Comment on IMF and World Bank Structural Adjustment Programs and Poverty by William Easterly
NBER Conference on Management of Currency Crises
© Peterson Institute for International Economics
Bill Easterly's paper on the impact of IMF and World Bank structural adjustment programs on poverty should be troubling to most readers because he believes he has found evidence of an adverse effect of adjustment lending on the link between growth and poverty. However, serious questions can be raised about his methodology and the interpretation of his results.
This paper is about adjustment lending by the international financial institutions (IFI); it is not about structural adjustment lending by those institutions. The author inappropriately, in my view, uses results about the former to criticize the latter. Easterly states that critics may conclude "growth under structural adjustment programs is less pro-poor than in economies not under structural adjustment programs." On the other hand, supporters of such programs may conclude "contractions under structural adjustment hurt the poor less than contractions not under structural adjustment programs."
It is important to understand what he finds: (1) growth reduces poverty, (2) adjustment lending has "no direct effect on poverty reduction," and (3) adjustment lending "has a strong interaction effect with economic growth," which means, in his words, "the poor benefit less from expansions during a structural adjustment program than in expansions without an adjustment program, while they are at the same time hurt less by contractions." It is the third statement that is the focus of the paper.
It is more surprising that Easterly finds a positive statistical association between adjustment lending and dampening of the effect of growth on poverty during downturns than that he finds a positive statistical association between adjustment lending and dampening of the effect of growth on poverty during an expansion. My prior would have been that adjustment lending was associated with relatively bad times and the lower the growth rate in those bad times the larger the impact on poverty.
The issue of the impact of adjustment lending on poverty is important. For that reason we should hold research on this topic to a high standard. The basic problem with this paper is that Easterly does not succeed in establishing the reason why an absence of adjustment lending should produce a larger impact of growth on poverty. The reader is left with two alternative explanations: first, adjustment lending does not affect the very poor because they are more often part of the informal economy. This is interesting, but not very useful. Second, a distinction is made between homegrown reform programs and programs mandated from outside the country. This distinction is not well supported by the facts. IFIs do not require members to borrow from them. Members come to them precisely when their policies have failed or continue to fail. It is Easterly's apparent lack of appreciation of this distinction that contributes to his misunderstanding of the nature of the selection bias that undermines his statistical results.
What is a Structural Adjustment Lending Program?
A major problem with Easterly's paper is that it adopts a broad and misleading definition of structural adjustment lending by the IFIs. It is wrong to lump essentially all types of IMF lending together with various types of World Bank structural adjustment lending for purposes of examining the impact of IFI structural adjustment lending programs. IMF and World Bank lending to support structural adjustment in member economies differs from lending to support macroeconomic stabilization. The former involves an attempt to attack some of the deeper problems that have affected the performance of these economies. Easterly is careful to distinguish in his conclusions between structural adjustment lending and structural adjustment policies, but he is not careful in his statistical analysis to distinguish between adjustment lending and structural adjustment lending, and he uses the two terms interchangeably in the main body of his paper. This is unfortunate when much of the current debate is about the structural content of IFI lending.
Easterly's lending data go back to 1980 when the conditionality in most IMF programs was aimed simply at macroeconomic stabilization and had little to do with structural considerations. As is reported in the recent IMF review of structural conditionality in programs (International Monetary Fund 2001), less than one-fifths of the upper-credit-tranche stand-by and extended arrangements approved in 1985 and 1986 had any performance criteria related to structural measures. As a consequence in its recent review the IMF's data on its structural conditionality do not start until 1987; the big surge in such lending was after 1994. Goldstein (2001b) also documents this point.
It is true that by 1980 the IMF had its Extended Fund Facility (which dates back to 1974), but the Structural Adjustment Facility was not established until 1986. It is also true that the World Bank has made structural adjustment loans for many years, but most of the loans for a good part of the 1980s were merely disguised balance of payments loans. In other words, Easterly should have sorted out the true from the ersatz structural adjustment loans from the Bretton Woods institutions before he started his statistical analysis. It is not enough, in my view, just to split the sample at 1989.
Second, even in the case of lending by the IFIs that all would agree was focused primarily on achieving structural objectives, those objectives come in various shapes and sizes. For example, what would we expect to be the impact on poverty of a program directed at restructuring the financial system of a member country? We would expect very little impact one way or another, but other factors (omitted variables) meanwhile may produce Easterly's statistical correlations. What about loans directed at reducing pricing distortions or rationalizing tariff structures? We would expect more of an impact, and we should be interested in the sign, but we should question whether the impact would be statistically similar across countries.
Third, structural adjustment lending has many different objectives depending in large part on the circumstances of the members. Structural adjustment lending for a country in transition (Russia and Ukraine) differs from structural adjustment lending for an emerging market economy (Korea or Thailand). Structural adjustment lending in Africa today also differs from such lending (primarily) to Latin American countries in the late 1980s and early 1990s. It is misleading to expect that lending to countries in substantially different amounts and circumstances will be associated with similar effects in a broad statistical analysis. Moreover, structural adjustment programs differ in size and in their degree of emphasis on structural adjustment; a better variable to try to capture the impact of such programs would be the size of the program in (SDR or dollars) per capita rather than just the number of programs approved by the IMF and World Bank Executive Boards. Easterly reports that he tried to control for any interaction effect with the absolute size of loans, but it was insignificant while interaction effect with the number of loans remained significant. This is puzzling, including why there was no multicollinearity problem in the estimation. One is left wondering whether the statistical results are dominated by a large number of small loans to a large number of small countries.
Fourth, it is important to worry about the contemporaneous impact of adjustment lending on poverty and to try to design programs that at a minimum cushion those impacts, but any significant positive payoffs from this type of program one would expect would be felt with a considerable lag after the lending occurs. Adjustment lending programs are not known as sources of instant gratification. Structural adjustment, in particular, is a complex process that even when it is most effective normally has effects over time. Easterly compounds this problem by his use of "poverty spells" (pairs of substantially similar poverty surveys) of different lengths and treating them as identical dependent variables. Moreover, as Easterly notes, in many instances adjustment programs are of different lengths, efforts may not be sustained, and the lending programs may be suspended or cancelled. One might expect these differences to affect the results. It is disturbing that a researcher as serious and respected as Bill Easterly did not take the time to refine his data.
In summary, my major criticism of this paper is that it combines apples, oranges, grapes, tomatoes, pasta, potatoes, red meat, and raw fish. It is not too surprising that the result is not particularly appetizing. We should be surprised that there are any "statistically significant" results at all.
Conterfactual and Other Methodological Issues
In addition to Easterly's misleading categorization of IFI adjustment lending programs, his paper raises serious methodological issues, some familiar to students of this literature and some less so. The familiar issue is the potential bias in the selection of the countries that have and have not had any or much lending from the IFIs. What we have is a problem of the counterfactual or control group. It is unfortunate in this connection that the reader is not provided with a full cross-classification of so-called adjustment lending with the "poverty spells." She is told in passing that India and China are countries that have had few or no such loans and Russia and Ukraine have had a large number.
I commented above about the inappropriate inclusion of lending to the latter two countries in the sample, but how are we to think about China and India? Calling China's economic reform program "home grown" is a distortion of the facts. Over the past 20 years, China has undergone a great deal of structural adjustment, often under the close tutelage of the IFIs. Moreover, China received three adjustment loans from the IFIs during the 1980s. China had two stand-by arrangements with the IMF in which the Chinese authorities laid out in considerable detail their reform plans. India, on the other hand, has undergone very little structural adjustment, and many observers wring their hands about the Indian situation. India had a substantial IMF program in the early 1990s, and it is generally regarded as a success.
My basic point is that Easterly does not seriously address the statistical problem of selection bias; his use of instruments drawn from the foreign aid literature does not do the trick. The issue is not which countries may have had a political leg up to help to obtain IFI financial assistance, but the nature of their economic and financial circumstances that drove them to seek assistance from the IFIs. Building a convincing statistical counterfactual is a complex issue, but researchers must try harder to come to grips with it in studies of this nature.
Second, researchers know a lot about IFI programs, and it is troubling, for example, when Easterly lists a number of speculations about what the IMF or World Bank may "fear" about contractions, or what "may be" the nature of a change in taxation that was part of a program, or what "may" have been included with respect to social safety nets when these are facts that are known or knowable.
Third, macroeconomic conditions have a lot to do with what is going on in the background with adjustment programs, but the amount of macroeconomic analysis in this paper is limited to a look at the countercyclicality of IFI lending. No attempt is made there to control for economic circumstances (for example, external financial difficulties) in the actual statistical tests; no attempt is made to differentiate trend from cycle. On the other hand, Easterly was careful to split his sample between periods with contractions and periods with expansions to see if there is a statistical difference in his interaction term.
Fourth, it would be useful to know more about the 150 (more or less) datapoints. Although Easterly summarizes the data, the reader would like to know more in order to evaluate his results. He makes much of the distinction between "poverty spells" during which there are expansions and those in which there are contractions in income or consumption. (Disconcertingly, the two concepts are used interchangeably.) We are told that median consumption growth is zero, the mean is minus 1.1 percent, and the standard deviation is 11.1 percent, but that is not a lot of information. Figure 2 provides summary information on perverse poverty-growth outcomes (expansions associated with increases in poverty, or contractions associated with a decline in poverty) sorted on the basis of the level of inequality and adjustment lending. In the cell with high inequality and high adjustment lending, 27 percent of the observations are perverse. This strikes one as rather significant, but we are not told whether the perversity is evenly distributed between expansions and contractions.
Finally, Easterly tries to tease out of his dataset information on four periods of currency crisis: Indonesia, Mexico, Russia, and Thailand. He does not "make much" (his words) of the datapoints. Each case involved a contraction (negative growth) and an above average amount of adjustment lending on Easterly's crude definition. He notes, almost in passing, "the increases in poverty were fairly modest except for Indonesia." In fact, the ex post elasticities for the four cases range from minus 2.16 to plus 0.1, compared with the estimated average of minus 1.9. Moreover, the mean of these four observations is minus 1.1, three standard deviations below the mean in the total sample. It would appear that on average in these four cases that involved heavy doses of structural adjustment lending the so-called dampening effect of such lending, through whatever mechanism, was unusually pronounced.
This type of gross cross section analysis has provided policymakers with valuable insights in the past, when the work is carefully done. One can appreciate the challenge involved in enriching the data used for this type of analysis, but in this case, too much useful and relevant information has been discarded or ignored.
Results and Policy Implications
What implications for policy should we draw from this paper?
First, Easterly generates a statistical result that adjustment lending appears to dampen the effect of growth (expansion or contraction) on poverty without, in my view, establishing a convincing story or mechanism that might produce this effect. He acknowledges that doing so is crucial to his analysis, but the issue is not fully resolved by his paper. He finds that adjustment lending alters the cycle for some policy variables, but also finds no evidence that these alterations affect poverty. Without a convincing mechanism, one worries about correlation without causation.
Second, the author's concluding remarks on the issue of a mechanism focus on the informal sector and suggest that adjustment lending is irrelevant to poverty alleviation because the poor are largely found in the informal sector. This is a rather narrow view of both poverty and adjustment lending. It is one thing to think that adjustment in an economy, no matter how defined or supported, has a minimal direct and immediate impact on the informal sector and, therefore, on poverty, but that is not the same as irrelevant. In the longer run, the overall efficiency of the economy does matter because we expect that as a consequence of growth the poorest will move from the informal sector to the formal sector.
Third, it is useful to be reminded that even in cases of classical stabilization programs that are supported by IFI lending, attention should be paid to the impact on poverty. However, in some cases, a country has been living beyond its means, and the growth of and, sometimes, even the level of aggregate expenditure needs to be curbed or reduced to restore overall balance to the economy. If this occurs evenly across income classes, poverty will increase. The question for the policymaker is whether the distribution of expenditure can be twisted even as the overall level or growth is adjusted. It is a reasonable goal, and one that deserves attention, but the near-term objective in programs undertaken in the context of overall economic stabilization efforts is to try to avoid expenditure cuts that have a disproportionate impact on the poorest, even though, as Easterly concedes, these are "recent vintage" concerns, and maybe not the best grounds for criticizing IFI lending over the past 20 years.
Fourth, in structural adjustment programs, appropriately defined, attention to details is even more important. It should not be difficult to improve the efficiency of existing programs ostensibly intended to assist the poorest while reducing their overall cost because too many of such programs are not really directed at reducing poverty but rather at subsidizing the middle class if not the upper class. The challenge to do better is not always easy to meet, in particular at a time of crisis where the design phase of structural programs is compressed.
Nevertheless, we know what has to be done. Take, for example, the matter of subsidized petroleum product prices in oil-producing countries. From an overall efficiency standpoint, the cost of the subsidy involved is often outrageous. Nevertheless, political leaders are reluctant to reduce the subsidy substantially and appeal to resistance by the poor to justify their reluctance when the true political resistance comes from a broader and more politically active segment of the population. The objective should be to design programs to rationalize petroleum product prices and use some of the fiscal savings to address more directly and effectively the needs of the poor, e.g., at the extreme via direct income transfers.
Finally, do we conclude from Easterly's paper that there should be more or less structural adjustment lending by the IFIs? There is an active debate on this issue, especially within the IMF, as we know from Goldstein (2001a), though that debate focuses not so much on support for structural adjustment as on which IFI should take the lead. On the basis of this paper, we are justified in concluding that (a) the issue of the impact on poverty and the poor needs to be further researched and (b) the overall effectiveness of such lending needs to receive greater scrutiny.
The author states that for many of the countries in his sample "adjustment lending has been so continuous . . . it is hard to speak of it as purely a transitional phenomenon." Whether one is talking about macroeconomic adjustment or structural adjustment properly defined, prolonged access to the IFIs is a problem for the countries because they are falling further behind, for the IFIs themselves because they are failing in their missions, and for the system as a whole because support for rational policies and instruments is being undermined. This paper does not help to advance that worthy agenda. It certainly does not add much to the debate when Easterly rests part of his criticism of prolonged use on the fact that the median growth rate of income per person in a group of 36 heavy IFI borrowers over the period 1980-98 was zero because the median consumption growth per household in his overall sample of 64 countries was also zero.
Easterly argues that "structural adjustment policies" promote poverty reduction but implies that "structural adjustment lending" by the IFIs in support of those policies is counterproductive. This is a rather curious distinction to make. To make it on the basis of a contrast between so-called homegrown programs and programs mandated from the top (of the IFIs or the borrowing governments) just does not hold water. It displays an incomplete understanding of IFI lending programs over the past 20 years; countries face external financial or other deep-seated economic difficulties and turn, with varying degrees of success, to the IFIs for financial and policy assistance. Researchers will have to come to grips better with those realities in their statistical analyses if they are to deal adequately with the problem of selection bias in studies of this type.
Easterly is right that more careful and detailed research needs to be done on these important matters, but I do not believe that his aggregate results yet provide much of a useful guide for further research.
International Monetary Fund. 2001. "Structural Conditionality in Fund-Supported Programs," Washington, DC: International Monetary Fund, April 16.
Goldstein, Morris. 2001a. "An Evaluation of Proposals to Reform the International Financial Architecture." NBER Conference on "Management of Currency Crises," March 28-31, 2001.
Goldstein, Morris. 2001b. IMF Structural Conditionality: How Much is Too Much? Institute for International Economics Working Paper 01-4, Washington, DC: Institute for International Economics, April.