The United States, Bilateral Debt-for-Nature Swaps, and Forest Loss: A Cross-National Analysis

Abstract

We engage with the theoretical and empirical literature on the effectiveness of debt-for-nature swaps in promoting environmental protection. We present cross-national evidence that US bilateral debt-for-nature swaps are associated with less forest loss. Using a two-stage instrumental variable regression model to analyse a sample of 85 low- and middle-income countries from 2001 to 2014, we find that higher amounts of debt reduction and higher amounts of conservation funds generated as a result of such swaps are associated with lower rates of forest loss.

1. Introduction

The debt crisis of the 1980s was highlighted by an inability of many nations to generate enough revenue to repay their foreign debts (George, 1992). To finance interest and amortisation payments, nations turned to economic policies that increased exports to generate revenue. However, such policies led to extensive forest loss because exports tended to be concentrated in extractive sectors like forestry and agriculture (Rich, 2013).

The connection between debt repayment and forest loss led Dr. Thomas Lovejoy, then of the World Wildlife Fund, to propose a ‘debt-for-nature swap’ as a potential solution. In a New York Times editorial, Lovejoy wrote,

The international debt crisis should remind us of the ecological as well as the economic links between rich and poor…debtor nations willing to protect natural resources could be made eligible for discount or credits against their debts…stimulating conservation while ameliorating debt would encourage progress on both fronts. (1984, p. 31)

The basic idea of such swaps involves cancelling a portion of a nation’s foreign debt in exchange for investment in conservation programmes (Visser & Mendoza, 1994).

Among heightened public concern about the loss of forests and calls to address the issue, the United States Congress passed the Enterprise of Americas Initiative, which authorised the United States to carry out debt-for-nature swaps with other nations, in 1990 (Sheikh, 2009). The United States reauthorized the programme in 1998 with the passage of the Tropical Forest Conservation Act (Sheikh, 2009). In total, United States debt-for-nature swaps cancelled approximately $1.8 billion owed by 21 low- and middle-income nations. The swaps generated $400 million for conservation. In comparison, debt-for-nature swaps carried out by all other high-income nations totalled $1 billion of debt cancelled and generated about $500 million for conservation. The United States is by far the largest originator of bilateral debt-for-nature swaps. It is responsible for 64 per cent of the total debt forgiveness and generating 44 per cent of the total conservation funding.

Despite the large amount of conservation funding created through this mechanism, to our knowledge there is little cross-national research that evaluates whether such swaps correspond with less forest loss. This is surprising for a few reasons. First, the United States provides the most funding for such programmes, and empirical research that establishes the effectiveness of such swaps should be of interest to policy-makers. Second, the existing cross-national evidence on this topic raises questions about their effectiveness. For instance, Kraemer and Hartmann (1993) analysed cross-national data for 34 low- and middle-income nations using ordinary least squares regression and found no relationship between debt-for-nature swap participation and forest loss. Kraemer and Hartmann (1993) concluded that any understanding of how debt-for-nature swaps impact forest loss needs to be done by case study research. As a result, a limited number of such studies have been published that evaluated the effectiveness of United States bilateral debt-for-nature swaps in specific locations. For instance, Gockel and Gray (2011) used surveys, interviews, and field work to evaluate if such swaps have been effective in the implementation of conservation projects and the monitoring for illegal forestry extraction in Peru. Cassimon, Prowse, and Essers (2011) used project and government documents to evaluate the impact of a United States debt-for-nature swap in Indonesia. However, the comparability of such studies, given their case-specific nature, makes overall generalisations difficult.

Third, there is a nascent but burgeoning and related literature that examines how bilateral aid in the environmental sector impacts forests. For instance, Arvin and Lew (2009) find that higher levels of bilateral aid for conservation actually correspond with increased, not decreased, forest loss. Bare, Kauffman, and Miller (2015) report a similar finding for a sample of sub-Saharan African nations and attribute this finding to programmes that create protected areas which displace local people, who then go on to clear forests elsewhere. However, Bare et al. (2015) argue that these counter-intuitive empirical findings may arise as a result of not addressing donor selection bias, examining only one geographical region, or using a measure that includes aid from several bilateral donors, whose aid may have differing effects on the environment. Hermanrud and de Soysa (2016) begin to address such issues in their study by using a two-stage model to control for potential donor selection bias. The authors find that Norwegian bilateral aid for conservation has no impact on forest loss when using such a model.

Fourth, there is now greater data availability on the United States financing of bilateral debt-for-nature swaps. Sheikh (2009) has gathered information on United States bilateral debt-for-nature swaps, and the reported information includes recipient nations, amounts of conservation funds generated, amounts of debt cancelled, and approval dates of the agreements. By utilising this information, detailed analyses are more possible to perform than ever before, yielding greater specificity regarding how United States bilateral debt-for-nature swaps affect forest loss in our empirical assessment.

These points provide the justification and starting point for our study. We integrate and build upon the existing research in this area in several novel ways. First, we focus only on United States bilateral debt-for-nature swaps. We do so in an attempt to isolate the impact of this form of financing on forests, seeing as how the United States has been the largest actor to use bilateral debt-for-nature swaps and various other governments may pursue differing policies guided by their own institutional mandates. Second, we do not restrict our analysis to one geographical region but rather include all low- and middle-income nations according to the World Bank (2016) income classification of countries. We expand the population of interest because forest loss is not concentrated in only one geographical place, and countries across the globe are eligible to participate in United States bilateral debt-for-nature swaps. Third, to address the potential impacts of donor selection bias when evaluating the effects of debt-for-nature swaps on forest loss, we use a two-stage instrumental variable regression model (Easterly, 2006). Thus, we seek to isolate and evaluate the effects of such swaps on forests, using the appropriate and sophisticated methodology to overcome limitations in previous analyses.

We now turn to a discussion of United States bilateral debt-for-nature swaps and why they may be associated with less forest loss. We then describe the variables and methodology that allows us to address potential problems with selection bias. We conclude by talking about the findings along with the theoretical, methodological, and policy implications.

2. The structure of United States bilateral debt-for-nature swaps

The core idea of a bilateral debt-for-nature swap involves reducing debt that a low- or middle-income nation owes in exchange for that nation making investments in conservation and environmental protection (Hamlin, 1989). The concept is simple but, in reality, swaps are complex financial transactions that involve several different actors. The actors may include indebted governments, creditor governments, as well as international non-governmental organisations, domestic non-governmental organisations, private foundations, and commercial banks (Bryant & Bailey, 1997). There is some variation in terms of the details of how bilateral debt-for-nature swaps are conducted across different nations, however they all tend follow the same three steps (Visser & Mendoza, 1994).

First, an indebted nation works with an international non-governmental organisation to send a letter to the appropriate United States government officials to express interest in carrying out a debt-for-nature swap (United Nations Convention for Biological Diversity, 2001). If there is interest between indebted and creditor nations, the governments enter into negotiations pertaining to eligibility and the structure of the transaction.

The negotiations begin with determining eligibility of an indebted nation (Cartwright, 1989). There are various economic, political, and environmental requirements a nation must meet in order to participate in a bilateral debt-for-nature swap with the United States (Sheikh, 2009). In terms of economic requirements, a debtor government must have met thresholds for debt service and show progress toward implementing reforms involving an International Monetary Fund (IMF) or World Bank (WB) structural adjustment agreement (United Nations Convention of Biological Diversity 2001). In the political realm, a debtor nation must have a democratically elected government and demonstrate a commitment to ensuring human rights (United Nations Convention for Biological Diversity, 2001). Most importantly, an indebted nation must meet certain environmental criteria including the existence of at least one tropical forest with substantial biodiversity or a forest that represents a large intact block of forest on a regional, continental, or global scale (United Nations Convention for Biological Diversity, 2001). There is also consideration of the urgency of the request that centres upon threats to biodiversity in the area to be protected (Sheikh, 2009).

The discussion also focuses on the technical aspects of the transaction (Cartwright, 1989). The topics negotiated include amount of debt available for forgiveness, instrument of redemption, and a time frame for completion of the agreement (Visser & Mendoza, 1994). Further, government officials discuss where conservation projects will be implemented, what the projects will involve, and how they will be managed (Visser & Mendoza, 1994). There is emphasis on local stakeholder involvement, which tends to be done well in advance of the negotiations (Sheikh, 2009). A non-governmental organisation then prepares an independent feasibility assessment of the agreement that highlights any potential issues to address before the agreement enters into effect (United Nations Convention on Biological Diversity 2001).

Once negotiations are complete, the newly approved agreement restricting debt is put into place (Visser & Mendoza, 1994). This may take the form of debt reduction or buy back (United Nations Convention for Biological Diversity, 2001). The debt reduction procedure involves the United States reducing a recipient nation’s debt by the negotiated amount (Sheikh, 2009). A country’s outstanding debt to the United States is reissued with a new repayment (Sheikh, 2009), while the indebted nation pays what would have been paid on the forgiven debt into a conservation trust fund (Visser & Mendoza, 1994). In a buy back, the indebted nation pays the United States a negotiated single payment and a similar payment to a local conservation trust fund (Sheikh, 2009). In most bilateral debt-for-nature swaps, the United States agrees to cancel more debt than an indebted nation will pay into a trust fund (Visser & Mendoza, 1994). Further, the United States provides various tax incentives to private lenders to cancel debt in conjunction with the bilateral debt-for-nature swap (Visser & Mendoza, 1994) in an effort to leverage participation by foreign governments and commercial banks with the goal being to reduce a country’s need to export natural resources to generate revenue for debt repayment (Fuller & Williamson, 1989).

The trust fund tends to be financed with bonds rather than cash to avoid possible inflation that may accompany flooding local currency into the economy (Cartwright, 1989). The bonds also provide a more reliable source of financing for conservation projects over the long-term (Visser & Mendoza, 1994), because the trust fund is administered by a board of directors that is made up of local citizens, international non-government organisations, and at least one representative of the indebted and creditor governments (Sheikh, 2009). It is important to note that conservation funds generated as part of a United States bilateral debt-for-nature swap will usually far exceed government spending for conservation (Patterson, 1990).

The interest from the bonds allows the trust fund to provide grants for local non-governmental organisations involved in conservation and environmental protection (Visser & Mendoza, 1994). The initial investment is used to establish an endowment or source of long-term funding for domestic conservation groups when the bonds reach maturity (Sheikh, 2009). This helps ensure that conservation projects are not disrupted and new projects will be implemented. There are different oversight mechanisms built into the agreements including review of projects at various times during the agreement of supported projects internally by the trust fund’s board of directors and externally by international non-governmental organisations or multilateral organisations (Visser & Mendoza, 1994).

There are several types of activities that can be funded under the United States Tropical Forestry Conservation Act (Sheikh, 2018). First, projects may involve the establishment or restoration of protected areas including demarcation of protected forests, expansion of existing protected forests, and monitoring of protected areas for illegal logging (Gockel & Gray, 2011). Second, the establishment or restoration of protected areas correspond with putting into place scientifically-sound systems of natural resource management along with training to increase the technical and managerial capacities of individuals and organisations involved in conservation efforts (Gockel & Gray, 2011). These may include carrying out forestry inventories, testing silvicultural techniques, starting extension services for communities, and implementing environmental education programmes. Third, the training and management programmes seek to develop a country’s civil society and promote the implementation of ‘people-first’ initiatives that simultaneously address a community’s livelihood concerns, especially among women and conservation (Sheikh, 2018).

3. United States bilateral debt-for-nature swaps and forest loss: considering the evidence

What have United States bilateral debt-for-nature swaps funded? Have they been effective in decreasing forest loss? We address these questions by considering the limited scholarly research on the topic. Gockel and Gray (2011) evaluate two projects funded by a United States debt-for-nature swap signed in 2002 that cancelled $14.3 million of Peruvian debt and generated $10 million for conservation. The first project assessed was a long-term integrated conservation and development project in Peru’s Pacaya Samiria National Reserve. The project involved a domestic non-governmental organisation working with villagers to develop species-specific management plans that centred on biological monitoring of threatened plants and animals. There was also an emphasis on decreasing forest loss by putting into place palm management plans that focused on establishing a replanting schedule and setting harvesting limits (Gockel & Gray, 2011).

The authors conclude from a survey of 57 villagers living near the reserve and from interviews with non-governmental organisation staff that the project did improve species conservation and socio-economic wellbeing. This corresponded with data from the project’s biological monitoring plan. However, the authors note that improvements in socio-economic wellbeing were difficult to confirm because little baseline data on household assets were collected – a common issue for debt-for-nature swaps because they rarely include in-situ monitoring (Gockel & Gray, 2011). This was further complicated by aid programmes by other non-governmental organisations being carried out in the area, including the Nature Conservancy’s Parks in Peril Program (Gockel & Gray, 2011).

A second set of projects were centred on efforts to prevent illegal logging of mahogany in Alto Purus National Park (Gockel & Gray, 2011). The project supported the construction of guard posts around the park and along neighbouring rivers to limit access to the park. There was also an emphasis on improving in-situ inspections of mahogany concessions to ensure that the wood was not coming from inside the park by creating a methodology to verify mahogany populations, establish baseline data of mahogany, and verify annual operation plans of forest concessions (Gockel & Gray, 2011). The project also sought to strengthen civil society in the region by working with indigenous communities to plant communal mahogany nurseries to regenerate future stocks and to monitor for illegal logging (Gockel & Gray, 2011).

The authors conclude from interviews with project staff and government officials along with participant observation that the projects were successful in improving the capacity of local people to manage forest resources while improving the capacity of domestic non-governmental organisations to manage the projects described above. Most notably, the debt-for-nature swap created a baseline of mahogany stores, which provided a tool for monitoring illegal logging into the future (Gockel & Gray, 2011). Nevertheless, the effectiveness of projects was hindered by short-term funding cycles of grants financed by the debt-for-nature swap, such as guard posts being left vacant once the grant ended (Gockel & Gray, 2011).

Another study by Cassimon et al. (2011) evaluated the effectiveness of the largest United States bilateral debt-for-nature swap that was carried out with Indonesia in 2011. The authors analysed swap documents and statistical data to determine how it performs across several commonly identified limitations of these transactions. First, the authors determined that Indonesia had approximately $22 million of debt owed to the United States cancelled. However, the debt-for-nature swap did not generate any additional fiscal space in Indonesia’s budget because payments were redirected into a conservation fund (Cassimon et al., 2011). Further, because of a lack of statistical data, it is unclear the degree to which the debt-for-nature swap generated subsequent investment by the United States in conservation because of the fungibility of the conservation funds generated by the swap (Cassimon et al., 2011). At the same time, the amount of debt cancelled by this debt-for-nature swap represents only a fraction of Indonesia’s outstanding debt. Thus, the authors conclude that it is unlikely that the amount of debt cancelled has eliminated the need for Indonesia to cash in on natural resources to meet its remaining debt obligations.

Bilateral debt-for-nature swaps tend to be most effective when the goals of the swap align with policy priorities of a government (United Nations Conference on Biological Diversity 2001). In this regard, Indonesia has published its action plan to address climate change by limiting illegal logging, improving reforestation, and establishing protected areas in Sumatra (Cassimon et al., 2011). It has also integrated similar insights into its Poverty Reduction Strategy Papers, indicating that any attempts to increase economic growth and, as a result, decrease poverty needs to take conservation seriously (Prowse, Grist, & Sourang, 2009). Although policy alignment may appear present, Cassimon et al. (2011) argue that this debt-for-nature swap suffers from a lack of system alignment or the failure to work with existing government structures in an effort to build capacity and ensure maximal funds are being spent on conservation initiatives. This is largely the result of the debt-for-nature swap creating a separate oversight structure rather than being integrated into existing conservation programmes in Indonesia (Cassimon et al., 2011).

Our review of the major studies of bilateral debt-for-nature swaps highlight some of the challenges in evaluating their effectiveness. Close analysis of specific cases provides us with a richer understanding of both the on-the-ground conditions as well as policy-level considerations that yield the intended conservation results in specific instances. However, such studies can only inform but not answer the broader macro-comparative question of whether United States bilateral debt-for-nature swaps do produce appreciable benefits in terms of less forest loss in the countries that participate in such conservation programmes. To address this question, we now turn to a discussion of our methodology and the sources of data that we use in our analysis.

4. Methods

We begin by describing the sources for the data we use to address our research question of whether United States bilateral debt-for-nature swaps are associated with less forest loss.¹ In addition to using newly available data that we combine from a number of sources, we develop a more sophisticated analysis that attempts to account for one potential source of erroneous inference about the effectiveness of such swaps, which is endogeneity or selection bias in terms of which countries enter into bilateral debt-for-nature swaps with the United States. We now turn to our data by describing our variables.

4.1. Dependent variable

4.1.1. Forest loss

We use newly available data from the World Resources Institute’s Global Forest Watch (2016) online database. The forest loss data are derived from high resolution satellite imagery, where each pixel of the image represents an area of 30 square meters. These data eliminate reliability issues with previous estimates of forest loss such as data from the United Nations Food and Agriculture Organisation (2010), which rely on various methods of collection including being based upon expert opinion, extrapolated from an old forest inventory, or estimated from outdated remote sensing surveys (Grainger, 2008). See Hansen, Stehman, and Potapov (2010) for a discussion of the methodology used to arrive at the estimates.

We calculate forest loss using the three-step procedure described by Rudel (2013). First, we obtain the hectares of forest cleared from 2001 to 2014 based on a 75 per cent tree cover canopy density level. The tree cover density for a nation represents the estimated percentage of a pixel taken from satellite imagery that is covered by tree canopy (World Resources Institute, 2016). The 75 per cent tree cover canopy density level is the level associated with gain or loss of ‘wet’ forests (Rudel, 2017). We also calculate a second measure of ‘dry’ forest loss using a 50 per cent tree cover canopy density level. Second, we obtain a nation’s total forest area, measured in hectares, in 2000. Third, the difference in tree cover canopy density from 2001 to 2014 is an unstandardized measure of forest loss; to obtain the rate of forest loss, we divide the number of hectares cleared (from 2001 to 2014) by the country’s total forest area in 2000. This yields a rate of forest loss for the specific period in each nation in our sample. The descriptive statistics for this and all other variables are reported in Table 1.

4.2. Independent variables

4.2.1. Bilateral debt-for-nature swap conservation funding

We include a variable that measures the total amount of conservation funds generated by a United States bilateral debt-for-nature swap within a nation in the period between 1990 and 2001, per capita. The conservation funding data may be obtained from Sheikh (2009). This source provides information on the location of United States bilateral debt-for-nature swaps, the dates of the transaction, and conservation funds generated. The funding data are aggregated over this period because of the amount of time necessary to negotiate and initiate conservation projects as well as the potential for multiple agreements to be made with the same nation. The population data for 2000 come from the World Bank (2016). We log the variable because it is skewed. We anticipate that higher levels of conservation funds per capita should correspond with less forest loss.

4.2.2. Bilateral debt-for-nature swap debt reduction

The other way that a United States bilateral debt-for-nature swap should decrease forest loss is by reducing a country’s foreign debt. Thus, we include the total amount of debt cancelled by the United States as part of a bilateral debt-for-nature swap. We standardise this measure by dividing it by a country’s population size in millions. The data also come from Sheikh (2009). We log this variable to correct for its skewed distribution.

4.2.3. Bilateral environmental aid

It is also important to evaluate the impact of United States debt-for-nature swaps independent of other forms of conservation aid that a nation receives. Thus, we include bilateral environmental aid received from 1990 to 2000 as an important control against potential confounding. We expect that higher levels of environmental aid should correspond with lower levels of forest loss. This is because such aid often funds conservation projects including protected areas, demarcation of forest boundaries, eco-tourism, agroforestry, and monitoring of illegal logging, among others (Bare et al., 2015).

4.2.4. International Monetary Fund structural adjustment

We include a variable that measures the total amount of money that a nation receives as part of an International Monetary Fund (IMF) structural adjustment loan for 2000. We divide this value by a country’s population size to standardise it. This variable is logged. The data may be obtained from the World Bank’s World Development Indicators online database (2016). We expect higher amounts of structural adjustment lending should be related to increased forest loss. This is because the International Monetary Fund requires borrowing nations to implement certain macro-economic reforms in return for the loan. These reforms include decreasing government spending for conservation while increasing exports of natural resources, which may both adversely impact forests (Rich, 2013).

4.2.5. Debt service

We include debt service or the sum of principal and interest payments in foreign currency, goods, or services on long-term public and publicly guaranteed private debt or debt with a maturity of two years or longer as a percentage of exports of goods and services. The data may be obtained from the World Bank (2016). We expect that higher levels of debt service are associated with higher rates of forest loss because indebted nations tend to increase exports of natural resources to earn cash for repayment (Marquart-Pyatt, 2004).

4.2.6. Non-governmental organisations

We include the number of international non-governmental organisations working on environmental and animal rights issues in a nation for 2000. The data are collected by Smith and Wiest (2005) from the Yearbook of International Associations. We divide it by a country’s population size in order to standardise the variable. Schofer and Hironaka (2005) find that higher levels of non-governmental organisations are associated with lower rates of deforestation. This may be the case because non-governmental organisations finance local conservation projects, support social movement activity, and shape the language of environmental laws (Shandra, 2007).

4.2.7. Democracy

We use the average of Freedom House (2005) political rights and civil liberties scales to measure the level of democracy within a nation. According to Freedom House (2005), political rights refer to the degree to which a nation is governed by democratically elected representatives and has fair, open, and inclusive elections. The civil liberties scale measures the level of freedom of press, freedom of assembly, general personal freedom, freedom of private organisations, and freedom of private property within a nation (Freedom House, 2005). The variables have the following coding: free (1–2), partially free (3–5), and not free (6–7). We multiply the index by negative one so that high scores correspond with high levels of democracy. According to Li and Reuveny (2006), democracy should reduce forest loss because democratic nations have more political activism than repressive nations. This situation plays out in democratic nations because they guarantee freedoms of speech, press, and assembly while being able to hold leaders accountable via elections (Li & Reuveny, 2006).

4.2.8. Protected forest area

We control for the percentage of protected forest area within a nation. These data may be obtained from the Food and Agriculture Organisation of the United Nations (2010). We log this variable because it is skewed. We expect that higher levels of protected forest area should correspond with less forest loss because such areas should be off limits to various forms of forestry extraction (Bryant & Bailey, 1997).

4.2.9. Total population growth

We include the average annual percentage change in total population growth from 1990 to 2000. These data come from the World Bank (2016). We expect that higher rates of population growth correspond with more forest loss with demand for resources from a growing population leading to increased consumption of resources that adversely impact forests (Rudel, 1989).

4.2.10. Agricultural land area

We include the percentage of land within a nation used for agriculture or land that is permanently under crops or pasture (World Bank, 2016). This variable measures the size of the agricultural sector. We take the natural log of this variable to help deal with skewness. We expect that higher amounts of agricultural land correspond with more forest loss because forests are converted to pastures or fields (Austin, 2010).

4.2.11. Gross domestic product per capita

We employ a measure of gross domestic product per capita adjusted for purchasing power parity for 2000. The data may be obtained from the World Bank (2016). We log this variable to correct for its skewed distribution. Burns, Kick, and Davis (2003) find that higher levels of economic development are associated with lower rates of forest loss based on wealthier nations externalising their environmental costs by importing natural resources and moving toward economic activities like services that have less impact on forests.

4.2.12. Economic growth

We include the average annual economic growth rate from 1990 to 2000. These data may be obtained from the World Bank (2016). The cross-national research that examines how economic growth affects forest loss yields contradictory findings. On the one hand, economic growth is associated with higher rates of deforestation (Jorgenson, 2006). This is because there are large amounts of capital available for investment in activities (for example, infrastructure) that accelerate forest loss during periods of economic expansion (Rudel, 1989). On the other hand, an absence of economic growth may increase forest loss. The lack of growth offers rural populations little incentive to migrate to urban centres for work and rather expand agriculture and forestry to survive (Ehrhardt‐Martinez, Crenshaw, & Craig Jenkins, 2002).

4.3. Analytic strategy

We note earlier that issues of endogeneity must be taken seriously when examining the impact of bilateral aid in cross-national research (Easterly & Easterly, 2006). In this instance, donor selection bias on the bilateral debt-for-nature swap variable may lead to distorted coefficients and inefficient tests of statistical significance (Easterly & Easterly, 2006). The problem of endogeneity arises when one of the ‘explanatory’ variables, on the right-hand side of the regression equation, is jointly determined with the left-hand side ‘dependent’ variable to be explained (Wooldridge, 2015). Consider the ordinary least squares regression equation for forest loss in Equation (1) to illustrate this point:

(1) $\mathrm{y}1=\beta 0+\beta 1\mathrm{y}2+\beta 2\mathrm{x}1+\mathrm{e}$(1)

where y1 represents the dependent variable, forest loss, y2 represents the endogenous variable or bilateral debt-for-nature swap financing, x1 represents variables assumed to be exogenous, and e represents the error term. In this case, y2 may be considered endogenous if the same processes that drive forest loss in a country are also those that affect the country entering into a debt-for-nature swap agreement. Using ordinary least squares regression to estimate the effect of bilateral debt-for-nature swaps on forest loss would lead to biased estimates because investments are not randomly assigned, and so the regression equation is capturing the selection on receiving financing as well as the effects of swap financing in one parameter (Wooldridge, 2015). The estimate is also inconsistent since the error term e will be correlated with the endogenous predictor y2 (Wooldridge, 2015).

One way to address this problem is to find a variable (or variables), called an instrument or instrumental variable, that satisfies the property of being correlated with receiving aid but uncorrelated with forest loss (Easterly & Easterly, 2006). In a two-stage model, the predicted values for the instrument are obtained in the first stage, which are then used in the second stage to obtain an unbiased estimate of forest loss. We carry out this analysis using Stata version 13 and its ‘ivreg2’ command (Baum, 2006). We can write the equation for our endogenous explanatory variable as follows:

(2) $\mathrm{y}2=\pi 0+\pi 1\mathrm{z}1+\pi 2\mathrm{z}2+\mathrm{v}$(2)

where z1 is an instrument for y2 if it is still correlated with y2 after partialing out the effects of other exogenous predictors (z2) (Wooldridge, 2015). Note, the set of exogenous predictors (x1) from Equation (1) need not be the same as those included in the second equation (z2). The residuals v are the part of the endogenous predictor y2 that are uncorrelated with the instrument z1, which are inserted in the second stage to obtain the unbiased estimates (Wooldridge, 2015):

(3) $\mathrm{y}1=\beta 0+\beta 1\mathrm{y}2+\beta 2\mathrm{x}1+\beta 3\mathrm{z}1+\beta 4\mathrm{v}+\omega$(3)

In the first stage, we include two exogenous predictors of receiving United States bilateral debt-for-nature swap funding or debt cancellation. They are the circular distance from Washington D.C. in miles and the number of threatened birds within a nation.² We log both variables because they are skewed. When using this modelling strategy, there are assumptions that must be met concerning the choice of instruments. First, we must determine that the instruments are ‘relevant’ or that they are correlated or statistically dependent with the endogenous variable but not the dependent variable (Wooldridge, 2015). To test this assumption, we look at the coefficients for the Anderson canonical correlation. This chi-square test reaches a level of significance in every model. Thus, we reject the null hypothesis and conclude that our instruments are relevant (Wooldridge, 2015).

Second, we determine if the instruments are ‘valid’ by testing to ensure that they are uncorrelated with the error term (Baum, 2006). We consider the Sargan chi-square statistics. As the null hypothesis is statistical independence, the test statistic does not reach a level of significance, thereby indicating the instrumental variables are not correlated with the error term (Baum, 2006).

Third, we test to see if the instruments are ‘weak’ or explain only a small amount of variation of the endogenous variable. This does not appear to be the case here. We calculate a Cragg-Donald F-statistics for each equation. We then compare these values against Stock-Yogo critical values to determine if the instruments are weak (Wooldridge, 2015). In the models with conservation funding, the test statistic for the F-tests are greater than the Stock-Yogo critical values for a 15 per cent bias level. For the models including the amount of debt forgiven, the F-tests are greater than the Stock-Yogo critical values at the 10 per cent maximal bias level.

We must also ensure that we are not violating other regression assumptions. First, we calculate mean and highest variance inflation factor scores for each model using ‘ivvif’ command (Roodman, 2014). We report the values in Table 2. There do not appear to be any potential problems with multicollinearity because mean and highest variance inflation factor scores do not exceed a value of 2.5 (Tabachnick & Fidell, 2013).

Table 1. Descriptive statistics and bivariate correlation matrix for forest loss analysis N = 85

	Mean	SD	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)	(14)
(1) Wet Forest Loss, 2001–2014	.146	.234	1.000
(2) Dry Forest Loss, 2001–2014	.104	.197	.464	1.000
(3) Bilateral Debt-For-Nature Swap Conservation Funding, 1990–2000	.461	1.747	−.103	−.076	1.000
(4) Bilateral Debt-For-Nature Swap Debt Reduction, 1990–2000	.491	1.983	−.099	−.071	.978	1.000
(5) Debt Service, 2000	43.371	25.393	.033	.177	−.027	−.025	1.000
(6) International Monetary Fund Loans, 2000	2.782	1.208	.039	−.003	−.056	−.011	.043	1.000
(7) Bilateral Environmental Aid, 1990–2000	.001	.001	−.056	−.013	.117	.137	.098	.131	1.000
(8) Environmental Non-Governmental Organisations, 2000	.001	.001	.013	.030	−.007	.018	.235	.120	.146	1.000
(9) Democracy, 2000	4.082	1.566	.025	−.041	−.233	−.231	.026	.038	−.138	−.168	1.000
(10) Protected Forest Area, 2000	2.389	.883	.118	.005	.116	.100	−.031	−.168	.165	−.220	−.063	1.000
(11) Population Growth Rate, 1990–2000	.200	.139	.390	.323	−.063	−.074	.114	−.344	.019	−.092	.141	.235	1.000
(12) Agricultural Land Area, 2000	3.683	.610	.108	.006	−.012	−.037	−.165	.014	−.128	−.252	.026	.084	−.167	1.000
(13) Gross Domestic Product per capita, 2000	6.617	1.019	−.241	−.203	.258	.276	−.351	−.086	.058	.254	−.443	−.003	−.284	−.078	1.000
(14) Economic Growth Rate, 1990–2000	3.449	3.696	−.077	−.180	−.074	−.089	−.248	.074	−.025	−.307	.251	.183	−.228	.385	−.056	1.000

Table 2. Second stage instrumental variable regression estimate of debt-for-nature swaps on wet forest loss

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model 9	Model 10
Independent Variables
Bilateral Debt-For-Nature Swap Conservation Funding, 1990–2000	−.074*		−.077*		−.094*		−.091*		−.088*
	(.038)		(.039)		(.046)		(.043)		(.049)
Bilateral Debt-For-Nature Swap Debt Reduction, 1990–2000		−.620**		−.063*		−.077*		−.075*		−.073*
		(.033)		(.033)		(.038)		(.035)		(.041)
Debt Service, 2000			.001	.001	.001	.001	.001	−.001	−.001	−.001
			(.001)	(.001)	(.001)	(.001)	(.001)	(.001)	(.001)	(.001)
International Monetary Fund Loans, 2000			.001	.006	.008	.013	.034	.040*	.033	.039
			(.024)	(.023)	(.025)	(.024)	(.025)	(.024)	(.024)	(.024)
Bilateral Environmental Aid, 1990–2000			.001	.001	−.001	−.001	−.001	−.001	−.001	−.001
			(.001)	(.001)	(.001)	(.001)	(.001)	(.001)	(.001)	(.001)
Non-Governmental Organisations, 2000					1956.158	2875.303	5157.933	5894.512	4984.996	5509.505
					(9016.68)	(8827.93)	(8554.01)	(8362.75)	(9375.45)	(9169.23)
Democracy, 2000					−.019	−.017	−.029	−.027	−.026	−.024
					(.022)	(.021)	(.020)	(.020)	(.021)	(.020)
Protected Forest Area, 2000					.056	−.054	.031	.029	.036	.034
					(.035)	(.156)	(.034)	(.033)	(.034)	(.034)
Population Growth Rate, 1990–2000							.751***	.761***	.696**	.707**
							(.223)	(.217)	(.234)	(.230)
Agricultural Land Area, 2000							.068	.064	.076	.072
							(.047)	(.046)	(.048)	(.048)
Gross Domestic Product per capita, 2000									−.009	−.007
									(.040)	(.041)
Economic Growth Rate, 1990–2000									−.006	−.006
									(.009)	(.009)
Constant	.180	.177	.171	.152	.106	.080	−.262	−.271	−.212	−.238
R-Squared	.147	.160	.134	.153	.079	.112	.209	.239	.233	.256
Number of Countries	85	85	85	85	85	85	85	85	85	85
Highest Variance Inflation Factor Score	1.00	1.00	1.12	1.13	1.54	1.46	1.58	1.51	2.39	2.48
Mean Variance Inflation Factor Score	1.00	1.00	1.07	1.07	1.23	1.21	1.28	1.26	1.57	1.57
Anderson Under-Identification Test	15.519***	16.176***	15.478***	16.817***	12.472**	14.003***	12.590**	13.999***	10.091**	10.965**
Sargan Over-Identification Test	1.062	1.390	1.137	1.351	.853	1.065	.876	1.165	1.123	1.397
Cragg-Donald Weak Instrument Test	9.157**	9.636**	8.794**	9.742**	6.535*	7.495*	6.433*	7.295*	4.849	5.332*

Notes: * Indicates p < .05, ** indicates p < .01, and *** indicates p < .001 for a one-tailed test. The first number is the unstandardized coefficient, the second number in parentheses is the standard error.

Second, we examine scatterplots of each independent variable against the dependent variables to determine if there are any problems with linearity (Allison, 1999). We transform variables when appropriate (Tabachnick & Fidell, 2013). We noted any transformations in the preceding section.

Third, we calculate Pagan-Hall tests to determine if heteroscedasticity is problematic. We use the ‘ivhettest’ command following estimation of the instrumental variable model (Schaffer, 2013). The null hypothesis for this chi-square test is that the error variances are homoscedastic or equally distributed (Schaffer, 2013). The test statistics do not reach a level of statistical significance in any of the models, indicating no potential problems with heteroscedasticity (Tabachnick & Fidell, 2013).

Fourth, we calculate standardised residuals for each model to determine if there are any potential problems with outliers. If a standardised residual exceeds an absolute value of three, then it may pose problems for the regression model (Tabachnick & Fidell, 2013). We do not find any potential problems with outliers for any of the models.

5. Findings

In Table 2, we present the second stage estimates of forest loss from the two-stage least squares regression model. The first number reported is the unstandardized regression coefficient and the second number in parentheses is the standard error. Because of our theoretical expectations, we use one-sided tests of significance.

The dependent variable in Table 2 is tropical or wet forest loss. The odd-numbered equations estimate the impact of conservation aid generated by United States debt-for-nature swaps on forest loss. The even-numbered equations estimate the impact of debt reduction from United States debt-for-nature swaps on forest loss. In models (2.1) and (2.2), we include only these two variables in the first two equations. In models (2.3) and (2.4), we add international factors including debt service, structural adjustment,³ and total bilateral conservation aid received.⁴ In models (2.5) and (2.6), we add political factors including non-governmental organisations, democracy, and protected forest area. We add demographic variables including population growth and agricultural land area in models (2.7) and (2.8). In models (2.9) and (2.10), we include our economic measures – gross domestic product per capita and economic growth.

Let us begin by focusing on the statistically significant findings for our main independent variables. In the odd-numbered models, we find that higher levels of conservation funds generated as the result of a swap transaction are related to lower levels of wet (tropical) forest loss. Across all models, the coefficients for the conservation funding variable are negative and significant. We also find that higher levels of debt reduced by a swap are related to less wet (tropical) forest loss. In all of the even-numbered models, the coefficients for this variable are negative and significant. Taken together, these results support our hypothesis concerning the beneficial impacts of bilateral debt-for-nature swaps on forests.

Next, we turn to other factors that are significantly associated with forest loss. We find that higher levels of population growth correspond with increased wet forest loss. The coefficients are positive and significant in models (2.6) through (2.10). This is consistent with the human ecological perspective which emphasises that ecological factors, such as population growth, act as fundamental drivers of environmental impacts (York, Rosa, & Dietz, 2003).

There are several non-significant findings that we should also mention. First, we find that none of the international factors are related to wet or tropical forest loss. The coefficients for debt repayment, structural adjustment, and bilateral conservation aid do not reach a level of statistical significance in any model. There is one exception in Model 8, where the association between the amount of IMF structural adjustment lending and forest loss is positive and significant. This finding is consistent with political economy theories which suggest that countries may need to exploit their natural resources, such as by exporting forestry products, to increase exports and provide monies to service their debts. Other political factors included in our model, such as democracy, non-governmental organisations, and protected forest area, are not related to forest loss. The coefficients do not reach a level of significance in any models of Table 2. Likewise, gross domestic product and economic growth are not related to wet forest loss.

In Table 3, we report the second stage estimates of dry or temperate forest loss. The modelling strategy in Table 3 is organised in the same way as Table 2. We find that higher levels of conservation funding generated by United States bilateral debt-for-nature swaps financing is related to less temperate forest loss. The coefficients are negative and significant in every odd-numbered model. We also find that higher levels of debt reduction generated by United States bilateral debt-for-nature swaps also corresponds with less forest loss. The coefficients are negative and significant in every odd-numbered model. Finally, we find that higher population growth rates are also related to increased dry or temperate forest loss. The coefficients are positive and significant in every model in which this variable is included. The findings for dry (temperate) forest loss in Table 3 are consistent with those for wet (tropical) forest loss in Table 2. The same international and political economic factors remained non-significantly associated with forest loss.

Table 3. Second stage instrumental variable regression estimate of debt-for-nature swaps on dry forest loss

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model 9	Model 10
Independent Variables
Bilateral Debt-For-Nature Swap Conservation Funding, 1990–2000	−.074*		−.072*		−.100*		−.106*		−.120*
	(.038)		(.037)		(.032)		(.047)		(.059)
Bilateral Debt-For-Nature Swap Debt Reduction, 1990–2000		−.060*		−.058*		−.081*		−.086*		−.100*
		(.033)		(.031)		(.038)		(.039)		(.049)
Debt Service, 2000			.001	.001	.001	.001	.001	.001	.001	.001
			(.001)	(.001)	(.001)	(.001)	(.001)	(.001)	(.001)	(.001)
International Monetary Fund Loans, 2000			−.009	−.004	−.006	−.001	.010	−.016	.012	.020
			(.021)	(.020)	(.023)	(.022)	(.025)	(.023)	(.026)	(.025)
Bilateral Environmental Aid, 1990–2000			.001	.001	.001	.001	.001	.001	.001	.001
			(.001)	(.001)	(.001)	(.001)	(.001)	(.001)	(.001)	(.001)
Non-Governmental Organisations, 2000					−2414.550	−1423.456	−1217.088	−351.688	−4912.59	−4243.357
					(8358.40)	(8071.32)	(8560.335)	(8250.22)	(10,178.36)	(9851.77)
Democracy, 2000					−.031	−.029	−.039	−.037*	−.031	−.028
					(.021)	(.020)	(.021)	(.020)	(.022)	(.021)
Protected Forest Area, 2000					.017	.014	.004	.001	.012	.009
					(.033)	(.144)	(.224)	(.033)	(.037)	(.036)
Population Growth Rate, 1990–2000							.451*	.463*	.418*	.432*
							(.224)	(.215)	(.251)	(.244)
Agricultural Land Area, 2000							.024	.019	.042	.037
							(.047)	(.045)	(.052)	(.051)
Gross Domestic Product per capita, 2000									.025	.028
									(.045)	(.045)
Economic Growth Rate, 1990–2000									−.009	−.010
									(.010)	(.010)
Constant	.138	.134	.010	.081	.198	.168	.049	.036	−.199	−.236
R-Squared	.038	.001	.001	.039	.211	.137	.211	.133	.352	.285
Number of Countries	85	85	85	85	85	85	85	85	85	85
Highest Variance Inflation Factor Score	1.00	1.00	1.12	1.13	1.54	1.46	1.58	1.51	2.39	2.48
Mean Variance Inflation Factor Score	1.00	1.00	1.07	1.07	1.21	1.21	1.28	1.24	1.57	1.57
Anderson Under-Identification Test	15.519***	16.176***	15.478***	16.817***	12.472**	14.003***	12.590**	13.999***	10.091**	10.965**
Sargan Over-Identification Test	4.225*	4.930*	3.878*	4.318*	3.008	3.514	3.393	4.099	3.082	3.694
Cragg Donald Weak Instruments Test	9.157**	9.636**	8.794**	9.742**	6.535*	7.495*	6.433*	7.295**	4.849	5.332*

Notes: * Indicates p < .05, ** indicates p < .01, and *** indicates p < .001 for a one-tailed test. The first number is the unstandardized coefficient, the second number in parentheses is the standard error.

6. Discussion and conclusion

The impetus for this study is to offer clarification about the impact of a particular form of conservation aid, United States bilateral debt-for-nature swaps, on forest loss. We situate our study at the crossroads of two literatures: the theoretical literature regarding the importance of conservation aid to improve the natural environment; and the existing cross-national and case study research that appears to contradict much of the theoretical expectations.

In an effort to resolve the apparent mismatch between the theory and empirical research on the topic, we refine the existing research in two ways. First, we limit our focus to only debt-for-nature swaps involving the United States in an attempt to isolate the effect of this type of conservation aid on forests. The United States is the largest originator of this type of aid, but other countries who use debt-for-nature swaps may have policy intentions or implementations of conservation programmes that differ from the United States. Hence, an analysis that does not isolate policies by a specific donor may yield unreliable estimates due to aggregation of potentially differential effects. Second, we adopt an improved analytic strategy that attempts to address the potential problem of endogeneity. This can arise if the same processes that determine the outcome being measured, forest loss, are also associated with the likelihood of a country receiving a United States bilateral debt-for-nature swap. By using a two-stage instrumental variable model, we can reduce the possibility that the regression estimates are biased due to endogeneity.

In the final results of our regression models after addressing selection bias, we find consistent support for the notion that United States bilateral debt-for-nature swaps are associated with lower rates of forest loss. These findings are robust across wet and dry forests. Our findings contradict much of the cross-national research and supports the theoretical arguments regarding the effectiveness of conservation aid in improving the natural environment. It does not appear that such swaps are limited by their small size relative to the large scale of the problem of forest loss, nor are they rendered completely ineffective because of potential failures to address livelihood issues pertaining to local people and capacity issues of swap participants. However, these remain important issues to address when formulating policy in regard to future use of swaps as a form of conservation.

We offer three interrelated theoretical and methodological implications that follow from the results of our study. First, it is important for scholars, policy-makers, and activists to conceive of conservation aid, not in totality, but rather in particularities in terms of how funding from specific donors are used to achieve specific policy aims. Our focus on the United States bilateral debt-for-nature swaps demonstrates how the impacts of one set of policies may differ from those implemented by other countries or in aggregate, as prior studies have attempted to show.

Second, we agree with Hermanrud and de Soysa (2016) about the importance of using an appropriate statistical method that accounts for potential problems that can arise due to donor selection bias. A failure to account for this form of endogeneity may lead to biased regression coefficients or inefficient tests of statistical significance. By using a two-stage instrumental variable approach, we have greater confidence that the empirical findings reported here are not the result of faulty inference (due to the problem of endogeneity) regarding the underlying theory of aid effectiveness (Easterly, 2006).

Third, we stress to other scholars studying the natural environment that the findings of any empirical analysis have validity in concordance with the underlying reliability of the evidence on which they are based. We use newly available data on forest loss derived from satellite imagery that eliminates the error introduced with countries using different collection methodologies (World Resources Institute, 2016).

By taking into account these three issues – specificity in what conservation aid programmes are assessed, accounting for donor selection, and more reliable estimates of forest loss – we may be able to explain in part the discrepancy between our findings here, which showed a beneficial effect of bilateral debt-for-nature swaps on conserving forests, and prior research which has shown no or adverse impacts of such aid.

We offer some policy recommendations that follow from our study. The results indicate that United States bilateral debt-for-nature swaps appear to reduce forest loss by generating funds for conservation as well as reducing debt burdens. Thus, lawmakers in the United States should reauthorize the Tropical Forest Conservation Act, which has not been financed since 2014, in an effort to reduce forest loss. If we truly want to address forest loss, however, then the programme should be reauthorized and expanded. This could be done by allocating more funds to the programme or by loosening the eligibility criteria to increase the number of nations that qualify for the programme. Of particular note, the United States may want to eliminate the requirement that a nation be under a World Bank or International Monetary Fund structural adjustment loan because these loans have been linked to forest loss themselves (Shandra, Restivo, Shircliff, & London, 2011).

To conclude, we mention two potential limitations of this study and directions for future research. In our study we were forced to rely on cross-sectional regression models because the satellite imagery data on forest loss that we used are not available over multiple time points (Rudel, 2017). When panel data become available, cross-national researchers should re-examine the relationship examined here by integrating country-level and period-specific fixed effects in their regression models. This would allow for more stringent control over invariant characteristics of nations that may affect forest loss and can further ward off potential problems with endogeneity (Wooldridge, 2015). Longitudinal models would also allow social scientists to incorporate the time dimension in terms of the implementation and effectiveness of bilateral debt-for-nature swaps (and other forms of conservation aid). At the same time that we advocate for expanded cross-national quantitative analysis once additional data become available, we argue that it is important to supplement this work with in-depth case studies and other qualitative research ‘on the ground’ to get a better understanding of the ways in which the implementation of such policies are conducted and how this pertains to conservation efforts (Gockel & Gray, 2011). There may be significant local variation, and such in-depth studies would help elaborate on the most effective ways in which the natural environment can be successfully conserved (Cassimon et al., 2011).

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. Our sample includes the following 85 nations: Albania; Angola; Argentina; Armenia; Azerbaijan; Bangladesh; Belarus; Benin; Bhutan; Bolivia; Bosnia and Herzegovina; Botswana; Brazil; Bulgaria; Burkina Faso; Burundi; Cambodia; Cameroon; Central African Republic; Chad; China; Colombia; Congo, Rep; Costa Rica; Cote d’Ivoire; Dominican Republic; Ecuador; Ethiopia; Fiji; Gabon; The Gambia; Georgia; Ghana; Guatemala; Guinea; Guyana; Haiti; Honduras; Hungary; India; Indonesia; Jamaica; Kazakhstan; Kenya; Kyrgyz Republic; Lao PDR; Lesotho; Macedonia; Madagascar; Malawi; Malaysia; Mali; Mauritania; Mexico; Moldova; Mongolia; Mozambique; Nepal; Nicaragua; Nigeria; Pakistan; Panama; Papua New Guinea; Paraguay; Peru; Philippines; Romania; Rwanda; Senegal; Sierra Leone; Solomon Islands; South Africa; Sri Lanka; Sudan; Tajikistan; Tanzania; Thailand; Turkmenistan; Uganda; Ukraine; Uzbekistan; Venezuela; Vietnam; Zambia; and Zimbabwe.

2. We include circular distance from Washington D.C. because nations are closer to the United States geographically may be more likely to align politically with the United States and receive bilateral aid (Lewis, 2000). We include the number of threatened bird species because it captures a country’s need for a bilateral debt-for-nature swap (Deacon, 1994).

3. We also evaluate the impact of the receiving debt reduction via the joint International Monetary Fund-World Bank Heavily Indebted Poor Countries Initiative and Multilateral Debt Reduction Initiative. However, we do not find that higher levels of debt reduction provided to qualifying nations by these programmes are related to less wet or dry forest loss. The coefficients fail to reach a level of statistical significance in any model. We do not present the results for sake of space, but they are available from the authors upon request.

4. We measure the independent variables for 2000 in order to establish temporal ordering of the variables. However, the reviewers correctly pointed out that the independent variables may have contemporaneous impacts on forest loss. Therefore, we estimated the models measuring the independent variables for the most recent year for which they are available (2014). The results are similar to the models that we present here. Notably, the coefficients for conservation funding and debt reduction generated by participating in a United States bilateral debt-for-nature swap are negative and significant across models of wet and dry forest loss. The coefficients for population growth also remain positive and significant. These results are available upon request.