World Bank Conditional Loans and Private Investment in Recipient Countries

World Development Vol. 36, No. 10, pp. 1692–1708, 2008 Ó 2008 Elsevier Ltd. All rights reserved 0305-750X/$ - see front matter www.elsevier.com/locate/worlddev

doi:10.1016/j.worlddev.2007.10.012

World Bank Conditional Loans and Private Investment in Recipient Countries MARIAROSARIA AGOSTINO * University of Calabria, Arcavacata di Rende (CS), Italy Summary. — World Bank conditional loans might affect private investment in recipient countries not only through the funds they provide, but also via the policy conditions they include and the transfer of knowledge they imply. This work investigates the impact of these channels on private investment, considering also the particular effect of the formal commitment to reform, which necessarily comes along with conditionality. Taking into account the selection problem posed by participation in World Bank programs, the results indicate that backed commitments are associated with lower investment ratios in the short-run, and none of the other potential channels of influence seem to counterbalance this negative impact. Ó 2008 Elsevier Ltd. All rights reserved. Key words — World Bank, conditional loans, private investment, commitment to reform, non-random selection

1. INTRODUCTION World Bank and IMF conditional loans are designed to support policy and institutional reforms in developing countries. Therefore, the governments that sign this kind of loans have to commit themselves to implementing certain policy menus of reforms before they will actually receive the funds agreed to. The question of whether in the past two decades the conditional aid provided by these two institutions has fostered sustainable reforms, and promoted economic development is still controversial. In fact, policy-based lending presents a mixed record (see World Bank, 1998), and its real impact is difficult to assess for two main reasons. First, it is difficult to separate the economic effects of the loans from the effects of other observable and unobservable circumstances that brought the country to ask assistance. In addition, conditional loans may affect the economic performance of recipient countries through several channels, such as the amount of funds they provide, the policy conditions they include, and the transfer of knowledge and advice that they imply. This paper aims at investigating the effect of World Bank Structural (and Sector) Adjustment Loans on private investment, by distin-

guishing the channels aforementioned and by controlling for the fact that the countries receiving loans are a non-random sample of all possible countries. By doing so, this work combines the features of two recent strands of the literature on the subject: one, which emphasizes the importance of disentangling the impact of different channels through which IMF and World Bank programs may influence macroeconomic outcomes (see Bookmann & Dreher, 2003; Dreher, 2004), and another that advocates the use of appropriate econometric techniques in order to account for the potential selection bias implicit in participation

* This paper was partially written during my visit to the Department of Economics, University of Aarhus. Their hospitality is gratefully acknowledged. An earlier version was presented at the ‘‘Research on the World Bank’’ workshop, Central European University, Budapest, April 2nd 2005. Special thanks go to Jesper Bagger and the workshop participants for their discussion and useful suggestions. I am also indebted to three dedicated referees, whose precious observations reshaped the original manuscript. Finally, I am grateful to Damiano Silipo and Francesco Aiello for their comments. Any errors are solely my responsibility. Final revision accepted: October 3, 2007. 1692

WORLD BANK CONDITIONAL LOANS AND PRIVATE INVESTMENT

in structural programs (see Przeworski & Vreeland, 2000; Vreeland, 2003). Differently from most previous studies, which focus their analysis on the long-term impacts, this work adopts a short-run perspective. This time horizon is considered given the crucial role that the investment response can play not only in the long term for the economic growth, but also in the short term for the survival of the reforms themselves. Considering only the longterm investment response, by appealing to the difficult economic conditions characterizing new recipient countries, implies neglecting an important aspect: a short-run investment response is highly desirable, as it may make the adjustment effort socially more acceptable and increase the probability that the reforms will be maintained. Indeed, for a Structural Adjustment Program (SAP) to survive, it is extremely important that good effects are quickly visible, as negative effects tend to prevail: packages usually include harsh economic reforms involving budget cuts, privatization, and labor ‘‘flexibility,’’ which are likely to increase unemployment in the short run. Additionally, given the intense cooperation between the two institutions, frequently Bank loans have been coordinated with stabilization programs administered by the Fund. Hence, if a positive investment response materializes in the shortrun, this is likely to smooth the progress also of the Fund’s targets. Considering a short time horizon allows also the assessment of which of the mentioned channels exerts a prompt influence, and therefore might be considered to be key determinants of how long the program will endure. Focus is particularly on the potential short-term role of the formal commitment to reform, which necessarily comes along with conditionality and may represent a signal that the governments send to private investors in order to boost their investment response. To shed light on these issues, a sample of both recipient and non-recipient countries is employed, and the impact of SAPs on investment is assessed by using a methodology that accounts for the selection problem posed by participation in World Bank programs, by allowing the investment and the selection processes to be potentially correlated. In particular, the first process is modeled as an empirical investment equation for developing countries, while the second is modeled following the recent literature on the IMF and World Bank loans determinants (see, among others,

1693

Abouharb & Cingranelli, 2005; Knight & Santaella, 1997; Vreeland, 2003). The main finding of this paper is that backed commitments to reform appear associated with lower investment ratios, and none of the other potential channels of influence seem to counterbalance this negative impact. The remainder of the paper is organized as follows: the next section reviews the literature on the impact of adjustment programs on economic outcomes. After a brief description of the traditional approaches used to analyze the Bank programs effects, Section 2 focuses on the two strands of the literature that guide the empirical analysis. Section 3 describes the empirical question and the methodology employed. Furthermore, it illustrates the theoretical assumptions underlying the adoption of the estimated equations, with the definition and justification of each variable. Section 4 presents the data, describes the results obtained, and reports the robustness checks performed. Section 5 is a conclusion. 2. LITERATURE REVIEW: SAPS EFFECTS A large number of studies have analyzed the economic and social consequences of World Bank programs on recipient countries (for a detailed review of this literature see Abouharb & Cingranelli, 2005). Most of them concern single case studies or small-n comparisons, and adopt two main methodological approaches: the before–after and/or the with–without approach. The before–after approach compares macroeconomic variables before and after the implementation of SAPs (see e.g., Andriamananjara & Nash, 1997). Any differences discovered are attributed to the programs. An evident weakness of this method is that it does not distinguish between the influence of SAPs and that of other factors, which may affect macroeconomic outcomes over time. Indeed, the implicit assumption is that non-program determinants of macroeconomic performance are constant over time. The second approach compares adjusting and non-adjusting countries by employing matching pairs of countries or, more generally, a ‘‘treated’’ and a ‘‘control’’ (non-treated) group of countries (see, for instance, Harrigan & Mosley, 1991). In an ideal setting, for each adjusting country there should be at least another nonadjusting country experiencing exactly the same political and economic circumstances. As this

1694

WORLD DEVELOPMENT

cannot be the case in the real world, deciding on which countries can be considered similar enough (i.e., setting up the counterfactual) always involves some element of subjective judgment. Furthermore, even though some countries may show similar economic indicators, they will always differ because of unobservable characteristics, which may have led the government to ‘‘self-select’’ into the program. Some contributions to the literature on IMF lending effects represent large-n studies, and adopt appropriate methods in order to solve the selection problem posed by participation in the programs. Goldstein and Montiel (1986) and Khan (1990) represent earlier studies acknowledging the need to control for issues of selection when analyzing the impact of IMF programs. Khan (1990) refines and applies the Goldstein and Montiel’s (1986) generalized evaluation approach, which controls for the pre-program macroeconomic performance and the link between policy instruments and targets. Some years later, Haque and Khan (1998) still emphasize the need to take into account the endogenity of the decision to enter into a program in any evaluation exercise. More recently, Przeworski and Vreeland (2000) and Vreeland (2003) account for non-random selection into IMF programs, and find that IMF programs negatively affect yearly economic growth; moreover, they conclude that there is no evidence of a positive effect of the programs in the long run. In particular, Vreeland (2003) devotes a large part of his book to ‘‘the selection story,’’ illustrating the politics of IMF agreement by means of a game involving three players: the government, the Fund, and the veto player (i.e., the domestic opposition). On the demand side, the executive may approach the Fund not only when it needs foreign reserves, but also when it wants conditions to be imposed, in order to carry out reforms, which the veto player opposes. The existence of high rejection costs and low sovereignty costs is likely to weaken the power of the internal opposition. Rejection costs are defined in terms of the negative signal for creditors and investors that the rejection of an IMF program represents, while sovereignty costs are political costs, which the executive faces in approaching the IMF. Indeed, the opposition may always accuse the government of bowing to ‘‘the forces of international capitalism or selling out.’’ As regards the supply side, the author individuates three main factors affecting the attitude of the Fund in negotiation. First, the institution has to respond to its mandate, thus it will be more sensitive to the demand of countries with

large and potentially destabilizing balance of payment deficits. Moreover, its negotiation posture will be tougher when its budget constraint is more binding. Finally, the Fund will be reluctant to bear high negotiation costs, which democratic regimes could imply. Barro and Lee (2005) recognize that any study on IMF lending impact ‘‘has to sort out the directions of causation, that is, distinguish the economic effects of the loans from the effects of economic conditions on the probability and size of the programs.’’ With this aim, they first analyze the determination of IMF programs. As they find that some political and institutional variables have a significant explanatory power for loan approval, size, and participation rate, they employ them as instruments in order to disentangle the impact of IMF loans on growth and other macroeconomic indicators. 1 Their main finding is that participation in IMF programs lowers economic growth, negatively affects democracy and the rule of law, and does not have significant effect on investment and other economic variables. According to Edwards’s (2006) study, Fund agreements are associated with capital flights. Since the author corrects for non-random selection, he concludes that it is possible to ascribe such a negative impact to the ‘‘medicine’’ (the IMF program), rather than to the ‘‘disease’’ (the crisis bringing the governments to ask the Fund assistance). Another current in the recent literature highlights the importance of disentangling the impact of different channels through which IMF and World Bank programs may influence macroeconomic outcomes. In particular, following Bookmann and Dreher (2003) and Dreher’s (2004) arguments, conditional loans are expected to exert their influence not only through the funds they provide, but also via the policy conditions they include and the transfer of knowledge and advice that they imply. The amount of money that loans make available may have opposite effects. On the one hand, it should help countries to restructure their economies. Indeed, as Bird (2002) points out, a large amount of funds may allow countries to implement longer-run adjustment reforms. Besides, ‘‘other things being equal, a well-financed program will carry greater credibility. Advocates of reform will then be able to refer to the financial return to accepting conditionality’’ (Bird, 2002, p. 802). On the other hand, the amount of finance could bias government incentive to implement policy reforms, as it softens their budget constraints (Bookmann & Dreher, 2003).

WORLD BANK CONDITIONAL LOANS AND PRIVATE INVESTMENT

The policy conditions, which conditional loans entail, are meant to stabilize and/or liberalize the economy, and thus foster sustainable growth in recipient countries. Furthermore, conditionality may provide political cover for structural changes, which otherwise would not be carried out. Indeed, a government may desire conditions to be imposed, in order to push through its reform agenda (Vreeland, 2003). Conversely, conditionality may signal that government and donor preferences diverge, thus contaminating the credibility of a reform (see Collier, Guillaumont, Guillaumont, & Gunnning, 1997). It is worth noticing that conditionality might exert an influence on private investors even before any policy condition is implemented. As a matter of fact, the recipient countries policy makers have to express a formal commitment to reform, which may affect investment. On a theoretical level, two main hypotheses have been formulated on the link between backed commitments and investment response. According to the first one, because of the deal on loan conditionality, the private sector may expect a substantial improvement in future domestic policies. This ‘‘may raise the expected level of both social and private investment returns and hence crowd in additional private sector investment’’ (Gilbert, Powell, & Vines, 1999). To give a further example, Franc¸ois (1997) argues that negotiated external policy bindings have been employed by some developing countries in order to relieve investors concerns about policy uncertainty. In the alternative hypothesis, the commitment to policy reform, prompted by conditional aid, may jeopardize the credibility of the policy makers, and it is likely to impair supply responses (see, for instance, Killick, 1998; Rodrik, 1991). Collier et al. (1997) subsume both the hypotheses, maintaining that conditional aid may have several different rationales. Among others, it might serve as a signal for attracting investment, reducing the costs which private agents have to bear in order to obtain information about a government’s performance. The same authors point out, however, that the dominant rationale of conditional aid has been inducement, which signals that government and donor preferences diverge. This may limit the credibility of a reform, because if national policy makers reform in order to comply with some donor’s conditions, ‘‘private agents cannot tell whether the government is genuinely committed to liberalization, has already

1695

decided to take the aid donors for a ride, or has simply welcomed the money and deferred the agony of decision’’ (Collier & Gunnning, 1994). Finally, turning to the third channel of potential influence, the transfer of knowledge may support domestic policy makers in their plans of reform, and help developing countries to build the capacity to implement economic reforms. Bookmann and Dreher (2003) analyze the effect of IMF and World Bank loans on the economic freedom of a country. Their dependent variable is the indicator elaborated by Gwartney, Lawson, and Samida (2000), which summarizes several dimensions of economic freedom, such as the size of the government and the freedom to trade with foreigners. Their key explanatory variables (number of projects and amount of credit) are meant to encompass three broad channels of influence: conditionality, transfer of knowledge, and advice, and softening of the budget constraint. They find that the first two channels have a positive impact on economic freedom, while the third one has a negative impact, but only in the case of the World Bank. Dreher (2004) studies the relationship between IMF programs and economic growth, distinguishing the effects of disbursed funds, compliance with conditionality, and advice. According to this work evidence, despite the positive influence that the compliance seems to exert on growth, the overall impact of IMF programs is negative. Since the research so far considered has been primarily concerned with IMF programs, it appears relevant to provide evidence on the impact of World Bank loans. In particular, this study contributes to the literature investigating the impact on the private investment response, which is of paramount importance to accomplish the Bank mandate of poverty alleviation through the promotion of economic growth. The lessons drawn from the literature contributions mentioned above motivate the empirical question and methodology here adopted. Indeed, this work distinguishes different channels of influence and, at the same time, controls for the fact that the countries receiving loans are a non-random sample of all possible countries, that is, it controls for selection bias. Finally, differently from Bookmann and Dreher (2003) and Dreher (2004), which focus on long-term impacts, this paper focuses on some important aspects that a short-run perspective may bring up.

1696

WORLD DEVELOPMENT

3. EMPIRICAL QUESTION AND METHODOLOGY In the light of the arguments so far described, one can maintain that World Bank conditional loans might affect the investment response in recipient countries through the money, the policy conditions, and policy advice that they entail. The net effect of these factors will represent the long-run effect of SAPs on investment. How about the short-run? There is always a time lag between the approval of structural loans and their disbursement, which depends on the implementation of the promised reforms, and it probably takes even longer before the transfer of knowledge and advice affects domestic policy management. In contrast, even before these policy reforms are put into practice, the commitment to implement policy reforms, which each conditional loan implies, may immediately alter private expectations on future policies, and possibly impact on current investment decisions. Therefore, the formal commitment could exert a faster impact on investment than the other channels. On the other hand, following Vaubel’s (1983) argument, the availability of World Bank funds may also have short-run effects. Indeed, moral hazard phenomena might lead governments, which cope with a crisis, to maintain (or even worsen) the status quo in order to stay eligible for further subsidized support. Hence, macroeconomic performance in general, and aggregate investment in particular, could deteriorate even before the funds are disbursed. Considering a short-term horizon these hypotheses will be tested, shedding light on the sign and the relative importance of the different channels. (a) The model and the estimation method The investment equation adopted in this study has the following specification: X PFI it ¼ a þ X 0i;t1 b þ ht T t þ dSAP it t

þ cWBfundsi;t1 þ pNoPastSAPsi;t1 þ vit ; ð1Þ where PFI stands for the ratio of private fixed investments to GDP, the vector X includes some control variables, Tt is a comprehensive set of time fixed effects, and the subscript t refers to yearly observations. Following Rama (1993), the aforementioned control variables are intended to capture not only the conventional

accelerator effect and the cost of capital goods, but also some other factors that are likely to affect the investment decision process in less developed countries. The group of conventional determinants includes the real GDP growth and the real interest rate. As concerns the second group, the domestic credit to the private sector (as a share of GDP) is used to account for the so-called ‘‘overall tightness of credit markets’’ (credit rationing), given that in a less developed economy firms do not enjoy an unlimited supply of credit at a given interest rate. Further, the international reserves are employed to consider the possibility that foreign exchange shortage translates into a rationing for the demand for capital goods. Indeed, many developing countries have to import a considerable proportion of the machineries and equipment that they employ in their production. Moreover, the public investment variable is meant to capture the effect of inadequate infrastructure, which represents a severe obstacle firms face in developing countries. Finally, since the degree of economic instability may heavily affect the investment climate, a measure of volatility of GDP growth is included. This measure is based on the one-step ahead forecast errors of a random walk model and represents the unpredictability of demand. 2 All variables are lagged one year to limit simultaneity problems. Turning to the key explanatory variables of the analysis, the dummy SAP is used to test whether the fact of entering a structural program (therefore committing to reforms) has an impact on private investment. It is equal to one each year that a country enters a SAP, and zero otherwise. As Edwards (2006) summarizes: ‘‘Focusing on the first program year allows us to isolate other confounding factors (such as the degree of programme implementation) and focus purely on the effects of the signal sent by the announcement of a program. The WBfunds variable represents the amount of funds disbursed, measured as the ratio of World Bank loans to GDP. The other two channels of influence—policy conditions and transfer of advice—cannot be directly measured. In the light of Bookmann and Dreher’s (2003) argument, however, the number of arrangements concluded (NoPastSAPs) may be considered as a proxy for both transfer of know how and the effect of conditionality. 3 If the distribution of countries over the categories of committing and non-committing was random, under the usual regularity conditions, the OLS method could be employed to estimate

WORLD BANK CONDITIONAL LOANS AND PRIVATE INVESTMENT

(1) and test the significance of the dummy SAP and the other key variables mentioned above. In fact, countries could be selected (and/or self-select) into programs on the base of both observable and unobservable characteristics. Thus, the investment ratios of committing and non-committing countries could be different due to some different conditions the countries face and not to the backed commitment to reform. Formally, the selection into SAPs may be modeled as follows: X SAP it ¼ 1 if f  ¼ h0 þ Z 0it h1 þ ht T t þ git > 0; t

SAP it ¼ 0 otherwise; ð2Þ where f* is a latent variable representing the values of participation in a SAP, the vector Z includes observable determinants of a program (described in details in the next subsection), and the error term g capturing unobservable determinants of a program is assumed to be iid N(0, 1). If structural loans are not randomly assigned, the error terms v and g may be correlated: unobserved variables driving participation may also affect investment. If this is the case, the dummy variable SAP will be correlated to the error term v, leading to biased parameters estimates. Such a potential correlation between the ‘‘treatment’’ (captured by the dummy SAP) and the error v justifies the name of ‘‘endogenous treatment effects model’’ given to Eqns. (1) and (2). 4 In this work the average treatment effect d is estimated by applying two alternative approaches typically employed in the literature (see Vella & Verbeek, 1999): the Heckman (1979) method, which assumes all other regressors are exogenous and is based on the normality assumption, and an instrumental variable (IV) approach, which allows to control also for the potential endogeneity of other regressors included in the investment equation. As Vella and Verbeek (1999) summarize, the choice between these alternatives ‘‘is mainly a tradeoff between efficiency and robustness against nonnormality.’’ When using the Heckman two-step consistent estimator (1979), a probit is employed to model the selection process, under the assumptions that v  N(0,r); g  N(0,1), and corr(v, g) = q. The residuals of this estimation are used to construct a selection bias control factor (lambda), equivalent to the inverse Mills ratio, which is added to the substantial equation (1). 5 If its

1697

coefficient is significant, a selection bias exists. Once lambda is included, the coefficient associated to the dummy SAP expresses how large the difference in investment is between committing and non-committing countries, conditional on the other independent variables and the investment-related unmeasured characteristics, which are also related to the participation decision. With regard to the IV alternative, the fitted probabilities from the above-mentioned firststage probit model are used as instrument for the endogenous dummy SAP. 6 The Arellano and Bond (1991) estimator is adopted, which allows taking into account not only the ‘‘endogeneity’’ of the treatment effect, but also that of many investment determinants and the other World Bank variables. Moreover, this estimator accounts for the role of country-specific effects, and the inertia characterizing yearly data via the inclusion of a lagged dependent variable. 7 In conclusion, it can be noticed that the use of different methods not only allows being more confident about the main findings of the analysis, but also enables to compare the magnitude of the programs effects under different estimators (see Khan, 1990, p. 14). (b) Determinants of World Bank loans This subsection illustrates the theoretical assumptions that underlie the estimated selection equation. Before describing and justifying the variables employed, it is worth emphasizing that each structural loan represents the result of the negotiation process between two agents: the government on the demand side and the Bank on the supply side. Therefore, it appears appropriate to account for factors that potentially influence each party. Previous studies concerning the Bank lending practices are not numerous and do not systematically examine the determinants of its loans. Research is limited and focuses either on the demand side or on the influence of its major contributors, potentially affecting the supply side. 8 To the best of my knowledge, Abouharb and Cingranelli’s (2005) study represents the first attempt to scrutinize the selection biases implicit in any World Bank structural adjustment loan. 9 More attention has been paid to the IMF programs determinants. According to some recent studies, the approval of a loan is the joint outcome of both the country’s desire to enter a program and the Fund’s decision to approve one (see, for

1698

WORLD DEVELOPMENT

instance, Bird & Rowlands, 2001; Knight & Santaella, 1997; Przeworski & Vreeland, 2000; Vreeland, 2003). Hence, in order to control for both supply and demand determinants of a World Bank loan, this work integrates the (available) variables suggested by the Bank related literature with some other (available) indicators, which can be found in the IMF literature and may be regarded as potentially relevant also for the Bank. The resulting list of variables is described and sorted by Abouharb and Cingranelli’s (2005) categories: economic, political, conflict proneness, and human rights selection criteria (see Table 4 in Appendix A). (i) Economic selection criteria The GDP per capita, change in GDP per capita, and extent of international trade (imports plus exports to GDP) are considered to account for some pre-existing economic circumstances, which are likely to affect the probability of negotiating a loan. Since the Bank’s mandate is poverty alleviation, low-income conditions are expected to positively affect the supply of structural adjustment loans. Moreover, the Bank considers the free trade regime as a cornerstone for economic growth. Hence, it will likely intervene to foster trade liberalization relieving some important constraints to its development. According to the Dependency theory, however, international agencies would serve the interests of their major contributors (namely, US, Germany, and Japan), which would have established a core-periphery relationship with the developing world. Thus, in the dependency theorists’ view, the Bank would reward those countries more open to international trade, granting to the core multinational companies cheap raw materials and new markets for extra profits. As concerns the demand side, the government foreign reserves and total debt service variables are employed since countries tend to approach the Bank when they have a balance of payment need of financial resources. As the level of reserves is likely to shrink when the real exchange rate is overvalued, the annual official exchange rate is also included. Finally, a low investment level (gross fixed capital formation) may be thought as an indicator of low attractiveness for investors, due to ‘‘limited access to international capital markets, but also limited imports of capital and intermediate goods as well as distorted domestic credit markets’’ (Knight & Santaella, 1997). A country characterized by low attractiveness for investors is more likely to seek the Bank’s assistance. Przeworski and Vreeland

(2000) too expect a low ratio of gross investment to be associated with a higher probability of requiring an IMF loan. However, they justify their prediction by referring to the cost that the rejection of an IMF program represents for the opposition (the so-called rejection costs). The lower the investment the higher the country sensitiveness to the decisions of investors, hence, the higher the costs of sending a negative signal to investors by rejecting the Fund’s arrangement. The authors apply the same reasoning to the debt service variable. In this case, the higher the debt service, the higher the country sensitiveness to creditors’ decisions, the more likely the achievement of an agreement. (ii) Political selection criteria Most scholars agree on the relevance of politics in each stage of a SAP, from its negotiation to its implementation (see, for instance, Bird & Rowlands, 2001; Vreeland, 2003). Different theories, though, suggest alternative hypotheses on the expected impact of political considerations on the selection process. Following Putnam’s (1988) seminal work, the World Bank selection process might be biased in favor of authoritarian regimes. 10 Przeworski and Vreeland (2000) and Vreeland (2003) found that this is the case for the IMF, and they ascribed this result to possible higher negotiation costs in dealing with democracies. A competing theory, however, leads to the opposite expectation: democracies are likely to be preferred by international agencies. The accountability mechanisms and the higher availability of information characterizing any democratic country should make their commitments more credible (see Martin, 2000), and this in turn should encourage World Bank loan approval. Dollar and Svensson’s (2000) evidence appears to support Martin’s argument since, for a large sample of developing countries, they found that democratic regimes are more likely to implement Bank’s conditions, included in SAPs. To account for these opposite claims, the variable measuring the level of democracy is considered. Furthermore, the trade flows with US variable are employed to proxy for the potential US influence on the World Bank decisions. Indeed, according to a strand of the literature, the Bank would be heavily affected by its major shareholder (see, for instance, Bello, Cunningham, & Rau, 1994), and Fleck and Kilby (2001) found that the American economic interests have an impact on the geographic distribution of World Bank loans.

WORLD BANK CONDITIONAL LOANS AND PRIVATE INVESTMENT

Besides, the population variable is considered as a political determinant of World Bank loans for Abouharb and Cingranelli (2005) maintained that more populous countries are likely to be more influential and be favored by the Bank. These authors also argue that the post Cold War era should be characterized by a higher probability of receiving a loan from international financial institutions for all countries. Thus, a dummy variable is included, assuming the value of one for the post cold war period (from 1992 onwards), and zero otherwise. As far as the demand side is concerned, the budget balance variable is used to account for the fact that a country with a high deficit is more likely to approach the Bank, presumably because in need of fiscal discipline. Some governments may require a structural loan because they want conditions to be imposed. According to Vreeland (2003), ‘‘this gives them the political muscle to push through unpopular spending cuts.’’ Moreover, the number (of other countries) under a SAP, the number of previous loans and the election variables are meant to capture the potential influence of the sovereignty costs. These are political costs, which are expected to be higher before elections and to depend on the behavior of both past governments and other countries. When few other countries enter agreements or when past executives have never approached the Bank, this kind of costs may represent an important obstacle to participation (see Przeworski & Vreeland, 2000; Vreeland, 2003). (iii) Conflict proneness selection criteria As Abouharb and Cingranelli (2005) point out, in principle, any factor rendering the implementation of the loan conditions more uncertain is likely to make the Bank reluctant to negotiate a program. In particular, countries in conflict represent an uncertain investment. If there is domestic unrest, the incumbent could be replaced by a new executive, who is unwilling to respect previous agreements. A similar argument applies to interstate conflicts. Thus, the expected impact of the variables interstate conflict level and internal conflict level on the probability of receiving a loan is negative. (iv) Human rights selection criteria How the probability of entering a SAP is affected by the governments respect for the physical integrity rights of their citizens is controversial. According to the Dependency theory, the

1699

practice of imposing conditions through structural programs, especially those requiring a more flexible labor market, is meant to favor multinational corporations based in core countries. Since these conditions often involve harsh economic reforms, violent protests may arise and menace the stability of the government. In this view, the Bank would tend to reward those policy makers who repress riots and violate workers’ and personal integrity rights, in order to carry on with the policy reforms that the core dictates. Thus, dependency theorists would expect both the Bank to prefer countries that disrespected workers’ rights and that abused the physical integrity rights of their citizens. Other authors, however, maintain that the World Bank has always defended rights, which are considered essential components of economic growth (see for instance Nelson, 2000). To account for such opposite claims, the workers’ rights and the physical integrity rights variables are included in the selection equation. Finally, the number of years since previous loan variable and its square are employed as controls for temporal dependence, as the approval of a SAP today could affect the subsequent probability of receiving a SAP. 4. DATA AND RESULTS The present work employs the sources listed in Table 4 (Appendix A). The different data availability for different variables leads to an important loss of observations and this has to be taken into account when analyzing the results. The final sample comprises 403 annual observations of 41 developing countries. The panel is unbalanced, and spans the years 1982–99. The total number of agreements is 98, and Table 5, in Appendix A, shows their distribution across the sample countries. 11 Table 1 reports the Heckman model estimates, based on yearly observations. To attenuate potential endogeneity problems, most explanatory variables are lagged by one year. Moreover, country-specific effects are included (see Table 1 for further details). The hypothesis of selection bias appears to be supported by the evidence, as the lambda coefficient is statistically significant at 5% level. Moreover, its positive sign indicates that committing as compared to non-committing countries have unobservable characteristics positively related to private investment. In the substantial equation, all the control variables are significant, except for the international

1700

WORLD DEVELOPMENT

Table 1. Heckman model estimates: short-run impact of World Bank channels of influence on private fixed investment Substantial equation. Dependent variable: private fixed investment to GDP GDP Growth 0.317*** Real Interest Rate 0.021* Domestic Credit 0.048*** Public Fixed Investment 0.091 GDP Growth Uncertainty 0.320*** International Reserves 0.139 SAP 2.88*** World Bank funds 0.003 Number Previous Loans 0.255** Constant 9.633***

(0.0387) (0.0113) (0.0114) (0.0743) (0.0554) (0.1045) (0.8733) (0.0510) (0.1265) (3.0950)

Selection equation. Dependent variable: dummy SAP, coded one each year that a country commits to reform in the context of a SAP, and zero otherwise (0.0003) GDP per capita 0.0009*** Percentage change in GDP per capita 0.031 (0.0238) International trade 0.018 (0.0109) Polity 0.038 (0.0379) (4.9030) (log) Population 10.789** (1.2430) Post cold war 2.561** Interstate conflict level 0.369 (0.4007) (0.2705) Internal conflict level 0.580** (0.2141) Workers’ rights index 0.628*** Physical integrity rights index 0.096 (0.0828) (0.0999) Number of years since previous loan 0.245** (0.0093) Number of years since previous loan squared 0.032*** Trade with United States 0.174 (0.5261) International reserves 0.041 (0.0748) Exchange rate 0.0003 (0.0004) Debt service 0.004 (0.0094) Budget balance 0.054 (0.0407) (0.0323) Gross capital formation 0.061* Legislative election 0.293 (0.2307) (0.0449) Number of countries under SAPs 0.095** (0.0851) Number of previous conditional loans 0.294*** Constant 219 (99.7556) (0.5482) Lambda 1.340** Number of observations (countries) 403 (41) Note: country and time fixed effects included. All explanatory variables (except post cold war, number of years since previous loan and its square) are lagged by one year. Standard errors are reported in parentheses. (*), (**), and (***) denote statistical significance at the 10%, 5%, and 1% level, respectively.

reserve and the public fixed investment indicators. With regard to the selection equation, higher levels of internal conflicts, and larger number of other countries under a SAP positively affect the probability of receiving a World Bank conditional loan. In addition, the post Cold War era seems to be characterized by a higher probability of receiving a loan. By contrast, higher GDP per capita, larger populations, higher government respect for workers’ rights, larger number of previous loans, and higher gross capital formation decrease this probability. Finally, and consis-

tently with Abouharb and Cingranelli’s (2005) results, no evidence is found of bias against democracies, or in favor of allies with the United States, while evidence of non-linear temporal dependence is found. With regard to the dummy variable SAP, the coefficient is negative and strongly significant. The number of previous loans coefficient is also negative and significant (at 5% level), but its absolute value is much lower than that of the SAP dummy. 12 The Arellano and Bond (1991) results are reported in Table 2. The two tests for first

WORLD BANK CONDITIONAL LOANS AND PRIVATE INVESTMENT Table 2. Arellano–Bond dynamic panel-data estimates, one-step results: short-run impact of World Bank channels of influence on private fixed investment Lagged private fixed investment ratio GDP growth Real interest rate Domestic credit Public fixed investment GDP growth uncertainty International reserves SAP World Bank funds Number previous loans

0.619***

(0.0611)

0.251*** 0.016** 0.022 0.203** 0.186*** 0.100 0.886* 0.011 0.093

(0.0552) (0.0082) (0.0145) (0.0851) (0.0569) (0.1183) (0.4696) (0.0323) (0.1239)

F test 80.12*** Arellano–Bond test for AR(1) 3.23*** in first differences Arellano–Bond test for AR(2) 0.2 in first differences Hansen test 11.87 Number of observations 390 (40) (countries) Note: time fixed effects included. Standard errors, in parentheses, are heteroskedasticity consistent. (*), (**), and (***) denote statistical significance at the 10%, 5%, and 1% critical level, respectively. The dummy SAP is coded one each year that a country commits to reform in the context of a SAP, and zero otherwise.

and second order serial correlation, in the first difference residuals, support the assumption of white noise errors that underlies the estimator. Moreover, the Hansen–Sargan test of overidentifying restrictions is not significant, thus the joint null hypothesis that the instruments are valid, and that the excluded instruments are correctly excluded from the estimation cannot be rejected. 13 Focusing on the variables of interest, these results still indicate a significant (at the 10% level) effect of the formal commitment variable, while the other two World Bank regressors are now both positive and statistically insignificant. In addition, it is worth highlighting that the magnitude of the SAP coefficient is much lower than the corresponding Heckman model estimate. 14 (a) Robustness checks Thus far, the average treatment effect has been estimated by applying two alternative approaches typically employed in the literature (see Vella & Verbeek, 1999; Wooldridge, 2002,

1701

Table 3. Arellano–Bond dynamic panel-data estimates, one-step results: short-run impact of World Bank channels of influence on private fixed investment Lagged private fixed investment ratio GDP growth Real interest rate Domestic credit Public fixed investment GDP growth uncertainty International reserves SAP World Bank funds Number previous loans Lambda F test Arellano–Bond test for AR(1) in first differences Arellano–Bond test for AR(2) in first differences Hansen test Number of observations (countries)

0.556***

(0.0766)

0.257*** .019** 0.033* 0.174* 0.233*** 0.249* 1.550** 0.055 0.032 0.465 82.59 *** 3.14**

(0.0615) (0.0089) (0.0181) (0.0976) (0.0628) (0.1285) (0.7839) (0.0411) (0.1291) (0.4127)

0.37 10.07 368 (38)

Note: time fixed effects included. Standard errors, in parentheses, are heteroskedasticity consistent. (*), (**), and (***) denote statistical significance at the 10%, 5%, and 1% critical level, respectively. The dummy SAP is coded one each year that a country commits to reform in the context of a SAP, and zero otherwise. Lambda is a selection bias control factor retrieved from a probit model where the dependent variable is SAP and the explanatory variables are those described in Table 4.

p. 621). Yet, as an anonymous referee pointed out, it is also possible to use an instrumental variable method with the Heckman procedure. When the selection control factor (lambda) is added to the regressors set of the Arellano and Bond (1991) estimation, the dummy SAP coefficient is still negative and significant, while the other two World Bank variables coefficients are positive and not significant. It has to be mentioned that since the lambda coefficient is not significant at the conventional levels, the standard errors do not need to be corrected to account for the generated nature of the lambda regressor. Moreover, as Table 3 shows, the SAP coefficient exhibits an absolute value in between the two estimates reported in Tables 1 and 2. 15 To sum up, controlling for the potential endogeneity of other regressors in the investment equation (as is the case in Tables 2 and 3) produces more conservative estimates of the ‘‘treatment’’ (i.e., formal commitment) effect than that

1702

WORLD DEVELOPMENT

obtained when employing the simple Heckman model (Table 1). Furthermore, the analysis so far carried out is based on a treatment effects model, where the recipient status has merely an intercept effect on investment. The other control variables coefficients are supposed to be the same for committing and non-committing countries. Besides, no distinction is made between supply and demand determinants of SAP. To allows for different betas and accounts for the distinction mentioned, one can adopt a bivariate version of the Heckman’s sample selection model. Following this approach, first an Abowd and Farber (1982) bivariate probit model is estimated, including the variables reported in Table 4. Second, two investment equations are estimated, one for committing countries, the other for non-committing ones, including the selection bias control factors retrieved from the bivariate probit aforementioned. Making use of these estimates, the hypothetical investment ratios if selection were random are computed. In Vreeland’s (2003, p. 122) words, the expected ratios are obtained by counterfactually matching the country-years for all conditions—observable and unobservable—affecting aggregate investment. As further checks, the bivariate probit is estimated allowing for correlation between the error terms of the government and the Bank’s values of participation, that is, adopting the Poirier’s (1980) version of the model. Besides, a ‘‘stripped model’’ is considered, in order to maximize the number of observations available. Indeed, the availability of data may translate into another source of non-random selection since ‘‘it turns out that countries for which information is available differ in some systematic ways from those for which it is not’’ (Przeworski & Vreeland, 2000). 16 The large sample estimates make use of 928 country-year observations including 60 countries and 172 formal commitments, and are not emphasized given the possible misspecification of the investment equation. The finding that a formal commitment negatively impacts on investment holds across all the above-described checks. The relative estimates are not reported, but are available from the author upon request. 5. CONCLUSION The main finding of the present work is that formal commitments to reform prompted by

World Bank structural loans seem to be associated with lower investment ratios in the shortrun in recipient countries. This result is robust across different specifications, and alternative approaches used to estimate a treatment effect model, which is adopted to account for the potential selection bias implicit in the participation in structural programs. Such a harmful effect is likely to add to other short-run negative effects, which usually characterize an adjustment period. Therefore, the probability of program survival is bound to decrease. It might be argued, however, that a temporary slowdown in private investment could be rational during an adjustment period, when a high marginal return exists for postponing investment until the uncertainty is resolved. Furthermore, when the beginning of a structural program is announced, a positive investment response might not materialize because the reforms envisaged by the program are credible but centered on austerity. Therefore, the expectation of a period of austerity leads investors to hold back on their investment decisions, and wait for better demand conditions. In both the cases, once the economy is stabilized, investors will respond positively to SAPs. In other words, a possible interpretation of the negative impact detected here is that it takes some time before all the channels of World Bank influence (policy conditions, funds and transfer of know how) exert their influence, and possibly counterbalance the short-run negative effect. In fact, when a longer-run perspective is adopted, by considering all variables as five-year averages, none of the mentioned channels appear to be relevant, and affect the aggregate investment positively. By contrast, there is some evidence of a negative impact of World Bank funds. These results, yet, are not tabulated as they must be interpreted with caution for the limited size of the estimation sample, and further investigation is ongoing on the topic. Indeed, as the non-recipient countries define the counterfactual on the basis of which the analysis is carried out, a larger cross-country data set might either confirm or refute the mentioned findings. Besides, the present findings (both in the short and long run) refer to the average impact of World Bank SAPs, therefore they may hide a certain degree of unevenness across countries, and only a case-study approach would allow the drawing of more reliable conclusions about the single country experience. Bearing in mind these caveats, the evidence produced by this paper appears to support the

WORLD BANK CONDITIONAL LOANS AND PRIVATE INVESTMENT

widespread criticism of conditional lending, according to which conditionality sounds at odds with the governments’ ownership of the programs. In other words, policy conditions might be perceived of as a signal that government and donor preferences diverge, and therefore they might compromise the credibility of policy reforms, and impair the supply responses. In particular, according to Collier et al. (1997), a detailed specification of policy reforms and a short period for which contracts apply may exacerbate the problem of limited ownership. In the authors’ view, the rationale of this design is to ‘‘price’’ reforms individually, but the private sector cannot be confident about ‘‘a policy environment, which has been purchased by donors.’’ If the domestic policy makers have initiated some processes of reform in response to ‘‘a carrot or a stick,’’ the private agents may expect a return to the status quo once the carrot or the stick incentive is removed. The recent World Bank move toward a programmatic approach could represent an important step toward overcoming the shortcomings of such a short-leash

1703

conditionality. The Poverty Reduction Strategy Papers (PRSPs) introduced by the IMF and the World Bank in 1999 are prepared by governments with the contributions of domestic stakeholders and these international financial institutions. They illustrate the policies and programs that a government intends to carry out over several years to foster sustainable and equitable growth. They also specify the amount and the sources of financing that the country intends to employ. Their aim is to increase public participation in order to strengthen country ownership and commitment to reform. With the same intent adjustment lending has recently been replaced by development policy lending. This switch is considered as a mirror of a new policy course: the Bank appears determined to support sound reforms that are genuinely owned by the governments and their citizens. Future research on the effectiveness of this new approach is called for, in particular to assess whether the participatory process facilitates private investment, therefore enhancing the prospects for future development.

NOTES 1. More precisely, Barro and Lee (2005) see the Fund as a bureaucratic and political organization, which is greatly affected by its major shareholders (US, Japan, France, Germany, and UK). They expect, and find, that a loan is more likely for the more influential countries (as reflected by the size of a particular country’s quota and the number of the Fund’s employees who come from that country) and for those better connected to the above-mentioned shareholders (in terms of trade and voting patterns in the United Nations). 2. If one is willing to assume that economic agents develop forecasts of economic variables based on past rates (E[yi(t+1)] = yit since yit = yi(t1) + vit; t = 1, . . . , T), the forecast errors are of the form y i(t+1)  E[yi(t+1)] = yi(t+1)  yit. The (3 year) standard deviations of them represent the measure of uncertainty employed, which, as it is not a generated regressor, does not compromise the inference on the remaining variables (see Pagan, 1984). 3. Dreher (2004) uses some proxies to measure the compliance with IMF conditionality (the share of agreed

money disbursed; a dummy variable indicating substantial percentages of undrawn funds, and another dummy for program suspensions). Unfortunately I lack the data for the World Bank case. Incidentally, Dreher (2004) found that the compliance effect is small in magnitude and dependent on the proxy employed. 4. The terminology could generate some confusion between selection and endogeneity problems. The latter issue is more general and arises whenever the error terms are correlated with the explanatory variables (i.e., the orthogonality assumption is violated either for omitted variables, measurement errors or for simultaneity bias, see Wooldridge, 2002, p. 50). The usual way of addressing this problem, which results in inconsistent estimators, is the use of IV methods. As concerns the selection issue, it derives from the correlation between a non-random selection process and another relationship, which is of interest for the researcher. As illustrated in the body of the above section, such a link, in turn, implies a correlation between the error and an independent variable in the equation of substantial interest, hence non-random selection may be regarded as another source of endogeneity (see also Winship &

1704

WORLD DEVELOPMENT

Mare, 1992, p. 329). It is worth mentioning, yet, that selection biases may occur under two main forms. The standard one materializes when information on the dependent variable is missing for some individuals, which are a non-random sub-sample of all respondents. The second type, which applies to this study, is sometimes called heterogeneity bias. In this form, information is available for all individuals, but a nonrandom selection process splits the observations into two (or more) sub-samples (e.g., treated and nontreated individuals), interfering with the estimation of the relationship under study (the link between investment and SAPs in this work). In the latter case, as summarized by Vella and Verbeek (1999), the correlation between a regressor and the error term in the substantial equation may be addressed either by IV methods or by two-step procedures a` la Heckman (1979). Moreover, it is also possible to split the sample and estimate two separate substantial equations, so that the potential treatment effect does not show up as a dummy variable coefficient, but rather in the fact that the parameters are allowed to be different for the different sub-samples. This alternative model (called endogenous switching regime or split population model, see Millimet, 2001; Edwards, 2006) is adopted in the robustness checks subsection.

5.

Lambdai ¼

h1  f ½^ h0 þZ 0i ^ h 1F ½^ h0 þZ 0i ^

h1  f ½^ h0 þZ 0i ^ h F ½^ h0 þZ 0i ^

if

SAPi = 1;

Lambdai ¼

if SAPi = 0, where f is the probability density

function and F the cumulative distribution of the standard normal distribution.

6. We refer to Vella and Verbeek (1999) for the alternative ways of implementing the IV procedure. As Wooldridge (2002, p. 623) points out, using fitted probabilities as instrument for the endogenous treatment dummy presents several nice features, first ‘‘the usual 2SLS standard errors and test statistics are asymptotically valid. Second, (. . .) the IV estimator is asymptotically efficient.’’ Furthermore, it has an important robustness property: the binary response model does not have to be correctly specified.

7. This estimator consists of two steps: the data are first differenced in order to eliminate the unobserved individual effects, and then valid instrumental variables are employed in order to cope with the endogeneity problem. More precisely, Arellano and Bond (1991) propose a GMM procedure, exploiting the entire set of internal instruments, which the model generates, under the assumption of white noise errors. This assumption is verified by two tests for first and second order serial correlation, in the first difference residuals. Indeed, if the errors in level are characterized by the lack of serial

correlation, the error in differences is expected to display first order autocorrelation and to be uncorrelated at all other lags. 8. Ratha (2001) found that the request of a loan ‘‘is positively related to an increase in debt service payments and inversely related to a borrowing country’s level of reserves.’’ Fleck and Kilby (2001) found that US interests influence the geographic distribution of World Bank loans, even after controlling for country characteristics expected to affect the loans distribution according to the Bank’s stated apolitical allocation mechanisms. Dollar and Levin (2004) indicated that, in the last decade, the majority of donors have tended to favor countries with sound policies and institutions. According to this study, the World Bank IDA belongs to this type of donor.

9. Abouharb and Cingranelli (2005) considered the theoretical claims that help in understanding the selection biases of the Bank. Building on these theories and previous empirical studies (mostly concerning the IMF), they distinguish four categories of issues (economic, political, conflict proneness, and human rights issues), which affect the probability of receiving a SAP. By estimating a logit model, they show that economic need, large population, and government respect for workers’ rights positively affect the probability of receiving a loan. In contrast, higher levels of interstate and domestic conflicts decrease this probability. Finally, they find no evidence of bias against democracies, nor in favor of US allies.

10. Putnam (1988) described the interaction between a government and an international agency as a two-level negotiation game. The first-level players are the institution and the government, while the second-level players are the government and its citizens. The author argues that opposition could prevent any agreement with the agency, if the latter doubted about the government ability to implement the reforms, due to a strong internal opposition. Hence, Putnam maintains that the probability of signing an agreement at level I increases as the autonomy of the government increases at level II.

11. The estimation sample, however, is bound to change when using different estimators. The Arellano and Bond (1991) estimator, for instance, implies a loss of observations due to the first differing. Besides, since few countries sign agreements over most of the sample periods (about the 24%), one might have reservations about the predictive power of the selection equations here employed. In fact, following Bird and Rowlands’s

WORLD BANK CONDITIONAL LOANS AND PRIVATE INVESTMENT (2001) argument, a simple guess of no agreement might be accurate. Yet, as Edwards (2006, p. 36) acknowledged, the selection models used by the empirical literature do not ‘‘answer all our questions about how states initiate programmes, for there is still a great deal of unexplained variance,’’ and they are usually specified as the best ones given the available data.

12. Furthermore, as some selection variables are not significant, a more parsimonious probit specification is selected through a general-to-simple approach by dropping the most insignificant regressor at each stage, and ending up with a set of variables significant at least at the 5% level. When this ‘‘stripped’’ probit specification is used, the number of past SAPs variable is no longer significant, while the amount of funds becomes negatively significant, with a coefficient of 0.13. The dummy SAP is still negative and significant, and its absolute value increases from 2.85 to 6.59.

13. It is worth recalling that the treatment dummy is instrumented with the fitted probabilities retrieved from the selection equation estimates. Besides, all explanatory variables different from SAP are treated as endogenous, and instrumented using a subset of the available instruments. This is because, as Altonji and Segal (1994) pointed out, the use of all internal instruments implies small-sample downward bias of the coefficients and their standard errors. Moreover, if the number of observations is not very large, many instruments may imply little power of the Sargan test. This potential weakness of the test, yet, has to be acknowledged in the present case, as, despite the limitation of the number of

1705

instruments, the number of overidentifying restrictions appears quite high (300).

14. The same absolute value of about 0.8 is obtained when, in order to expand the estimation sample, the dummy SAP is instrumented with its first and second lags, instead of the fitted probabilities above-mentioned. This estimation, using 673 observations on 54 countries, confirms the non-significance of the other two World Bank variables, while the dummy SAP regains significance at the conventional levels. To economize on space, and also as they lead to the same conclusions, these results are not reported, and are available from the author on request.

15. Similar results are obtained when a FE-2SLS regression is estimated, with all the variables different from the treatment dummy instrumented with their first and second lags. To avoid cluttering, these estimates are not reported here but are available from the author upon request.

16. The stripped model is made of a simple investment equation, where accelerator effects are captured through two lags of the GDP growth, and a measure of its volatility accounts for the demand unpredictability. Moreover, the most parsimonious specification of the bivariate probit is selected through a general to simple search: the most insignificant regressor is dropped at each stage, until all the remaining variables are significant at least at the 5% level.

REFERENCES Abouharb, M. R., & Cingranelli, D. L. (2005). When the World Bank says yes: Determinants of structural adjustment lending. In G. Ranis, J. R. Vreeland, & S. Kosack (Eds.), Globalization and the nation state: The impact of the IMF and the World Bank. Routledge. Abowd, J. M., & Farber, H. S. (1982). Job queues and the union status of workers. Industrial and Labor Relations Review, 354–367. Altonji, J., & Segal, L. (1994). Small sample bias in GMM estimation of covariance structures. NBER Working Paper no. 156. Andriamananjara, S., & Nash, J. (1997). Have trade policy reforms led to greater openness in developing countries? Evidence from readily available trade data. World Bank Working Paper no. 1730.

Arellano, M., & Bond, S. (1991). Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Review of Economic Studies, 58, 277–297. Barro, R., & Lee, J. W. (2005). IMF programs: Who is chosen and what are the effects? Journal of Monetary Economics, 52(7), 1245–1269. Beck, T., Clarke, G., Groff, A., Keefer, P., & Walsh, P. (2001). New tools in comparative political economy: The database of Political Institutions. World Bank Economic Review, 15(1), 165–176. Bello, W. F., Cunningham, S., & Rau, B. (1994). Dark victory: The United States, structural adjustment and global poverty. London and Oakland: Pluto Press.

1706

WORLD DEVELOPMENT

Bird, G. (2002). The credibility and signalling effect of IMF programmes. Journal of Policy Modeling, 24, 799–811. Bird, G., & Rowlands, D. (2001). IMF lending: How is it affected by economic, political and institutional factors?. Policy Reform, 4, 243–270. Bookmann, B., & Dreher, A. (2003). The contribution of the IMF and the World Bank to economic freedom. European Journal of Political Economy, 19(3), 633–649. Cingranelli, D., & Richards, D. (2004). The CIRI Human Rights Database. . Collier, P., Guillaumont, P., Guillaumont, S., & Gunnning, J. W. (1997). Redesigning conditionality. World Development, 25, 1399–1407. Collier, P., & Gunnning, J. W. (1994). Trade and development: Protection, shocks and liberalization. In D. Greenway, & L. A. Winters (Eds.), Surveys in International Trade. Oxford, UK: Blackwell. Dollar, D., & Levin, V. (2004). The increasing selectivity of foreign aid. World Bank Policy Research Working Paper no. 3299. Dollar, D., & Svensson, J. (2000). What explains the success or failure of structural adjustment programmes?. The Economic Journal, 110, 894–917. Dreher, A. (2004). IMF and economic growth: The effects of programs, loans, and compliance with conditionality. WUSTL Economics Working Paper no. 0404004. Edwards, M. (2006). Signalling credibility? The IMF and catalytic finance. Journal of International Relations and Development, 9, 27–52. Fleck, R., & Kilby, C. (2001). World Bank independence: A model and statistical analysis of US influence. Vassar College Department of Economics Working Paper no. 53. Franc¸ois, J. F. (1997). External bindings and the credibility of reforms. In A. Galal, & B. Hoeckman (Eds.), Regional Partners in Global Markets. CEPR. Gilbert, C., Powell, A., & Vines, D. (1999). Positioning the World Bank. The Economic Journal, 109, 598–633. Goldstein, M., & Montiel, P. (1986). Evaluating Fund stabilization programs with multicountry data: Some methodological pitfalls. IMF Staff Papers no. 33, 304–344. Gwartney, J., Lawson, R., & Samida, D. (2000). Economic Freedom of the World 2000. Annual Report. . Haque, N. U., & Khan, M. (1998). Do IMF-supported programs work? A survey of the cross-country empirical evidence. IMF Working Paper no. 169. Harrigan, J., & Mosley, P. (1991). Assessing the impact of World Bank structural development lending 1980– 1987. The Journal of Development Studies, 27(3), 63–94. Heckman, J. (1979). Sample selection bias as a specification error. Econometrica, 47(1), 153–162. Khan, M. (1990). The macroeconomic effects of Fundsupported adjustment programs. IMF Staff Papers no. 37, 195–231.

Killick, T. (1998). Aid and the political economy of policy change. London: Routledge. Knight, M., & Santaella, J. A. (1997). Economic determinants of Fund financial arrangements. Journal of Development Economics, 54, 405–436. Martin, L. (2000). Democratic commitments: Legislatures and international cooperation. Princeton, NJ: Princeton University Press. Millimet, D. (2001). Endogeneity versus sample selection bias. . Nelson, P. (2000). Whose civil society? Whose governance? Decision making and practice in the new agenda at the Inter-American Development Bank and the World Bank. Global Governance, 6, 405–431. Pagan, A. (1984). Econometric issues in the analysis of regressions with generated regressors. International Economic Review, 25, 221–247. Poirier, D. J. (1980). Partial observability in bivariate probit models. Journal of Econometrics, 12, 209–219. Przeworski, A., & Vreeland, R. J. (2000). The effect of IMF programs on economic growth. Journal of Development Economics, 62, 385–421. Putnam, R. D. (1988). Diplomacy and domestic politics: The logic of two-level games. International Organization, 42(3), 427–460. Rama, M. (1993). Empirical investment equations for developing countries. In L. Se`rven, & A. Solimano (Eds.), Striving for Growth after Adjustment. Washington DC, US: The World Bank. Ratha, D. (2001). Demand for World Bank lending, World Bank Working Paper no. 2652. Rodrik, D. (1991). Policy uncertainty and private investment in developing countries. Journal of Development Economics, 36(2), 229–242. Rose, A. (2004). Do WTO members have more liberal trade policy? Journal of International Economics, 63(2), 209–436. Strand, H., Wilhelmsen, L., & Gleditsch, N. P. (2004). Armed Conflict Dataset Codebook, version 2.1, March 2004. . Vaubel, R. (1983). The moral hazard of IMF lending. In A. Meltzer (Ed.), International lending and the International Monetary Fund: A conference in memory of Wilson Schmidt. Washington DC, US: Heritage Foundation. Vella, F., & Verbeek, M. (1999). Estimating and interpreting models with endogenous treatment effects. Journal of Business and Economic Statistics, 17(4), 473–478. Vreeland, R. J. (2003). The IMF and economic development. Cambridge: Cambridge University Press. Winship, C., & Mare, R. D. (1992). Models for sample selection bias. Annual Review of Sociology, 18, 327–350. Wooldridge, J. M. (2002). Econometric analysis of cross section and panel data. Cambridge, MA: The MIT Press. World Bank (1998). Assessing aid: What works, what does not, and why, Policy Research Report. New York: Oxford University Press. World Bank (2003). World Development Indicators, CDROM. Washington DC: IBRD.

WORLD BANK CONDITIONAL LOANS AND PRIVATE INVESTMENT

1707

APPENDIX A. VARIABLES USED IN THE EMPIRICAL ANALYSIS AND SAMPLE PROJECTS Table 4. Description and sources of the selection equation variables Variables name Economic selection criteria GDP per capita Change in GDP per capita International reserves Exchange rate International trade Debt service Gross capital formation Political selection criteria Budget balance Trade with United States

Polity Population Post cold war Number under Number previous loans Legislative election Conflict proneness selection criteria Level of external conflict

Level of domestic conflict

Human rights selection criteria Respect for workers’ rights

Respect for personal integrity rights

Description GDP per capita, using current US dollars based on PPP Percentage change in GDP per capita, using current US dollars based on PPP International reserves, in terms of the number of months of imports which could be paid for Annual official exchange rate in local currency per US dollar Import plus export of goods and services, as a share of GDP Total debt service, as percent of exports of goods and services Gross fixed capital formation as a share of GDP

WDI 2003 WDI 2003 WDI 2003 WDI 2003 WDI 2003 WDI 2003 WDI 2003

Budget balance as a share of GDP WDI 2003 Country imports from US plus exports to United Comtrade States, as a percentage share of the United States total trade flows Score ranging from 10 to +10, for strongly Rose (2004) autocratic and democratic systems, respectively Number of residents WDI 2003 Dummy variable coded 1 from 1992 onwards, zero otherwise Number of other committing countries WBPD Number of previous conditional loans WBPD Dummy coded 1 if legislative election takes place, and DPI 2001 zero otherwise Interstate conflict level, based on a four-point scale: no SWG 2004 conflict (0), minor conflict (1), intermediate conflict (2), war (3) Internal conflict level, based on a four-point scale: no SWG 2004 conflict (0), minor conflict (1), intermediate conflict (2), war (3) Workers’ rights, based on a three-point scale: severely restricted (0), somewhat restricted (1) or fully protected (2) Physical integrity rights index, ranging from 0 (no respect) to 8 (full government respect of these rights)

Temporal dependence No. of years since previous loan Number of years since previous loan No. of years since previous loan squared Number of years since previous loan squared a

Sourcea

CIRI 2004

CIRI 2004

WBPD WBPD

WDI stands for World Development Indicators; DPI for Beck et al. (2001) Database of Political Institutions; SWG for Strand, Wilhelmsen, Gleditsch; CIRI for Cingranelli–Richards Human Right Database; WBPD for World Bank online Projects Database.

1708

WORLD DEVELOPMENT Table 5. Number of SALs and SECALs for each sample country (from 1982 to 1999)

Argentina Bangladesh Bolivia Brazil Bulgaria Chile China Colombia Costa Rica Cote d’Ivoire Croatia Czech Republic Dominican Republic Ecuador Egypt, Arab Rep. El Salvador Estonia Gambia, The Guatemala Guinea-Bissau India Total number Mean Min Max

4 6 4 1 3 3 0 1 4 3 0 0 0 3 1 0 0 0 0 1 3

Indonesia Kenya Lithuania Madagascar Malawi Malaysia Mauritius Mexico Nicaragua Panama Papua New Guinea Paraguay Peru Philippines Poland Thailand Trinidad and Tobago Tunisia Uruguay Venezuela, RB

98 2.39 0 7

Note: SAL stands for Structural Adjustment Loan, SECAL for Sector Adjustment Loan. Source: World Bank online Projects Database.

Available online at www.sciencedirect.com

6 7 1 2 4 1 1 4 3 2 2 0 5 7 1 5 0 4 4 2