Randomized Field Experiments • The trial proceeds by taking a group of volunteers and randomly assigning them to either a “treatment” group (the group that gets the intervention), or a “control” group (a group that is denied the intervention). • Because it is random, the assignment of the intervention is not determined by anything about the subjects. • As a result, the treatment group is identical to the control group in every facet but one: the treatment group gets the intervention. • Hence, there is no BIAS (the two groups are not different in any consistent way)