Policy Research Working Paper 11177 Protecting Forests in the Congo Basin An Empirical Basis for Performance-Linked Financing for the Republic of Congo Dieter Wang Philipp Kollenda Veerle de Smit Kanta Kumari Rigaud Tsegaye Ginbo Gatiso Alexander Golub Environment Global Department July 2025 Policy Research Working Paper 11177 Abstract The Republic of Congo, a country with extensive tropical land-intensive activities; however, transitioning from oil forests and low deforestation rates, needs to balance export- dependency requires diversification into the forest sector driven development, especially through timber production, which in turn requires strengthening sustainable logging and sustainable forest management. Despite national practices and more robust institutional frameworks. This commitments to conserve and restore forests, such efforts research provides a model-based benchmark to define key remain underfunded. Empirical analysis shows that histor- performance indicators for deforestation reductions and ical deforestation is closely tied to timber and agricultural to set feasible and ambitious targets for protecting forests export prices, the real effective exchange rate, dry weather, while pursuing diversified economic growth. Transparent and demographic trends. Under a business-as-usual sce- performance indicators along with feasible, but ambitious nario, deforestation could rise sharply without effective targets are a critical for results-based financing instruments. policy interventions. Sangha and Likouala provinces, which They are critical to unlock public and private capital to are rich in undisturbed forests and new concessions, are par- support economic growth and conserve the standing forests. ticularly at risk. Past oil-driven revenues have contributed to This model has relevance to the other Congo Basin and lower deforestation by shifting economic focus away from tropical forest countries with extensive forests. This paper is a product of the Environment Global Department. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at dwang5@ worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Protecting Forests in the Congo Basin: An Empirical Basis for Performance-Linked Financing for the Republic of Congo ‗ Philipp Kollenda, Veerle de Smit, Kanta Kumari Rigaud, Dieter Wang, Tsegaye Ginbo Gatiso, and Alexander Golub World Bank Group JEL codes: Q23, Q56, F34, O13, C52 Keywords: High Forest Low Deforestation (HFLD), Congo Basin, avoided deforestation, key performance indicators, performance-based financing, feasibility and ambitiousness ‗ Corresponding author. E-mail: dwang5@worldbank.org. This paper has been produced in support of the World Bank’s Advisory Service and Analytics (ASA) "Leveraging Natural Capital Accounting and Climate Finance for the Congo Basin Forests" (P180767) under the leadership of Kanta Kumari Rigaud with Tsegaye Gatiso. The authors are deeply grateful to Chakib Jenane, Ellysar Baroudy, Louise Mvono for their overall guidance and Jonathan Coony, Richard Damania, Laurent Damblat, Andres B. Espejo, Gianleo Frisari, Marek Hanusch, David Maleki, James Seward, Robert Johann Utz, Vincent De Paul Tsoungui Belinga and Nabil Chaherli, for constructive and helpful comments throughout the review process. The authors are also grateful for the financial support of the Central African Forest Initiative (CAFI) and the Climate Investment Funds (CIF) under the Congo Basin Program, the financial support from the Global Program for Sustainability (GPS), and to GMV for the geospatial data used in the subnational analysis, supported by the European Space Agency (ESA) through its Global Development Assistance (GDA) program. 1 Introduction 9Almost 70 percent of the Republic of Congo (RoC) is covered by forest ecosystems that form part of the Congo Basin, the world’s second largest tropical rainforest. Most forest ecosystems in RoC are characterized as lowland and submontane forests. However, dense moist forests exist especially in the southwest and north of the country (World Bank Group et al., 2025). The northeast of the country also contains the world’s largest tropical peatland complex (Dargie et al., 2017). Peatlands store large amounts of carbon as organic matter accumulates in waterlogged conditions. RoC’s peatlands are estimated to store between 9.3 billion and 10.42 billion tons of carbon—238 to 288 percent of the country’s above-ground forest carbon stock (Crezee et al., 2022; World Bank Group et al., 2025). Of the world’s large forest ecosystems, the Congo Basin is the only stable carbon sink. Southeast Asian forests are already net carbon sources, and the Amazon is about to become one (World Bank Group et al., 2025). Within the Congo Basin, the RoC accounts for 12 percent of the total forested area (FAO, 2020). The Congo Basin ecosystem is not only a vital carbon sink, but also a critical hotspot of biodiversity and ranked third in the world in terms of species richness (Beekmann et al., 2024; Mittermeier et al., 2003). A recent assessment of the RoC’s forest accounts indicates that from 2000 to 2020, the value of wood and other resources increased, with the most notable growth in carbon retention—a global public good (World Bank Group et al., 2025). Although Congo’s forests remain extensive, lowland forest losses accelerated between 2010 and 2020 due to land use changes driven by farming. Forest conditions, measured by canopy cover, connectivity, tree height, and naturalness index, show a slight decline but generally remain good, except for lowland forests and mangroves. Trends indicate increased wood and fuelwood exports as energy sources, coupled with a decline in bushmeat harvests, reflecting increasing resource depletion. The forest resource assessment also covered tourism, sedimentation regulation, and carbon regulation services (World Bank Group et al., 2025). The total value of the RoC’s natural forests is estimated to have almost doubled from US$1.7 trillion in 2000 to US$3.1 trillion, with a lion’s share driven by climate regulation as a global public good, and by the peatlands which have a high carbon intensity. However, unsustainable exploration of the country’s vast forest resource would significantly endanger the Congo Basin’s role as a global carbon sink and biodiversity hotspot and replace one extractive growth model with another. The sustainable management of agriculture and forest resources presents an opportunity for the RoC to create employment for local communities and leverage additional climate finance (World Bank Group, 2023). Currently, the RoC’s growth model depends on revenues from the oil sector, with oil rents accounting for 34.4 percent of gross domestic product (GDP) in 2021. Broadening the economic base is a stated objective of the government (World Bank, 2023) and could be achieved through engagement in sectors “such as agriculture and forestry, industry, special economic zones, tourism, digital, and housing” (World Bank Group, 2023). The RoC is a high-forest, low-deforestation (HFLD) country and as such faces unique challenges in man- aging forest resources and accessing climate finance. Definitions of HFLD areas differ, but a common definition considers countries with forest cover above 50 percent and average annual deforestation rates below the 10-year global average during that period (Fonseca et al., 2007; Maniatis, 2024).1 Thirty coun- tries meet this definition (FAO, 2020; Maniatis, 2024) with the 10-year global average deforestation rate at 0.22 percent. However, many of them have small land areas or are small island nations. Forest cover in RoC is almost 70 percent with deforestation rates around 0.1 percent of total land area (European Com- mission, 2024). For other countries in the Congo Basin, only the RoC’s northwestern neighbor Gabon qualifies. Equatorial Guinea qualifies under Architecture for REDD+ Transactions—The REDD+ Environ- mental Excellence Standard (ART TREES’) HFLD methodology, but not under the United Nations Food and Agriculture Organization’s (FAO) Forest Resources Assessment (FRA) methodology (see Figure 7 in Simon et al. (2021)). HFLD countries face increasing pressure on forests as economic development and increased state capacity 1 See UNFCCC statement on HFLD climate finance mobilization (Krutu of Paramaribo Joint Declaration on HFLD Climate Finance Mobilization, 2019) 2 enable the exploitation of forest resources. To safeguard the value provided by HFLD forests in terms of carbon retention, ecosystem services, and biodiversity, forest finance is needed at a scale that can offset the rising opportunity cost of conservation. This is a form of transition finance, because the pressure on forests decreases again as countries move to a productivity-led growth model focused on human capital rather than natural resource extraction. The forest transition curve (FTC) conceptualizes this theory of different stages of deforestation (Figure 18).2 However, the FTC fails to account for the loss of ecosystem services and biodiversity that accompanies a decrease in forest cover or subsequent changes in the condition of the forests. These losses cannot be reversed, even if forest cover recovers. It is therefore crucial to provide HFLD countries with sufficient incentives to prevent deforestation.3 The RoC’s overall external debt is classified as in distress following a period of high public expenditure for ambitious infrastructure projects and persistently low oil revenue (International Monetary Fund. African Dept., 2024; World Bank Group, 2023). However, the country is making progress in restructuring its debt and debt levels are assessed as sustainable in the International Monetary Fund (IMF) and World Bank’s 2024 Debt Sustainability Analysis (International Monetary Fund. African Dept., 2024). At end-2023, public debt levels were at 99 percent of gross domestic product (GDP), severely limiting fiscal space and increasing the country’s dependence on oil revenue which comes with significant downside risks. A mix of different climate financing instruments is needed to engage in the country’s debt challenges and to support the country to transition away from an oil dominated economy without accelerating deforestation rates to unsustainable levels (World Bank Group 2023). Additional financing mechanisms are essential to attract investors and to be able to meet the country’s financing needs. RoC has an annual need of US$820 million of climate finance. Yet, it only succeeded in raising 18 percent in 2021 and 8.8 percent in 2022 (Climate Policy Initiative, 2024). Innovative financing mechanisms, like a crediting mechanism for HFLD countries, innovative debt instruments, or jurisdictional reducing emissions from deforestation and forest degradation (JREDD+) programs are needed for forest protection and economic development. In this paper we outline the necessary building blocks for a sustainability-linked financing framework for the RoC. The building blocks consists of three parts. First, a key performance indicator (KPI),4 which measures how much of an observed outcome, such as annual deforestation, can be attributed to factors outside the sovereign’s control and how much to actual issuer performance (Wang et al., 2023). Second, the methodology needed to identify ambitious but feasible targets, given a specific indicator. Both parts are crucial to safeguard the environmental integrity of deforestation-linked carbon assets, as well as to protect issuers and investors from greenwashing accusations. Third, how to mobilize this KPI when seeking to deploy financing instruments. A key requirement for the success of climate and sustainability-related financial instruments is a robust, science-based, and transparent performance framework. The KPI developed and presented in this paper would be suitable for a range of climate finance instruments. For sustainability-linked instruments, such as sustainability-linked bonds (SLBs) or sustainability-linked loans (SLLs), the KPI can form the basis to determine whether contractually fixed increases or decreases in coupon payments (step-ups or step- downs) would be triggered.(Wang et al., 2023) For use-of-proceeds bonds, the KPI can be used as part of a monitoring, reporting, and verification (MRV) system to understand if the delivered outputs led to the expected outcomes. For carbon-emission reduction programs, the KPI could serve as a counterfactual to estimate carbon-credits related to avoided deforestation. In addition to this introduction, the paper has ten sections, including this introduction. Section 2 lays out a strategy to develop a KPI for avoided deforestation and the theoretical foundation of the underlying model of deforestation pressure in the RoC. It discusses the factors influencing deforestation outcomes in the RoC and the geospatial data used to measure deforestation on a yearly basis. Section 3 describes the forestry 2 The relationship described in the FTC is closely related to the concept of the environmental Kuznets curve (Stern, 2004) which argues that economic development first correlates with worsening environmental outcomes before an increased valuation of non-monetary environmental goods relative to incomes leads to a reversal. However, the fast adoption of environmental policies by developing countries has, at least in some areas, led to a significant flattening of the empirical environmental Kuznets curve. 3 This is still possible for the RoC, which has lost only 9 percent of undisturbed forests since 1990. By comparison, the Democratic Republic of Congo lost 21 percent of undisturbed tropical moist forest, the Central African Republic 27 percent, and Cameroon 16 percent (European Commission, 2024) 4 Note that we distinguish between indicators, which can be any data source with sufficient coverage, from KPIs, which are derived from indicators but can be interpreted to identify additionality. 3 data. Section 4 explains the econometric strategy to model and forecast deforestation pressure in RoC and presents the results of the baseline model. Section 5 describes how these results can be used to build a KPI using the REACH model. Section 6 discusses how to set feasible and ambitious targets for a hypothetical performance-linked instrument. Section 7 analyzes the model for the Sangha and Likouala provinces where new logging concessions supporting sustainable forest management were established in the last 15 years. Section 8 analyzes how (offshore) oil production determines RoC’s national income and indirectly shapes forest degradation and deforestation patterns by crowding out more land intensive industries. Section 9 presents the methodology used to undertake a qualitative analysis of financial instruments and their viability for the RoC highlighting the importance of strong performance indicators and KPIs; and presents a preliminary analysis of the feasibility of mobilizing financing through specific financing instruments. Finally, Section 10 concludes and suggests avenues for future research. 1.1 Incentivizing Forest Protection for HFLD Countries The RoC has committed to several conservation and restoration efforts as well as to supporting sustainable forest management practices and reduced impact logging (RIL). These commitments are reflected in the country’s updated NDC (Republic of Congo, 2021) as well as other (regional) projects, such as the Forest Carbon Partnership Facility (FCPF) emissions reduction program , the Bonn Challenge pledge , the Braz- zaville Declaration, and a well-established cooperation with Central African Forest Initiative (CAFI) and the Forest Code (Núñez del Prado et al., 2023). Many policies and law enforcements are in place to contribute to forest conservation, but implementation often lacks progress and the countries’ institutional challenges impede this process. In 2023, the RoC’s government effectiveness estimate was ranked among the world’s lowest at 193 out of 213 with high levels of perceived corruption (Transparency International, 2021). A recent forest resource assessment estimated the value of the RoC’s natural forests at US$3.1 trillion—with a large part coming from the role of forests as a global public good (World Bank Group 2025, forthcoming). However, HFLD countries have not been able to use this valuation to access forest finance on a scale sufficient to offset the economic incentives of increasing deforestation. In part, this has been due to concerns about the environmental integrity of HFLD crediting approaches. The ART-TREES’ crediting approach, for example, allows HFLD countries to credit an upwards adjustment to their historical baseline in relation to their respective remaining forest carbon stock. Whether such an adjustment accurately reflects the increasing pressure on forests for HFLD countries is an ongoing topic of discussion and research (Teo et al., 2024).5 For HFLD countries, the baseline of low deforestation rates causes a particular challenge. Figure 1 shows that Congo Basin countries have high undisturbed forest covers and low historical rates of deforestation, especially Gabon, Equatorial Guinea, and the RoC. In this situation there is little room to further reduce deforestation rates. Similarly, the methodology of extrapolating historical deforestation rates to estimate avoided deforestation, such as via moving averages, unfairly raises the bar for HFLD countries. It only consistently provides rewards when deforestation trends are downwards which means that HFLDs’ low deforestation rates do not benefit the economy. This not only distorts the flow of forest finance to jurisdic- tions with high historical deforestation rates but also undermines its effectiveness and integrity. Because countries like RoC will face increasing deforestation pressure as the economy grows (Megevand et al., 2013), a benchmark based on historical deforestation rates will not generate sufficient funding to counteract the economic incentives for deforestation. Targets based solely on historical deforestation rates can create unintended (perverse) incentives, as govern- ments would be rewarded for reducing deforestation after previously increasing it. “If the world rewards actors for stopping deforestation, there is a perverse incentive to start deforesting in the first place” (Ma- niatis, 2024). This in essence what HFLD countries like the Republic of Congo would need to do to fully benefit from Article 6.6 An alternative is to reward stable and undisturbed forest stock instead of focusing on deforestation (Funk et al., 2019). However, mechanisms focusing on stable forests using this approach have received relatively less attention and finance compared to areas perceived to be in immediate danger. 5 See details of the latest decision by the ICVCM to only approve the ART TREES crediting approach without the HFLD credits in Architecture for REDD+ Transactions (2024) 6 The issue has been recognized by REDD+ and the new ART TREES HFLD credits contain an upwards adjustment to account for the expected rise in deforestation pressure (Department of Public Information, Guyana, 2022) 4 Figure 1: Undisturbed Forest Areas and Deforestation Rates in the Congo Basin (a) Share of Undisturbed Tropical Moist Forest (b) Deforestation Rate Between 2001 and 2023 Extent in 2023 as Percentage of Total Land Area Source: Author’s calculations based on forest data from the Tropical Moist Source: Author’s calculations based on deforestation data from the TMF Forests (TMF) Country Statistics database (European Commission, 2024; Country Statistics database (Vancutsem et al 2021; European Commission Vancutsem et al., 2021) and the Land Area from the World Bank’s World 2024). Total deforestation rates were calculated using the TMF dataset Development Indicators (WDIs) Database ID: AG.LND.TOTL.K2). The and include direct deforestation and deforestation after degradation. The countries are labeled with ISO3 country codes: CAF = Central African TMF dataset only covers tropical moist forests, excluding dry forests. De- Republic, CMR = Cameroon, COD = Democratic Republic of Congo, COG forestation rates were calculated relative to the total undisturbed tropical = Republic of Congo, GAB = Gabon, GNQ = Equatorial Guinea. moist forest cover in the year 2000. One reason could be that it is not always clear to investors or donors how much forest was at risk (either immediately or in the medium-term). Therefore, it is difficult to argue what the additional impact of the investment or grant was on avoiding deforestation. Addressing the difficulty of HFLD countries’ efforts to leverage their vast forest resources to unlock forest finance is an area of ongoing discussion and research. Two recent proposals for methodologies to calculate crediting levels for HFLD countries come from the ART TREES 2.0 HFLD approach and the proposed Tropical Forests Forever Facility (TFFF) (ART 2021). Under the ART TREES HFLD methodologies, countries with deforestation rates below 0.5 percent and forest cover above 50 percent during the entire reference period are allowed an upwards adjustment to their crediting level relative to their total standing forest carbon stock. For the RoC, this corresponds to additional credits for 3,246,000 tons of carbon dioxide equivalent (teCO2) to 3,739,000 teCO2 per year, based on a calculated adjustment factor of 0.025 percent to 0.029 percent.7 Our model attempts to find a middle-ground between payment for forest stock and payment for reduced deforestation by explicitly acknowledging the economic factors that impact deforestation pressure indepen- dently of historical deforestation rates. This provides credible incentives for HFLD countries as estimates of rising deforestation pressure are based on climatological, demographic and economic factors and not on ad-hoc upward adjustments of historical deforestation rates. Deforestation pressure is estimated based on multiple factors that are largely outside of the governments’ control. This reduces the perverse incentive to increase deforestation and makes it difficult to manipulate the counterfactual in some other way to create an easier benchmark for the future. 7 The adjustment factor is the sum of the ART TREES forest score and the ART TREES deforestation score, averaged over the reference period from 2016-2022. Specifically, the formula is: (forest cover – 50) / 100 + (0.5 – deforestation rate). The average forest cover and deforestation rate in the reference period based on the TMF data used in this report are 64.23 percent and 0.1356 percent respectively. The adjustment factor is therefore ((64.23 – 50) / 100 + (0.5 – 0.1356)) * 0.05 percent = 0.02533 percent. 5 Figure 2: Theoretical Framework for Deforestation Causes Source: Geist and Lambin (2002) as visualized in Megevand et al. (2013) 2 Theoretical Framework This paper describes a framework of deforestation pressure as a proof-of-concept with data sources for the RoC. The framework supplies empirical evidence for a possible performance-linked financing instrument to help avoid deforestation and provide a basis for consultations with relevant stakeholders such as gov- ernments, international and national NGOs, development banks, and financial market participants (see identification phase in Figure 11). Our approach is based on the framework introduced by Geist and Lambin (2002) and re-evaluated in the Congo Basin context by Megevand et al. (2013). Geist and Lambin (2002) emphasize the distinction between: (1) proximate causes of deforestation such as shifting agricultural cultivation, infrastructure projects, and commercial logging; and (2) underlying causes including demographic factors, comparative cost advantages, and institutional factors. See Figure 2. In the context of the Congo Basin, the expansion of agriculture and infrastructure (including logging roads) appear to be the most relevant proximate causes (Megevand et al., 2013), together with the increase in logging. Megevand et al. (2013) also highlights the emergence of underlying deforestation drivers related to trade in the globalized economy in addition to population growth. The global demand for commodities, such as oil and mineral extraction, road development, agribusiness, biofuels, and the logging sector, has been increas- ingly contributing to deforestation, directly and indirectly through extension of infrastructure (Megevand et al., 2013).8 In the RoC, observed deforestation occurs in relation to shifting agriculture, slash-and-burn farming, logging and infrastructure development (such as logging roads), including for mining and logging activities. Clearly, understanding these proximate causes is important to explain deforestation outcomes in retrospect and to develop effective measures to support sustainable forest management. However, identifying only proximate causes does not give the full picture or help to build a model of deforestation pressure and forecast how deforestation may develop in the future. Here we focus on the underlying causes of deforestation and how they are projected to change in the future. 8 Increased logging does not necessarily have to lead to unsustainable deforestation and the uncontrollable loss of forest stock. In fact, several certified concessions engage in sustainable logging and reduced impact logging, harvesting well below the natural harvesting rate per species. 6 Table 1: Descriptive Statistics Variable Mean SD Min Max N Forest disturbances Total annual deforestation (ha) 34,870.13 21,240.75 10,655.00 97,934.73 24 Total annual degradation (ha) 63,961.23 28,962.12 24,261.46 149,202.74 24 Climatological factors SPEI-Index (12-month, index) –0.05 0.58 –1.13 0.94 23 Economic factors Real effective exch. rate (2017=100) 99.50 6.86 83.59 107.90 23 GDP per capita, PPP (2017 US$) 4,972.20 963.18 3,687.69 6,932.94 24 Hard sawn wood price (US$ / m3 ) 731.79 125.07 488.46 939.45 24 Tropical roundwood price (US$ / m3 ) 344.54 76.63 192.83 488.00 23 Round- and sawnwood price (2015=100) 85.27 14.86 56.05 109.76 23 Cocoa price (US$ / m3 ) 2,280.98 659.22 903.91 3,258.04 24 Sugar price (US cents / pound) 14.63 5.54 6.24 26.24 24 Coffee price (US cents / pound) 147.50 56.66 61.46 271.11 24 WTI oil price (US$ / barrel) 63.14 24.45 25.93 93.99 24 Demographic factors Urbanization rate (%) 63.95 3.23 58.70 69.19 24 Rural population growth rate (% change) 1.67 0.76 0.80 2.92 24 We identified proxy indicators for the underlying causes of deforestation and tested their relevance for the RoC’s context empirically. The next section presents our selected proxies, their historical correlation with deforestation outcomes in the RoC, and assesses their suitability as variables included in our econometric model of deforestation pressure. 2.1 Factors Contributing to Deforestation This study focuses on factors contributing to deforestation rather than forest degradation.9 We used geospatial data from the tropical moist forests (TMF) dataset (Vancutsem et al., 2021) from the European Union (EU) Joint Research Centre. The TMF dataset reports areas as deforested when there is no visible tree cover anymore within a 30 by 30-meter pixel (that is the disturbance is long-lasting). This is different from temporary forest degradation. Table 1 shows the summary statistics of the candidate variables for the sample period of 2000 to 2023. Table 2 shows an overview of all candidate variables, their expected influence on deforestation, their reporting frequency, and the most recent value and data sources. Section 3 describes the deforestation data in more detail. 9 Underlying causes for forest degradation are partly different from causes for deforestation. We explore the factors influencing forest degradation and their historical relationship to observed forest degradation in an extension of our baseline model (see Appendix) 7 Table 2: Overview of candidate explanatory variables for deforestation and degradation modeling Variable Expected sign Explanation Data Source Frequency Most Recent Value Demographic factors Rural population growth + Rural population growth increases the expansion of UN World Urbanization Yearly 0.8% (2023) rate agricultural land for subsistence farming and the Prospects; World Bank WDI demand for charcoal. ID: SP.RUR.TOTL.ZG Urbanization rate – Urban population less likely to engage in UN World Urbanization Yearly 69% (2023) slash-and-burn agriculture and more likely to use Prospects; World Bank WDI modern cooking methods. ID: SP.URB.TOTL.IN.ZS Economic factors Real effective exchange rate – Appreciation reduces exports and increases imports, Bruegel (Darvas, 2021) Yearly, Monthly 98.2 (2017=100) (2022) (REER) lowering demand for forest products. Export price of industrial, + Higher log prices incentivize logging and forest FAO Forestry Production Yearly 369 USD/m3 (2022) tropical roundwood disturbance. and Trade Global export price of hard + See above. FRED Monthly* 692 USD/m3 (2024) sawn wood 8 Average global commodity + Higher prices incentivize forest conversion to FRED Monthly* – price for sugar, coffee, cocoa plantations. Climatological Factors SPEI Index (12 month) – Wetter conditions reduce wildfire risk and hinder – Yearly 0.98 (2023) logging operations. Singular Events Wildfires in 2016 + Drought from El Niño led to fires causing Mongabay Erickson-Davis, 2016 15,000 ha affected degradation. 2016a Road development of N2 + Roads drive deforestation directly and indirectly by – 2010 onwards – (2010 onward) increasing accessibility. Forest-Related Policies Policies since 2018 (REDD+, – REDD+ and declaration in force since 2018. – 2018 onwards – Brazzaville Declaration) Agreement since 2019 (CAFI – Caps annual conversion at 20,000 ha; peatland Krutu of Paramaribo Joint 2019 onwards – Letter of Intent) protection. Declaration on HFLD Climate Finance Mobilization (2019) Updated Forest Code 2020 – Modernization of legal forestry framework. – 2020 onwards – Updated NDC 2021 – RoC committed to 17.09%–39.88% GHG reductions. Republic of Congo (2021) 2021 onwards – Figure 3: Demographic Factors between 2000 and 2023 in the Republic of Congo The urbanization rate increases steadily to around 71 percent in 2023. Rural population growth was between 2 percent and 2.7 percent from 2004 to 2010 and then decreased significantly to around 1 percent since 2013. Sources: Author’s calculations based on data from the World Bank’s WDIs and TMF (European Commission 2024). Urbanization is the share of urban population relative to the total population (ID: SP.URB.TOTL.IN.ZS), data is measured every five years and interpolated for years in between. Population growth is measured as yearly percentage changes. The database IDs for population growth are SP.POP.GROW (total) and SP.RUR.TOTL.ZG (rural). Deforestation was measured as “total deforestation (direct or after degradation)” from the TMF database. 2.1.1 Demographic factors Unsustainable shifting agriculture due to subsistence farming (such as slash-and-burn farming) is a major driver of deforestation. The expansion of agricultural land has been reported frequently as a proximate cause of tropical deforestation (Busch & Ferretti-Gallon, 2017, 2023; Megevand et al., 2013; Zhang et al., 2002). Around 6.2 million people live in the RoC, mostly along the southern corridor between Brazzaville and Pointe Noire. In addition, the population is growing with rises in annual population of up to 4 percent before 2013 and growing steadily at around 2.4 percent since then, putting more pressure on forests. At the same time, the RoC is steadily urbanizing. The urbanization rate increased steadily to around 71 percent in 2023 and the share of the population living in urban areas steadily increased.10 Meanwhile, rural population growth was between 2 percent and 2.7 percent from 2004 to 2010 and then decreased significantly to around 1 percent since 2013. To proxy these demographic factors, we included the rural population growth rate and urbanization rates as candidate variables in our model. Figure 3 shows how demographic factors have developed in the RoC. The urbanization rate is defined as the share of the total population living in urban areas, where urban areas are defined by national statistical offices. We included the rural population growth rate from the World Development Indicators (WDIs) which are based on World Bank estimates using the United Nations (UN) World Urbanization Prospects 2018 report (United Nations, Department of Economic and Social Affairs, Population Division, 2019). Because total population growth rate and rural population growth rate develop similarly (see Figure 3) we only included the rural population growth rate.11 In addition to the need for more agricultural land, population growth is associated with a rise in the demand for charcoal which increases deforestation pressure (Megevand et al., 2013). Improved access to clean fuels and promoting alternatives to traditional cooking can ease deforestation pressure, which is why reducing the demand for charcoal is an important policy lever. However, the adoption of clean cooking technologies in the RoC is still low, especially in rural areas. It has been steadily increasing (see Figure 249). But, with low overall take-up and little year-over-year changes in the growth rate, the indicator is less useful to explain changes in deforestation trends and it was not included as a candidate variable. 10 See: https://africapolis.org/EN/country-report/Republic%20of%20the%20Congo. 11 Data on total population growth rate is also available from the World Population Prospects, which also provide long-term projections of population growth rates into the future. See: https://population.un.org/wpp/ 9 Figure 4: Export as share of GDP of Republic of Congo and Sub-Saharan Africa Source: Author’s calculations based on data from the World Bank’s WDIs. Export ratios represent the value of all goods and other market services provided to the rest of the world as a share of GDP. Export values are the current value of exports (free on board (FOB) converted to US dollars and expressed as a percentage of the average for the base period (2015). The database ID for export values is TX.VAL.MRCH.XD.WD and NE.EXP.GNFS.ZS for export ratios. 2.1.2 Macroeconomic factors Global trade in agricultural commodities, logging, and oil and mineral extraction affects the value of land and can lead to deforestation or the conversion of undisturbed forest ecosystems to plantations (Hanusch, 2023; Megevand et al., 2013). Export values of the RoC are still low (ranked 104 of 133 in the Atlas of Economic Complexity (Harvard Growth Lab, 2024; Hausmann et al., 2011) but have increased fivefold since 1990 (World Bank Databank, TX.VAL.MRCH.XD.WD). The RoC also has the highest export-to-GDP ratio in Sub-Saharan Africa (56.9 percent) – more than twice as high as the Sub-Saharan Africa average (25.8 percent) (World Bank databank NE.EXP.GNFS.ZS). See Figure 4. Real Effective Exchange Rate Trade competitiveness decreases when the local currency appreciates with respect to a country’s main trading partners because local export goods become relatively more expensive. Therefore, in countries with large, export-oriented agriculture sectors, the real effective exchange rate (REER) is negatively correlated with increased deforestation pressure (Hanusch, 2023; Wang et al., 2023). Even for countries with lower agricultural export volumes, like the RoC, the relationship can matter because an appreciation of the currency also decreases the relative price of imports, making imports (also of non- export goods) cheaper compared to local production. The REER increased between 2000 and 2008 and has stayed relatively constant since then, with a visible decrease since 2021. The biggest trading partners in terms of export volume for the RoC are China (44 percent), United Arab Emirates (16 percent) and India (7 percent). The REER is calculated using the exchange rates of a country’s main trading partners and inflation rates, based on the consumer-price index (Darvas, 2012, 2021). Therefore, it can vary even for pegged currencies such as Congo’s Central African franc (CFAF). We included the REER from the updated Bruegel database (Darvas, 2021) as a candidate variable in our model.12 Global Commodity Prices Rising global commodity prices make the extraction of natural resources or the conversion of land for agricultural production relatively more attractive and increase deforestation pressure (Hanusch, 2023). However, whether a given commodity exerts relevant deforestation pressure depends on the goods’ importance for exports—historically but also in the future. Even if the current total export volume of agricultural commodities is small for the RoC, this may change in the future under an 12 We used the REER from Darvas (2021) instead of the IMF or World Bank estimates as they are available for a longer historical period. For the period where the REER was reported by the IMF and Darvas the estimated values were very similar. 10 Figure 5: Real effective exchange rate between 2000 and 2022 Source: Author’s calculations based on data from Darvas (2021) and TMF (European Commission, 2024) The REER is indexed at 100 in the year 2007. An increase reflects an appreciation of the currency. extractive growth model as new deforestation frontiers are opened and undisturbed forest converted to agricultural land. The biggest export good for the RoC is crude oil, which accouned for over half of the total export value in 2021 (Harvard Growth Lab, 2024). Oil is mostly extracted through offshore operation, and its influence on deforestation is therefore only indirect through oil revenues contribution to GDP and by channeling resources away from other sectors. We explored the role of oil revenue in shaping the structure of the Congolese economy and its indirect influence on deforestation pressure in Section 8. For our baseline model, we did not include the oil price as a candidate variable because there is no direct relation with deforestation.13 Agricultural exports represent about 10 percent of non-oil export volume with tropical hardwoods ac- counting for more than 80 percent (Harvard Growth Lab, 2024). The export of timber products and the related logging operations can influence deforestation, both directly and indirectly (such as through the development of logging infrastructure, road development, and rural migration). In the Republic of Congo, the Okoumé and Sapelli trees are the most logged species and account for over 50 percent of total logging production in 2017 and 2018 (Eba’a Atyi et al., 2022). All species in the production top 10 are hard wood species. Therefore, we included the price of industrial non-coniferous tropical roundwood14 and of hard sawn wood as candidate variables in our model. Logging activities in the Congo Basin are directly related mainly to degradation rather than deforestation (Megevand et al., 2013). The exception is the development of logging infrastructure (such as roads) or when logging operations happen informally in an unsustainable way, sometimes in areas that were supposed to be closed off after official logging operations took place. Small-scale logging is also a factor contributing to forest disturbances in the important peatlands of the Cuvette Centrale (Nesha et al., 2024), although the total net decrease in peat forest ecosystems between 2000 and 2020 is still minimal with only 8,225 hectares (see Figure 3.3: The Change in Extent of Ecosystem Types, Measured in Hectares, in the Accounting Years, 2000, 2010, and 2020 in World Bank Group et al. (2025)). Overall, the case of wood products is complex in the Republic of Congo. In 2011, an estimated 70 percent of harvested timber was illegal (Hoare, 2015). However, tackling the problem of unauthorized production and trade was one of the primary challenges in the National Development Plan (NDP) 2018-2022 and there has been a considerable push for certification, sustainable forest management plans, and concession models 13 Estimating the model including the oil price showed a non-significant association with deforestation and its exclusion only had a small effect on the estimated elasticities of the other included variables. 14 We calculated the Congo-specific average price of exported roundwood using the reported export value divided by the reported export quantity. The data is available from the FAOSTAT-Forestry database and relies on a mix of official figures, figures reported by international organizations and estimates. See: https://www.fao.org/faostat/en/#data/FO 11 Figure 6: Gross Export Basket and Agriculture Sector Isolated, 2021 Data shows net trade flows at the four digit HS 1992 level for 2021 from the Harvard Atlas of Economic Complexity (Harvard Growth Lab, 2024) accessed in October 2024. Highlighted product categories added by authors. Figure 7: Global Export Prices for Wood Products Source: Global price of hard sawnwood: Federal Reserve Economic Data (FRED). Global roundwood price: Industrial roundwood, non-coniferous tropical from FAOSTAT. Because roundwood and hard sawn wood prices are highly collinear, we included an average of the two in our estimation. Deforestation was measured as “total deforestation (direct or after degradation)” from the TMF Country Statistics (European Commission, 2024). that align with reduced impact logging. We engage in detail with concessions in Section 7 where we focus on the two provinces of Sangha and Likouala in northwestern RoC where sustainable forest management concessions opened in the last decade. Global export prices for wood products peaked between 2010 and 2015 and have since decreased slightly. Figure 7 shows the development of prices for wood products over time, where hard sawn wood refers to the global export price and roundwood is the export price for RoC according to FAOSTAT. Because prices for roundwood and hard sawnwood are highly collinear we included an average of the two in our estimation. Other agricultural export goods are cocoa (8.3 percent), coffee (2.5 percent), and sugarcane (previously around 10 percent) (see Figure 250). Cocoa is argued to be one of the biggest potential factors leading to the disturbance of primary forest if the production is done unsustainably (World Bank, 2024). However, sustainable production of cocoa (and other perennial agroforestry crops) can also serve as an alternative for slash-and-burn agriculture, reducing deforestation pressure. Because non-wood agricultural commodities are a small percentage of Congo’s exports, we included an aggregate cash-crop commodity price as a candidate variable in our model instead of each commodity price separately.15 Figure 8 shows the price development of these agricultural commodities over time. 15 Technically,we took the first principal component of the time series for cocoa, coffee, and sugarcane prices. This first component reflected the trends that are common to all three commodities. 12 Figure 8: Global Export Prices for Sugarcane, Coffee, and Cocoa Because non-wood agricultural commodities are a small percentage of Congo’s exports and are collinear, we included an aggregate cash-crop commodity price as a candidate variable in our model. Source: Global prices for sugar, coffee and cocoa commodities from FRED. Deforestation was measured as “total deforestation (direct or after degradation)” from the TMF country statistics (European Commission, 2024). The second largest individual export for the RoC after oil is copper—26.4 percent in 2021 (Harvard Growth Lab, 2024). While mining does not lead to direct deforestation, it can lead to indirect deforestation through the development of roads, mining sites, and rural population growth. Figure 251 shows the price devel- opment over time for copper and gold. Because there is no direct relationship to deforestation, we did not include copper as a candidate variable in our model. In addition, we did not include gold because it constitutes only a small share of overall exports (0.33 percent in 2022, Harvard Growth Lab (2024)). Some recent deforestation events in Sangha province have been associated with gold mining, but they are too recent to be visible in our data. Nevertheless, in a possible application of the model, it is important to reevaluate whether gold mining emerges as a significant factor influencing deforestation pressure in the RoC. 2.1.3 Climatological factors Dry climatic conditions and droughts lower the yield of agricultural crops and raise the risk of wildfires. Wildfires and dry conditions are mostly associated with forest degradation rather than deforestation (Vancutsem et al., 2021). However, an extension of the dry season also allows logging to continue for longer throughout the year as roads remain accessible in the absence of prolonged rains. Hence, we included the Standardized Precipitation and Evaporation (SPEI) index (Beguería, 2022) to identify abnormally dry conditions. The SPEI index measures rainfall, so dry conditions are defined as negative values of the index. 2.1.4 Singular events and forest-related policies We included specific singular events that had a large historical impact on deforestation. We did this to improve the precision of our estimates and to prevent wrongly associating the effect of a singular event to an underlying cause. These singular events can be climate events such as the 2016 wildfires in the Sangha province or large infrastructure projects such as the paving of the N2 between Brazzaville to Ouesso from 2010-2013. In the modeling of deforestation pressure, these events were modeled as dummy variables that allow for a level-shift in deforestation outcomes in the specified years. 2.1.5 Forest-Related Policies We also included policy dummies to indicate a change in the policy environment and in institutions governing sustainable forest management. For the RoC during the period of 2000-2023 we included the National REDD+ Strategy and the Brazzaville Declaration, the CAFI Letter of Intent Central African Forest 13 Initiative and Republic of Congo, 2019 and the update Forest Code (Núñez del Prado et al., 2023). Different from singular climate events, we modeled these policy changes as staying in effect after their adoption.16 3 Forestry Data Remotely sensed geospatial data has the advantage of providing a transparent external measure of defor- estation outcomes available at a granular temporal and spatial resolution. Satellite imagery is also available with a minimal time delay, meaning that performance can be assessed quickly and delays in payments can be prevented. Additionally, many datasets follow a harmonized methodology across different coun- tries, further increasing transparency and accountability of the KPI modeling process. Of course, remotely sensed data still need to be validated using field surveys and consultations with forestry experts familiar with the local context. Several data sources exist for global or regional tree and forest cover usually reporting annual changes based on disturbances visible on satellite imagery. The difference between tree and forest cover is important in setting goals and monitoring changes. While tree cover counts any woody vegetation meeting a certain height and density threshold, whether natural or artificially planted, forest cover usually has a stricter definition that sets restrictions on the land use of the tree area. This means that plantations or tree crops would count as tree cover but often do not meet forest criteria. 3.1 Tropical Moist Forests The baseline specification uses deforestation data reported by the TMF dataset from the EU Joint Research Centre (European Commission, 2024). The TMF reports annual disturbances in the humid tropics since 1990, where no specific thresholds are used to define forest cover. Instead, the extent is based on closed primary and secondary forests in the humid tropics (excluding dry forests, young secondary forests and plantations). An important advantage of this dataset is that it monitors disturbances not only at the time they occur, but also in the years that follow. Therefore, it is possible to categorize short-term and long-term disturbances, categorizing less intense disturbances as forest degradation. A disturbance is reported when there is no visible tree cover within a 30 by 30-meters pixel, where all available Landsat images are surveyed. Such a pixel in one of the images of a certain year is enough to classify a disturbance.17 The TMF data is available on an annual basis with a long historical period, collected systematically for all tropical forest countries and publicly available. Deforestation data released by the government or national statistical offices are not always consistently available on an annual basis and do not lend themselves to time series analysis. A data validation exercise showed that deforestation reported by the TMF aligns well with official data for the period before 2010 and after 2017. See Figure 9. Between 2011 and 2016 satellite-based deforestation rates were higher than what was reported by the FAO’s FRA ((FAO, 2020). However, we noted that the FRA only provides three data points between 2000 and 2020 with minimal variation. The TMF data categorizes disturbances into degradation, deforestation after degradation, and direct de- forestation, as well as conversion to plantations and forest regrowth. Deforestation reflects concrete land cover changes, while degradation shows temporary disturbances in the tree cover, for example due to fires, extreme weather conditions, and selective logging. The TMF only covers changes in tropical moist forests, which represent 97.7 percent of total forest area in Congo. Dry forests extend the remaining 2.3 percent and are not included in the dataset. The TMF reports between 2000 and 2023 disturbances on 2.36 million hectares previously undisturbed forest cover, of which 0.72 million hectares is classified as deforestation (direct or after degradation). See Figure 9. There is a strong connection between the observed transitions of undisturbed forest cover to degraded and 16 We also present results based on a model without any policy dummies. Note that the policy dummies were only used in the estimation phase to isolate the effect of demographic, economic, and climatological factors. The policy dummies were not used to build the KPI in the evaluation phase as sustaining or improving the positive effects of policies constitutes policy efforts of the issuer that should be rewarded. 17 The indicator used was “total deforestation (direct or after degradation)”. We normalized deforestation using the undisturbed tropical moist forest cover in the year 2000 to facilitate easier comparison between different subnational areas or other Congo Basin countries. 14 Figure 9: Annual Deforestation Rates and Reference Rates This figure compares the data from Tropical Moist Forests (TMF) and Global Forest Watch (GFW). We note that the FAO data consists of only three data points with minimal variation that supposedly cover the period shown in the graph. Source: Annual data is from the TMF data showing “total deforestation (direct or after degradation)” (European Commission, 2024) from Global Forest Watch (GFW) showing tree cover loss with a 30 percent canopy density (Hansen et al., 2013). Officially reported deforestation rates according to Congo’s forest emission reference levels (Niveau d’emissions de reference pour les forets NERF) see Republic of Congo (2017) and the FAO FRA (FAO, 2020) are shown for comparison. Table 3: Definitions of Forest Cover Source Forest Definition RoC (Government) Tree height of 3 meter at maturity, canopy cover at least 30% and minimum area 0.5 hectare. FAO FRA Tree height of 5 meter at maturity, canopy cover at least 10% and minimum area 0.5 hectare. GFW Tree height of 5 meter at maturity, canopy cover at least 30% (data also available for other canopy covers). Changes detectable at 30 × 30-meter pixel. TMF Tree cover in the humid tropics, no minimum height or canopy density threshold. Changes detectable at 30 × 30-meter pixel. Differentiates between forest cover transitions into degradation, deforestation after degradation, direct deforestation, conversions to plantations and forest regrowth. Note: Tree height at maturity is above 3 meters in all RoC’s forests and canopy cover density is above 50 percent in almost all tropical moist forests globally. Therefore, in practice, the data by TMF also conforms to the same tree height and canopy cover density thresholds as the other definitions. deforested land. The TMF data shows that in the majority of land cover transitions in RoC, a deforestation record was preceded by forest degradation in the preceding years. A KPI that uses forest degradation as an indicator would therefore not be fundamentally different, but we prefer deforestation to degradation as it more accurately tracks the long-term loss of forest cover, ecosystem services, and biodiversity that we want to capture. The TMF data was most suited for our analysis despite the differences with the RoC’s own data, its lack of data on land cover transitions, and no sub-annual observations. For example, the RoC government applies a canopy density threshold of 30 percent and a tree height at maturity of minimum 3 meters to define tree cover. This differs from the TMF definition which has no tree height requirements. However, because tree heights in all of RoC’s forests exceed 3 meters, the height requirement is not practically applied. More customizable geospatial products such as those developed by the European Space Agency (ESA) and the World Bank are able to identify which land-use replaced undisturbed forests and can shed more light on the underlying causes of deforestation (Borlaf-Mena, Chunet, et al., 2025; Borlaf-Mena, Mérida-Floriano, et al., 2025). It also allows tracking sub-annual changes which helps differentiate factors that are relevant throughout the year, such as charcoal demand, from seasonal patterns, such as logging operations. 15 3.2 Alternative Forest Data Sources Global Forest Watch (GFW) provides worldwide, satellite-based data on tree cover loss from 2001 to 2023, updated annually (Hansen et al., 2013). GFW collects various forms of disturbances in one aggregated measure called tree cover loss. No separation is made between small collections of trees and forests, but forests require a minimum area of 0.5 hectares tree cover. GFW applies a threshold of five meters tree height and 30 percent canopy density18 , where a loss is reported when more than half of the 30-meter resolution pixels are cleared from tree cover. GFW also includes dry forests, provided the trees meet the tree cover and canopy density thresholds. The base extent of tree cover is measured in the year 2000. For RoC, a canopy density of 75 percent (also reported by GFW) may better represent the dense tropical forest present in the Congo Basin. Using this definition, GFW reports a tree cover loss of 600,000 hectares between 2000 and 2023, indicating a 2.8 percent decrease. However, these losses include plantations or tree crops. Table 3 shows different definitions of tree height and canopy density used by various publicly available datasets. To identify forest cover, datasets also usually contain a minimum tree area. The RoC uses 0.5 hectare, consistent with the definition used by the FAO’s FRA (FAO, 2020). However, the FRA is only available at 5-to-10-year intervals and not sufficient for this analysis. The minimum detectable area is 0.09 hectare for GFW and TMF, but no minimum areas are defined to identify forests. While the definitions are different, we still need to rely on the global datasets from GFW, TMF and others, given their global comparability (Sims et al., 2024). To assess the feasibility and ambitiousness of a KPI and its targets for the RoC, we need to compare historical data with that of peer countries. Country-specific data, like national forest inventories or ecosystem extent accounting, can help reconcile global datasets but is not sufficient for the financing framework. Additionally, in the context of the RoC’s forests, tree height at maturity is above 3 meters for all species and almost all tropical moist forests have canopy cover density above 50 percent. Therefore, the tree height and canopy cover density requirements are satisfied in RoC’s forests, even if not explicitly stated in the TMF data. 4 Empirical Framework This research uses the relative evaluation and benchmarking (REACH) and feasibility and ambitiousness (FAB) framework by Wang et al. (2023). The fundamental idea is that sustainable financing instruments should focus on performance rather than outcomes when evaluating sustainability targets. Using the REACH approach, the observed outcome is compared against the KPI benchmark to identify the issuer’s performance. In this way, an issuer can be rewarded when it outperforms the projected benchmark level and be penalized if it fails to meet the benchmark level. The model makes it possible to disentangle policy efforts from external effects. In case of extreme droughts, for example, a government should not be punished for higher natural disturbance rates if these do not lead to long-term deforestation. However, when a new forest policy is followed resulting in decreasing deforestation, the government should be rewarded for its performance. The purpose of the REACH model is to construct a benchmark, and causal interpretations are not always warranted. The model is used to see to what extent exogenous factors can be used to predict deforestation pressures and consequently provide incentives to avert it. The model can give an indication of what the drivers for deforestation in Congolese forests are, but good predictors are not necessarily the main deforestation or degradation drivers. The coefficients do not translate directly into causalities and should not be interpreted as such. The FAB methodology serves to assess the feasibility and ambitiousness of sustainability targets. Feasibility uses historical data, considering achievements of peer countries to evaluate if there is historical precedence for reaching the goal. The ambitiousness assessment projects a future business as usual (BAU) scenario to identify a deforestation trajectory that is likely without additional efforts or financing—as a benchmark to identify room for improvement. Ambitious targets are then those that meaningfully improve upon the BAU scenario. 18 The default canopy density for GFW data is 30 percent, but data is available for thresholds between 10 and 75 percent. 16 4.1 Econometric Model The REACH model uses an econometric approach to model a baseline scenario for deforestation pressures. It uses a linear specification (see Equation 1) and applies regularization to identify generalizable relationships that can be used to make near-term predictions. = + − + − + | ≥ + (1) =1 =1 =1 ∈ In Equation 1, is the outcome variable of interest—the annual deforestation rate from the TMF in the baseline model. is the intercept and is the autocorrelation coefficient of lag . − are the candidate variable that proxy factors influencing deforestation. Because their influence may not be immediate, we included up to two lags for all candidate variables in Table 2, except for the SPEI index and the singular event dummies. Commodity prices represent the yearly average, so we only included lags (one or two) to prevent including changes in prices that occur after the measured deforestation took place within the same year. | ≥ are dummies for singular events. These included policy changes, road construction, and wildfire events. The dummy variables were included to estimate generalizable elasticities of the remaining factors with respect to deforestation without spuriously associating the effect of singular events to simultaneous change in the explanatory variables. Due to the lack of historical deforestation data for the RoC, the model used a time series of 24 annual deforestation values from 2000 to 2023. Even if more historical data was available, the relationship between external factors and deforestation from that time in the past has limited informational value to predict future deforestation pressure. As shown in the next paragraph, the estimation procedure accounted for the small sample size to some extent via regularization and bootstrapping. However, an application that goes beyond the proof-of-concept presented in this paper could increase the effective sample size by: (1) modeling deforestation pressure for various HFLD countries in a panel framework; (2) using sub-annual data to account for seasonal patterns; or (3) disaggregating the deforestation rates based on the post- deforestation land use. The full list of candidate variables including their lags and singular event dummies was large relative to the length of the historical time series. With many parameters relative to data points, the model would be prone to overfitting—meaning it would fit the estimation period exceptionally well but contain limited generalizable information for future projections. We addressed the risk of overfitting with two strategies. First, we built up our full model slowly, starting with the most minimal specification. Second, we always used a variable selection method to do a data-driven pre-selection of those parameters that hold the most predictive power. Here we used elastic net (Tibshirani, 1996; Zou & Hastie, 2005) and where the penalty strength is determined imposed regularization penalties on the coefficients and through cross-validation. 19 We reported the adjusted coefficient of determination (adjusted R2 ) as the goodness-of-fit statistic for all results. 4.2 Empirical Results Table 4 shows the results of our baseline model, estimating how demographic, economic, and climatological predictors of deforestation are related to historical deforestation rates in the RoC between 2001 and 2023. We construct the full model incrementally over four specifications to examine its robustness and understand the effect of expanding the model. The main feature of the regularization step is that only the most predictive variables are included in the model. Less predictive variables that are not selected are indicated with n.s. (not selected). The resulting coefficients were also penalized towards zero to avoid overfitting the model to historical data. Model 1 only included a lagged dependent variable, the SPEI index, REER, and wood prices. The selected variables had the ‘expected sign’ wetter climatic conditions and an increase in the REER was associated with 19 We partitioned the time series into five folds, each using an expanding training set and a shrinking test set to preserve temporal ordering. Unlike random K-fold splits, this approach respects the data’s autocorrelation. 17 reduced deforestation.20 Increases in global export prices within the past two or three years for hard sawn wood and for industrial tropical roundwood were associated with an increase in deforestation. Because the estimation was conducted using a regularized linear regression, the statistical significance was calculated using a residual bootstrapper in a post-selection inference step, which differs from conventional parametric inference. Model 2 added a measure for the global agricultural commodity price cycle, proxied by a basket of sugar, cocoa, and coffee. These commodities contributed a small but not negligible share to RoC’s exports throughout the sample period. The estimated coefficients for the second lag were positive and significant, suggesting that increased commodity prices were associated with increased deforestation pressure with a time lag. We also noted that the inclusion of an average commodity price variable reduced the estimated coefficients for wood prices, suggesting that wood prices and cash-crop commodity prices had similar effects. Model 3 extended the model with urban and rural demographic factors. The results indicated that a higher urbanization rate—the proportion of the total population residing in urban areas—was consistently associated with a reduction in deforestation across all three temporal lags. This finding aligns with the hypothesis that urbanization may reduce direct reliance on subsistence agriculture or small-scale forest clearing, either through rural outmigration or shifts in economic activity. Conversely, rural population growth is positively and significantly associated with deforestation pressure, supporting the view that increasing rural populations increase demand for agricultural land, fuelwood, and settlement expansion. Model 4 included dummies to control for infrastructure projects, weather and policy events. This model contained all candidate variables described in Table 2 and constituted the baseline model for the remainder of this article (see also results in Table 5). For national-level data, the dummy variables for infrastructure projects and policy events did not explain a sufficient share of the variation and were therefore not included in our model after the regularization step.21 In the case of forest-related policies, this was likely because they only covered a small part of our sample (after 2018) and it was the implementation over time that led to measurable changes in deforestation outcomes, rather than the act of signing the policy agreement. However, in subsequent sections on sub-national deforestation outcomes (Section 5) the adjustment for singular events improves the model fit significantly. The climatological effects and the REER were extremely stable throughout the inclusion of additional factors influencing deforestation. Wood prices were associated with increased deforestation rates in all models, but the inclusion of other commodity prices and demographic factors reduced their estimated effect size.22 The magnitude of the coefficients can be interpreted as a x standard deviations increase on deforestation rates for a one standard deviation increase in the respective regressor. Having estimated a model for deforestation pressure we could now use the estimated coefficients to calculate the associated benchmark level for deforestation pressure in the RoC. In the evaluation phase, the KPI was defined as the difference between the benchmark level of deforestation pressure and the observed deforestation. Forest-related policies were not included in the calculation of the benchmark level, so the effect of their implementation was reflected in the KPI and any observed deforestation significantly different from the benchmark could be attributed to policy interventions (or the lack of). The deviations for the time series used to estimate the model are presented in Figure 10. We calculated deforestation pressure without using the coefficients of the forest related policies, as these were sources of differences between the model and the observed outcome that were connected to performance rather than exogenous factors. The height of the bars shows observed deforestation rates, while the color indicates how the observed data relates to the estimated deforestation pressure. The figure compares expected deforestation rates (blue distributions) based on the model of deforestation pressure with observed defor- estation rates (colored bars) from the TMF dataset. Observed deforestation rates below the counterfactual 20 All candidate predictors (except singular events and policy dummies) were standardized to zero mean and unit variance before model fitting, so each coefficient represents the effect of a one-standard-deviation increase in that regressor. 21 The coefficients for the included variables were slightly different between Model 3 and 4 because adding the indicator variables leads to different results in the regularization step and hence different weights for the included variables. 22 While the magnitude was smaller on a national level, when we estimated the model separately for the Likouala and Sangha provinces where logging is more relevant, the estimated effects for wood prices remained large. 18 Table 4: Results of Baseline Models for Deforestation Pressure Model 1 Model 2 Model 3 Model 4 Variable Climate and + Agroecon. + Demographic + Policies and econ. factors factors factors events Lagged relative deforestation rate, t–1 0.234** 0.209** 0.106** 0.099** Climatological factors SPEI-Index (12-month), t –0.267** –0.257** –0.223** –0.225** Economic factors Real effective exch. rate (%Δ), t–1 –0.230** –0.193** –0.194** –0.197** Real effective exch. rate (%Δ), t–2 n.s. n.s. n.s. n.s. Round- and sawnwood price (US$), t–1 0.145* 0.068** n.s. n.s. Round- and sawnwood price (US$), t–2 0.341** 0.232** 0.220** 0.219** Cash-crop prices (US$), t–1 . n.s. n.s. n.s. Cash-crop prices (US$), t–2 . 0.222** 0.550** 0.569** Demographic factors Urbanization rate, t . . –0.090** –0.095** Urbanization rate, t–1 . . –0.095** –0.102** Urbanization rate, t–2 . . –0.099** –0.107** Rural population growth rate, t . . n.s. n.s. Rural population growth rate, t–1 . . 0.043** 0.039** Rural population growth rate, t–2 . . 0.057** 0.053** Forest-related policy and singular events El Niño / wildfires 2016, t . . . n.s. N2 Road Project 2010 onwards, t . . . n.s. Policy dummies No No No Yes (n.s.) Adjusted 2 0.55 0.54 0.61 0.62 Note: n.s. = not selected: The variable was included as a candidate but not selected during the regularization step. Significance levels were based on residual bootstrapping, showing significance at a 5% (**) and 10% (*) level. Model 4 allows for policy dummies for the national REDD+ strategy and the Brazzaville declaration (2018), the CAFI letter of Intent (2019) and the updated Forest Code (2020) to be included in the estimation. However, because they do not explain a sufficient share of the overall variation they were not selected in the data-driven regularization step. The adjusted R2 accounts for the number of included regressors after the regularization step. Because the policy dummies were included but not selected, the statistic was negative for Model 4. The dependent variable was “total deforestation (direct or after degradation)” from the TMF (European Commission 2024). are colored green, those above red, and those that are not significantly different are colored blue. The re- sults showed that deforestation was higher than expected between 2010 and 2013 and lower than expected between 2007 and 2008, 2014 and 2015, 2017 and 2023. Observed deforestation outcomes have decreased significantly after 2014 and have so far remained at low levels although deforestation pressure increased since 2021. Our model would have predicted an increase in deforestation rates in 2022 and 2023 (see blue distributions in Figure 10) but observed deforestation rates are still at a 20-year low according to the TMF data. Observed deforestation in the 2010-2013 period was larger than expected, but the difference was only significant in 2013. The high commodity prices in that period (see Figure 7 and Figure 8) led to a significant upward adjustment of estimated deforestation pressure, meaning that the country would not have been penalized for worse-than-expected performance (except for in 2013). An indicator based solely on the observed outcomes in the previous years (such as a ten-year moving average) would have resulted in a much lower benchmark and consequently a much more pessimistic view on RoC’s performance in this period. 4.3 Extension: Forest Degradation Table 11 shows that when using TMF’s measure of short-term forest degradation as the outcome variable of interest, climatological and economic factors remain important but demographic factors are not. The forest degradation measure is more susceptible to detect short-term and temporary disturbances in forest cover (e.g. fires, extreme weather conditions, selective logging) and it is therefore not surprising that the effect of the climatological factor is stronger than for deforestation. Table 11 also shows that an increase in the REER is associated with lower forest degradation and that higher global wood prices are associated with 19 Table 5: Results Full Baseline Deforestation Pressure Model Variable t t–1 t–2 Relative Total deforestation (direct or after degradation) . +0.099 . Real effective exchange rate (%Δ) . –0.197 . SPEI-Index (12-month) –0.225 . . Roundwood and sawnwood price (US$) . . +0.219 Cash-crop commodity prices (US$) . . +0.569 Rural population growth rate . +0.039 +0.053 Urbanization rate –0.095 –0.102 –0.107 Note: The table shows the estimation results for the preferred model of deforestation pressure (Model 4 in Table 3). The columns refer to the lagged values that are included in the model. For example, the SPEI index is only included as a contemporaneous effect, while the effect of wood and cash-crop commodity prices are only significant with a two-year lag (the one-year lag was included as a candidate variable but not selected during the regularization step). Because the regressors are standardized before the estimation, their effect sizes can be compared to each other and interpreted as a x standard deviation increase in the annual deforestation rate for a one standard deviation increase in the respective regressor. higher degradation. For both factors the estimated magnitudes are comparable or slightly stronger than the effect estimated for deforestation. The prices of other agricultural commodities do not have an easily interpretable association with degradation with the first lag being negative (or not selected) and the second lag being positive. Demographic factors are included in Columns 3 and 4 but not selected by the model during the regularization step and hence do not explain a significant share of variation in degradation rates. The results for forest degradation are a proof-of-concept to show that the strategy used in this paper for deforestation can also be used for other types of forest loss. The exact definition of the outcome variable is ultimately a decision by the authorities and should reflect the type of forest loss that the government commits to reduce. Forestry experts should then be consulted to perform a similar analysis of contributing factors to identify the candidate variables for the model, similar as what we have done for deforestation. In the RoC, degradation and deforestation are closely related, and the TMF data shows that most recorded deforestation is preceded by degradation in our sample. 20 Figure 10: Modeled Deforestation Pressure Compared to Observed Outcomes This figure compares expected deforestation rates (blue distributions) based on the model of deforestation pressure) with observed deforestation rates (colored bars) from the TMF dataset. Deforestation was higher than expected in 2010-2013 and lower than expected in 2007-2008, 2014-2015, 2017 and 2023. Source: “Total annual deforestation (direct or after degradation)” from TMF (European Commission 2024) relative to the total area of undisturbed forest cover in 2000. Notes: The modeled deforestation pressure is based on the full model (Model 4). See Table 4. The bars plot the observed deforestation rate. We also normalized deforestation using the undisturbed tropical moist forest cover in the year 2000. The blue distributions represented the predicted deforestation pressure due to demographic, economic, and climatological factors. Confidence intervals around the predictions were based on a residual bootstrapper. The colors of the bar represent whether deforestation outcomes were higher or lower than expected given the modeled deforestation pressure. 5 Developing KPIs with REACH Based on the empirical model results we constructed a KPI for avoided deforestation that explicitly accounts for forecasted deforestation pressure due to changing macroeconomic conditions. In contrast to indicators based solely on backwards-looking data such as moving averages of historical deforestation rates, this strategy considered the fact that HFLD countries have large areas of undisturbed forest and historically low deforestation rates. Yet, they also face increasing pressure on forests as economic development and macroeconomic conditions make exploitation of forest resources more attractive. The KPI development can be separated into three phases: (1) identification phase for key model parameters; (2) estimation phase for modeling deforestation pressure; and (3) evaluation phase to calculate performance. See Figure 11. The core of the KPI is an econometric model that describes how, historically, observed deforestation was related to key drivers, such as macroeconomic variables, demographics and extreme weather conditions, accounting for singular contributors to changes in deforestation rates such as policy changes, infrastructure projects or logging concessions. This model is described in greater detail below. In the first phase (the identification phase), stakeholders’ engagement is key. Based on the data and information available, stakeholder input is solicited to agree on the key parameters of the model. For example, the suitability of the geospatial deforestation dataset needs to be validated with national sources and experts, including forestry experts, and, where possible, field studies. The variables that are included to model deforestation pressure and the suitability of other datasets is determined for the local country context. When the KPI is defined on a national level, stakeholders need to agree, for example, on how existing projects (such as the FCPF, REDD+, and private carbon certification initiatives) are incorporated. This is to avoid double-counting of avoided deforestation and prevent sovereigns from being penalized for exhausting logging quotas that are certified as sustainable under reduced impact logging schemes. 21 Figure 11: Flowchart of KPI Development and Measurement In the second phase (the estimation phase), a model of deforestation pressure is estimated using historical data and the key parameters agreed upon after stakeholder consultation. Here, the variables that have been identified as factors influencing deforestation are related to historical deforestation rates. For example, how have deforestation rates developed historically in years following increases in global timber prices? To isolate the influence of factors that are outside of the issuers control as precisely as possible, the estimation phase also accounts for historical singular events that led to large changes in deforestation rates. For example, large infrastructure projects can lead to visible deforestation in geospatial data and should not be attributed to changes in macroeconomic conditions. In the third phase (the evaluation phase), the estimated model coefficients are used to calculate expected deforestation pressure. This generates a model-based estimate of deforestation pressure during the tenure of the financial instrument given the policy environment during the years covered in the estimation phase. The KPI is defined as the difference between these estimates and observed deforestation rates. A positive KPI means that the counterfactual outcome (the expected deforestation pressure) is significantly higher than the observed deforestation and the issuer was successful in avoiding deforestation and should be rewarded. The model of deforestation pressure provides an estimate for what level of deforestation rates we would expect given the development of global and largely exogenous underlying demographic, economic, and climatological factors. We use this section to explain how the model can be used to build a measure of performance for performance-linked financing instruments. We also explain how to use the FAB framework to build targets that are achievable for the issuer but require additional effort and resources. 5.1 KPIs and Relative Performance With the benchmark model (Table 4, Table 5, and Figure 10) the estimated coefficients from the estimation phase are used to calculate ex post benchmark deforestation rates for the RoC during the evaluation phase. For example, for an evaluation of the 2024 to 2030 period, the estimation period is defined as 2001-2023 (as we have done) and the estimated coefficients are used together with the data from 2024 to 2030 to calculate the benchmark deforestation pressure. For evaluation, this would be done ex post, but for factors where forecasted variables are available it can also be done ex ante to assess a likely trajectory of deforestation pressure. The KPI is then defined as the difference between the benchmark deforestation pressure and the observed deforestation outcome. Using confidence intervals, we can determine whether the outcome is significantly better (or worse) than the benchmark. The empirical results laid the foundation for a KPI that is robust to exogenous variation from climatological, economic and demographic factors. The dashed lines in Figure 12 demarcate the benchmark, which depends on the factors described in Table 2. Note that the benchmark level of deforestation is always set one period in advance based on the estimated change in deforestation pressure, with the market prices and exchange rates from the current (or past) period. The difference between the observed outcome in the period of interest and the benchmark set in the preceding period constitutes the KPI. As illustrated 22 Figure 12: Visualization of Benchmarked KPIs A confidence interval around the benchmark levels indicates whether the observed outcome (here deforestation rate) is significantly better or worse than the benchmark outcome. Source: Wang et al. (2023) in Figure 12, the absolute difference itself is the sole determinant of whether we can speak of significant performance or not. The distributions around the dashed line show the likely range of outcomes given the historical relationship between forest loss and its determinants. In our model for the RoC, the distributions were calculated using residual bootstrapping. Only if the outcomes are significantly higher or lower than the benchmark can we speak of additionality. For further discussion, see Wang et al. (2023). 5.2 Incentive-Compatible KPIs Effective forest protection through performance-based financing hinges on well-calibrated KPIs that set strong incentives for the issuer. Sustainability-linked instruments, such as SLLs or SLBs, impose no conditions on the use of their proceeds, which differentiates them from labeled bonds. While green bonds or sustainability bonds earmark and track their expenses, proceeds from SLBs and SLLs are added to the general government budget. To incentivize the issuer to achieve sustainability performance targets, such as lowering emissions or improving forest resilience, the instruments tie KPIs to financial ‘carrots and sticks’. It is therefore critical for KPIs to be well-designed to safeguard the environmental integrity, set incentives for the issuing government, and protect investors from reputational risks. Ill-constructed KPIs may result in adverse selection before issuance and moral hazard afterwards. The deforestation benchmark describing the historical link between exogenous factors and forest loss (Table 5) could in theory be exploited if the coefficients overstate the true influence of factors such as commodity prices. An issuer that knows this bias exists would deem the SLB and SLL terms as attractive—that is when prices rise, the model inflates the expected deforestation benchmark, making it easier for the country to ‘beat’ the target.23 After issuance, moral hazard risk can reverse the intended incentive: that is an issuer might postpone resilience investments, or even allow vulnerability to commodity prices to grow, so that future shocks keep the benchmark high and the target easier to achieve. Such concerns are justified, but unlikely to be a major concern in practice. ‘Gaming the system’ would require significant levels of coordination and result in only marginally raising the probability of achieving the target. Robust MRV systems are the most practical safeguard against information asymmetry in KPI-linked in- struments. More intricate mechanism design features or more sophisticated econometric methods could dampen adverse selection and moral hazard risks, but only by sacrificing transparency and simplicity—a costly trade-off for real-world market transactions. Therefore, SLBs and SLLs rely on annual reporting 23 To illustrate this point, suppose the estimated sensitivity of deforestation with respect to wood prices is ˆ (see Table 5). However, assume that the ‘true’ sensitivity is < ˆ . When wood prices increase, the model will predict deforestation pressures using ˆ that is higher than the actual rate using . Consequently, it would be easier for the country to keep actual deforestation below the inflated expectations and therefore get rewarded without having to exert the necessary amount of policy effort. 23 that disclose KPI values and progress toward the targets.24 The expense of maintaining this infrastructure itself signals the issuer’s commitment, while the readily available data facilitates price discovery and boosts investor confidence. Once the MRV system is in place, the externally audited platform will exist for at least the lifetime of the bond and can support other national reporting needs, extending the MRV system’s value well beyond the issuance of SLBs and SLLs. While no benchmarking method can eliminate information asymmetries, the proposed KPI-linked approach offers advantages over moving average methods, particularly in setting incentives. Benchmarks based solely on moving averages are in principle more susceptible to manipulation. They are designed to gradually ratchet downward and promote reduced deforestation, while temporary spikes in deforestation can artificially raise future benchmarks, making subsequent targets easier to meet. In contrast, the KPI- linked approach links expected deforestation to exogenous drivers (see Table 5), thereby limiting the ability to inflate benchmarks through short-term policy lapses. More importantly, as argued in Wang et al (2023), adjusting KPIs for factors beyond the issuer’s control allows for more accurate attribution of outcomes to government effort. In a principal-agent context, this improves incentive alignment—that is risk-averse governments are more likely to exert consistent effort when they are evaluated against a KPI that reflects true performance rather than external volatility. In addition to informing market-based instruments such as SLBs and SLLs, the proposed KPI framework can also underpin broader results-based finance. Development partners and multilateral climate funds increasingly require verifiable, policy-relevant indicators to justify disbursements. By isolating the effect of government effort from exogenous shocks, this KPI provides a transparent and empirical basis for performance-linked transfers beyond capital markets. This strengthens the case for additional climate finance, whether through bilateral partnerships or mechanisms such as the Green Climate Fund (GCF) or the LEAF Coalition. Note that the benchmark model presented in Table 5 is primarily a proof of concept. In practice, the model can be simplified by using a smaller set of key variables, preserving most of its benefits while improving transparency and ease of implementation. 6 Setting Feasible and Ambitious Targets The Government of the RoC has made multiple commitments and set targets for forest cover, tree cover loss, and deforestation, all varying in scope. See Table ??. However, these targets do not specify any rates for deforestation or degradation. Instead, they focus on reforestation and avoided deforestation—which needs to be compared with benchmark levels, conversion limits, and restoration targets, all in hectares. However, the FAB framework, and our model of deforestation pressure estimated previously, can be used to determine a range of national targets for deforestation. The FAB framework helps identify targets that are feasible. This means there is historical precedent that peer countries have successfully followed such a trajectory. It is also ambitious as they constitute a significant improvement above a BAU scenario. The objective of the FAB framework is to help identify targets that, from a purely empirical standpoint, reflect ambitious improvements above a BAU scenario and at the same time are feasible for the country to achieve. Where Section 5 focused on how a deforestation KPI could be defined to isolate policy performance from exogenous climatological, demographic, and economic factors, this section focuses on setting targets. It is important to stress that the target range estimated in this analysis was based on quantitative defor- estation data of RoC and peer countries. It does not incorporate qualitative insights from local subject matter experts on recommended policies and their potential impact. As such, it should not be interpreted as the definitive target range for RoC, but as a starting point for more advanced discussion with relevant stakeholders. A crucial element of the FAB framework is selecting appropriate peer countries for the Republic of Congo. When deciding the number of selected peers two objectives should be considered: (1) specificity—choosing only peer countries that are like the Republic of Congo in terms of socioeconomic, environmental, and 24 Costly state verification (Townsend, 1979) underpins the need for monitoring; performance bonds/take-or-pay clauses (Grossman & Hart, 1980) and residual control rights under incomplete contracts (Grossman & Hart, 1986) credible commitment. In SLBs/SLLs, mandatory annual MRV enforces KPI-linked penalties and clear verification rights. 24 Figure 13: Deforestation Rates in Congo Basin and HFLD Peers Peer countries are Congo Basin countries (Central African Republic, Cameroon, Democratic Republic of Congo, RoC, Gabon, and Equatorial Guinea) and HFLD countries (Belize, Bhutan, Colombia, Guyana, Panama, Papua New Guinea, Peru, Solomon Islands, and Suriname). Note that the graph reflects deforestation in the tropical moist forests of the respective countries. Some countries, such as the Solomon Islands, had higher rates of forest disturbances, but those were concentrated in the dry domain. Source: “Total annual deforestation (direct or after degradation)” from TMF (European Commission 2024) relative to the total area of undisturbed forest cover in 2000. geographic conditions; and (2) robustness—choosing enough peer countries so that no individual country has disproportionate influence on the results. For this analysis we selected the other Congo Basin countries and all HFLD countries with available data as the set of potential peer countries. To define HFLD countries we followed the updated list in Simon et al. (2021). They updated the original list of HFLD countries from Fonseca et al (2007) with more recent data and the original definition of “a minimum of 50% forest cover and a rate of change in forest area less than the global average based on FAOSTAT data over a 10-year period” (see Figure 7 in Simon et al. (2021)). 6.1 Feasibility To assess what target range would be feasible to achieve, we investigated the historic development of deforestation rates in Congo Basin and HFLD peer countries. In the feasibility assessment, we asked the question: “What trajectory have peer countries followed in the past, when they had a similar rate of deforestation as ROC has today?”. To narrow down the list of peer countries to those that have been sufficiently similar, we focused on countries that had historical deforestation rates below 0.15 percent (recall that RoC’s deforestation rate in 2023 was 0.058 percent). Bhutan, Cameroon, Equatorial Guinea, Gabon, Guyana, Panama, Papua New Guinea, Peru, Solomon Islands, and Suriname satisfied that criterion. The latest year in which the deforestation rate was below 0.15 percent was chosen as the base year, and the following seven years from the base year were chosen as the relevant trajectory. Figure 14 shows that the deforestation trajectories of peer countries have remained stable in the past, when they started from a deforestation rate like the Republic of Congo in 2023. We also saw that among the peer countries, Gabon had been successful at maintaining an extremely low deforestation rate, whereas Cameroon’s deforestation rate had increased from 0.1 percent to 0.34 percent between 2009 and 2016 (the seven years after 2009—the year in which it had last had a similar rate of deforestation as the Republic of Congo). 25 Figure 14: Peer Countries with Similar Historical Deforestation Rates The feasibility analysis shows how peer countries deforestation rates have developed in the past, in the years following similar deforestation rates as RoC has today. Source: “Total deforestation (direct or after degradation)” from TMF (European Commission, 2024) to the total area of undisturbed forest cover in 2000. Selected peer countries are Congo Basin or HFLD countries with available TMF data and that had similar deforestation rates as ROC (below 0.15 percent) at any point in the past 20 years. When we shift our focus from annual deforestation to the total accumulated deforestation, we found that peer countries had deforestation rates that would lead to almost 150,000 hectares over the studied horizon if RoC were to deforest at a similar rate. Figure 15 shows how much deforestation was accumulated in peer countries trajectories. On average, peers accumulated deforestation at a rate that would translate to almost 150,000 hectares of deforestation for RoC within the seven years after which the peer countries most recently had similar deforestation rates as the RoC today.25 Fifty percent of all peer countries accumulated deforestation at a rate that corresponds to between 87,568 and 298,428 hectares for RoC (the 50 percent feasibility interval) and 90 percent accumulated between 54,417 and 577,647 hectares (the 90 percent feasibility interval).26 By investigating the distribution of accumulated deforestation in peer countries, we were able to classify three levels of outcome. These were: (1) highly feasible—between 87,568 and 298,428 hectares; (2) feasi- ble—between 54,417 and 87,568; and (3) low feasibility—below 54,417. These thresholds corresponded to the 50 and 90 percent feasibility intervals. Figure 246 shows what the feasibility intervals looked like for the annual deforestation rates. Because targets are usually set over a longer horizon (such as between 2023 and 2030), our focus for target setting was on the accumulated deforestation. Historical trajectories of worst-case scenarios in the selected peer countries had accumulated deforestation up to 23 percent higher than in our baseline assessment. Instead of selecting the most recent year in which a peer country had similar deforestation rates to the RoC today (as we did in the baseline assessment of Figure 14 and Figure 15), we also selected the worst historical trajectory (that is with the highest increase in deforestation). In this worst-case scenario, annual deforestation rates more than doubled over time—from 25 Peer countries differ in their total land area. To make them comparable with each other we looked at the accumulated deforestation rates (as a share of total undisturbed forest cover in the year 2000) in peer countries and the RoC. We then multiplied the accumulated deforestation rates with RoC’s area of undisturbed forest cover in the year 2000 to show how much hectares of deforestation we would expect if RoC were to deforest at a similar rate as peer countries. 26 There were also a significant number of historical trajectories that reflected large increases in accumulated deforestation. Figure 15 shows that more than 25 percent of peer countries had deforestation rates that would lead to more than 324,089 hectares of accumulated deforestation in seven years. 26 Figure 15: Feasibility Intervals from Accumulated Historical Deforestation of Peers The feasibility analysis shows how peer countries deforestation rates have developed in the past, in the years following similar deforestation rates as RoC has today. Source: TMF (European Commission, 2024) relative to the total area of undisturbed forest cover in 2000. The trajectories of peer countries are accumulated over seven years and the quantiles of the distribution plotted in the graph. Selected peer countries are Congo Basin or HFLD countries with available TMF data and that had similar deforestation rates as RoC (below 0.15 percent) at any point in the past 20 years. 0.06 percent in the base year to 0.17 percent seven years later. Total accumulated deforestation was also higher—on average 179,844 hectares within seven years (see Figure 247). 6.2 Ambitiousness To assess what target range would be ambitious, we estimated a likely BAU scenario based on our model of deforestation pressure and forecasted values for the climatological, demographic, and economic factors. In the ambitiousness assessment, we asked the question: “What business as usual scenario is likely and how would a meaningful improvement beyond BAU look?”. We used the model estimates from Table 4, Model 4, and forecasted values of the regressors to produce an out-of-sample forecast of deforestation pressure for the RoC between 2023 and 2030.27 Deforestation pressure is expected to increase for the RoC between 2023 and 2030. Figure 16 shows that our BAU scenario, based on the model of deforestation pressure, projected a significant increase in deforestation due to climatological, demographic, and global economic factors largely outside the government’s control. Projected deforestation rates in the BAU scenario increase from 0.14 percent in 2023 to 0.20 percent in 2030, which corresponds to annual deforestation of 46,069 hectares. In 2023, the RoC kept deforestation below expectations, and it is possible that successful implementations of forest-related policies and sustainable logging practices will lead to deforestation outcomes significantly below the BAU scenario. It is important to note that the BAU scenario in Figure 16 shows the expected deforestation based on deforestation pressure from climatological, demographic, and economic factors. For example, in 2023 the expected deforestation rate was 0.14 percent, but the observed deforestation rate was only 0.06 percent. This corresponds to avoided deforestation of almost 20,000 hectares. Hence, the values from the BAU median path in Figure 16 should be interpreted as an expectation without additional policy effort. However, successful policies may lead to observed deforestation significantly below the BAU scenario. For target setting, the total accumulated deforestation in the BAU scenario should be considered the bench- mark against which meaningful improvements need to be made. Figure 17 shows how the projected deforestation rates from the BAU scenario translated into accumulated hectares of deforestation over the seven-year period (2023-2030) assessed in this study. The most likely trajectory shows accumulated defor- estation of around 280,000 hectares until 2030, with 50 percent of the scenarios ranging from 190,880 to 27 As forecasts with uncertainty intervals are not available for most predictors, we rely on univariate ARIMA(p,d,q) time-series forecasts. The parameters , , were determined using Akaike information criteria 27 Figure 16: BAU Projections for Deforestation Pressures until 2030 The ambitiousness analysis produces a BAU scenario of deforestation rates in RoC from 2023-2030. Source: Authors’ calculations. The variable of interest was “Total deforestation (direct or after degradation)” from TMF (European Commission 2024) relative to the total area of undisturbed forest cover in the year 2000. The forecast is an out-of-sample prediction based on the model of deforestation pressure presented in Section 4 and autoregressive integrated moving average (ARIMA) forecasts of the climatological, demographic, and economic factors that are inputs into the model. 377,221 hectares (the 50 percent ambitiousness interval) and 90 percent of the scenarios between 67,907 to 519,996 hectares (the 90 percent ambitiousness interval). According to the expected deforestation from climatological, demographic and global economic factors, during the years 2023 to 2030, it would be an ambitious target to keep accumulated deforestation between 67,907 and 90,880 hectares, and a highly ambitious target to keep it below 67,907 hectares. The distribution of BAU projections tell us how likely different deforestation outcomes are without additional policy effort. For target setting, we considered outcomes that occurred in less than 25 percent of the projections as ambitious and those that occurred in less than 5 percent of the projections as highly ambitious. A comparison of our BAU scenario with a hypothetical benchmark based on the historical reference line showed how the latter unfairly raises the bar for HFLD countries like the RoC. HFLD countries have historically low levels of deforestation but may nevertheless face increasing pressure on forests from clima- tological, demographic, and global economic factors. When reference lines are only calculated based on historical deforestation rates, without incorporating rising deforestation pressure, the resulting benchmark leaves little room for improvement through successful policies. See Figure 244 for an illustration. A deforestation benchmark based only on the historical deforestation rates from the past five years would be 46 percent lower than the benchmark incorporating rising deforestation pressure. The average annual deforestation rate between 2019 and 2023 was 0.091 percent. Extrapolating this rate as a benchmark for the 2023 to 2030 period translates to total accumulated deforestation of just over 150,000 hectares.28 This is significantly lower than the BAU scenario of 280,961 hectares from Figure 17. 28 0.091 percent annual deforestation rate times 23.6 million hectares undisturbed forest area in the year 2000 × 7 years. 28 Figure 17: Accumulated Deforestation under BAU projections until 2030 The ambitiousness analysis describes a BAU scenario of deforestation in RoC from 2023-2030. Source: Authors’ calculations. The variable of interest was “Total deforestation (direct or after degradation)” from TMF (European Commission 2024) relative to the total area of undisturbed forest cover in the year 2000. The forecasted annual deforestation rates (see Figure 16) are accumulated for the seven years between 2023 and 2030. 6.3 Combined Assessment One benefit of the FAB framework is the ability to visualize and systematically address the trade-off between setting targets that are as ambitious as possible while still being feasible to achieve. Targets that are significantly better than the BAU scenario projects have high ambition, but if there is no historical precedent for peer countries following such a trajectory, they have low feasibility (long shots). Similarly, targets that many peer countries have been able to follow in the past have high feasibility but may happen anyway under the BAU scenario without additional policy effort (low hanging fruits). According to the combined FAB assessment, during the years 2023 and 2030 a target range of total accu- mulated deforestation between 54,417 and 67,907 hectares would be classified as high feasibility and very high ambitiousness. Figure 18 presents the combined FAB assessment. Targets below 67,907 hectares are considered very ambitious—and until 54,417 hectares there is sufficient historical precedent to also consider such a trajectory feasible. Targets that represent more accumulated deforestation until 2030 (between 87,568 and 190,880 hectares) would be classified as ambitious and highly feasible.29 The FAB assessment is not prescriptive, and the appropriate sustainability performance target ultimately needs to be decided by local authorities. Nevertheless, the assessment quantifies the trade-off between feasibility and ambitiousness and provides possible target ranges as a starting point for further discussion. 29 Total accumulated deforestation between 67,907 and 87,568 hectares would be ambitious and feasible. 29 Figure 18: FAB Assessment for Accumulated Deforestation until 2030 The forecasted accumulated deforestation until 2030 results from the FAB assessments. Comparable peers tended to experience a steady increase in deforestation. Assuming similar dynamics apply for RoC, then deforestation reduction targets between 88 to 298 thousand hectares are highly feasible (feasibility panel) since we found plenty of historical precedence. However, such targets would likely be achieved without any policy intervention under the BAU scenarios, which identifies targets between 191 and 377 thousand hectares as low ambition (ambitiousness panel). Targets above 88 thousand hectares are highly feasible compared to peers and below 191 thousand hectares are still highly ambitious under BAU and strike a good balance between dimensions (combined assessment panel). 7 Subnational Deforestation: Sangha and Likouala This section presents an extension of the baseline model to sub-national data to highlight deforestation trends in Sangha and Likouala, two key areas of focus for sustainable forest management in the RoC. It also provides a proof-of-concept for a panel data framework that models deforestation pressure in multiple jurisdictions simultaneously. Located in the northern part of the RoC, Sangha and Likouala constitute about one-third RoC’s land area (124,000 square kilometers) and are almost exclusively covered by forests. Land forest cover is 60 percent and wetland forests are 37 percent. In addition, the areas have historically low deforestation rates. However, the two provinces have seen increasing logging activities in recent years and the government, together with REDD+ and the World Bank, are advocating for sustainable logging management (FCPF & World Bank Group, 2018). The FCPF, together with Terra Global and the Congo National Coordination body for REDD+ has initialized an emissions reduction program for Sangha and Likouala in 2018.30 The program’s objective is to reduce 9,013,440 teCO2 between 2018 and 2023, and “enhance sustainable landscape management improve and diversify local livelihoods, and conserve biodiversity” (FCPF & World Bank Group, 2018). The program document lists four main intervention areas, focusing on: (1) reduced impact logging; (2) improving the livelihoods of communities; (3) providing alternative sources of income; and (4) improving the management of existing protected areas. In 2023, the two departments contained 17 forest concession areas covering 72,000 square kilometers (58 percent of total land area). National parks and reserves covered 30,000 square kilometers—a significant portion of the remaining area. Figure 19 and Table ?? show the names, location and starting year of operation of the 17 concession areas. To determine when concessions started their operations, we used the first year where logging volumes were reported in a validation workshop of the emissions reduction program. Note that the exact start year needs to be validated with official data, as in some cases the years in which certifications were registered on platforms such as the Forest Stewardship Council (FSC) did not align with the year where logging volumes were first reported. Total deforestation in Sangha and Likouala in 2022 was 2,756 hectares and 4,497 hectares respectively, which represents about 23 percent of ROC’s deforestation in that year. Figure 20 shows that deforestation follows a similar pattern as in RoC overall, a significant increase between 2010 and 2016 followed by a decrease of deforestation in recent years. Deforestation was considerably lower in 2023 in both departments with only 30 See: www.terraglobalcapital.com/emission-reductions-program-sangha-and-likouala-republic-congo. 30 Figure 19: Concession Areas in Sangha and Likouala The 17 concession areas in Sangha and Likouala cover about 58 percent of total land area in these two districts. A large part of the remaining area is covered by national parks and reserves. Source: Names and borders of concession from GFW logging concessions dataset (Global Forest Watch 2019). The Karagoua concession southwest of Jua-Ikie started operations in 2020 but was not yet registered in the 2019 GFW dataset. 601 hectares in Sangha and 668 hectares in Likouala. Given that the undisturbed forest area in northern RoC is so large, relative deforestation rates in Sangha and Likouala are still low—between 0.01 to 0.08 percent of total undisturbed forest cover in the year 2000. Figure 20 also shows short-term forest degradation where the influence of the 2016 wildfires on forests in the Sangha province is clearly visible (Erickson-Davis, 2016b). Forest degradation reflects short-term disturbances that do not persist for more than three years and are commonly due to fires, extreme weather conditions, and selective logging. With sustainable logging, forest disturbances should mostly appear as forest degradation in the satellite data (Vancutsem et al., 2021). However, the development of logging infrastructure can lead to forest disturbances that are picked up as deforestation (such as access roads, storage, and processing areas). Here, sustainable forest management could be easily detectable on satellite data as it requires a higher degree of organization with wider access roads. This is an important caveat when using geospatial data as an underestimation of illegal or unsustainable logging opposed to sustainable logging would punish countries for engaging in the later. Besides field validation of the satellite-detected deforestation values, a solution could be to apply post- processing to the geospatial data and indicate deforestation due to logging infrastructure separately. For example, road networks and associated deforestation are clearly visible on satellite images underlying the data for this analysis (see Figure 254). We assessed the impact of concession areas coming into operation by analyzing deforestation (and degra- dation) rates before and after a given concession area became active. Such an event-study analysis can be used to see if there is a discontinuity around the timing of the event. Because concession areas came into operations in different years, we aligned them by using the year of the start of operation as year 0. Figure 253 shows that, for Sangha province, there is no clear change in deforestation or degradation rates immedi- ately before and after the start into operations for different concession areas. However, the precision of the analysis suffers from the uncertainty around the actual starting of operations in the different concession areas. Next, we analyzed the impact of logging and logging concessions on deforestation. This was done by re-estimating our baseline model (see Section 4) for Sangha and Likouala and taking concession areas 31 Figure 20: Total Deforestation and Degradation in Likouala and Sangha Provinces (a) Total deforestation in Likouala province in- (b) Total degradation in Sangha province spiked creased fourfold between 2009 and 2016 since the in 2016 due to extremely dry conditions and asso- previous period and recently reverted back. ciated wildfires. Source: Author’s calculation based on data from TMF (European Commission 2024). Deforestation was “Total defor- estation (direct or after degradation)”; Degradation was “Total degradation (followed or not by deforestation)”. into account. Table 6 presents the results using province-level data for Sangha and Likouala (Columns 1 and 3).31 Here, we showed the results of an adjustment to Model 1 from the baseline model, including the climatological and most economic factors and controls for singular events. However, we excluded demographic factors and the price for cash-crop commodities. Due to data limitations, the available data for demographic factors reflected the situation in Congo overall and given the limited population density of Sangha and Likouala was less suitable for these two provinces. Sangha and Likouala province also did not have a large cash-crop sector, so their prices were less likely to be relevant factors.32 In all specifications, climatological factors and prices for roundwood and hard sawn wood were associated with increased deforestation rates. The magnitude of the effect for wood prices was also larger in those two provinces than for Congo overall, suggesting a stronger relationship between global wood commodity prices and deforestation. The effect of the REER was less clear compared to the estimation using national data. While the first lag was negative but mostly insignificant or not selected, the second lag for Likouala was positive and significant, which is contrary to the expectation that an appreciation of the currency leads to increased deforestation through higher competitiveness of timber products. An important singular event to control for is the road development of the N2 from Brazzaville to Ouesso between 2010 and 2013. The road goes through Cuvette-Ouest and Sangha and significantly increased accessibility to RoC’s northern provinces. Therefore, we included the control for Likouala (although the road is not located there) and modeled the effect as ongoing, to control not only for direct deforestation due to the road construction but also for indirect deforestation due to increased accessibility. The estimated coefficient was positive and significant, indicating that after 2010 deforestation rates had shifted upwards.33 Active concession areas are associated with a small and significant increase in deforestation rates in Sangha, but not in Likouala. We also looked at the interaction effect between active concession areas, the REER, and wood prices. While most interactions were not selected, the coefficient for the second lag of wood 31 The deforestation rates for Sangha and Likouala were extracted from the geospatial TMF data by restricting the analyzed area to the respective administrative boundaries. 32 When we included cash-crop commodity prices the results were inconclusive being associated with an increase in deforestation in the second lag, but also (for Sangha concession level estimations) a significant decrease in the first lag. 33 When we modeled the control to only cover the period between 2010 and 2013 the estimated coefficient was much larger. However, the period also coincided with higher global commodity prices and the estimation technique did not allow us to disentangle these two effects that occurred simultaneously. With geospatial data tailored to the Congo context, it is possible to improve this by, for example, excluding the road network from the analyzed data to estimate deforestation that is not directly due to road construction. 32 Table 6: Results of Deforestation Model: Sangha and Likouala Provinces Sangha Likouala Variable Province Concession Province Concession Timeseries Area Panel Timeseries Area Panel Lagged relative deforestation rate, t–1 0.025 n.s. 0.179** 0.179** Climatological factors SPEI-Index (12-month), t –0.287** –0.303** –0.182** –0.222** Economic factors Real effective exch. rate (%Δ), t–1 –0.030 –0.048* –0.045 0.002 Real effective exch. rate (%Δ), t–2 n.s. n.s. 0.110 0.084** REER x Active concession area, t–1 . 0.041 . –0.020 REER x Active concession area, t–2 . –0.038 . 0.032 Round- and sawnwood price (US$), t–1 0.122** n.s. 0.121 0.056 Active concession area, t–1 . n.s. . –0.370** Round- and sawnwood price (US$), t–2 0.383** 0.468** 0.417** 0.380** Active concession area, t–2 . 0.088* . 0.020 Forest-related policy, concessions and one-off events Active concession area, t . 0.066** . –0.000 El Niño / wildfires 2016, t n.s. n.s. . . N2 Road Project 2010 onwards, t 0.210** 0.009** 0.281** 0.360** Policy dummies Yes Yes Yes Yes Adjusted 2 0.24 0.29 0.51 0.50 Note: n.s. = not selected. The variable was included as a candidate but not selected during the regularization step. Significance levels are based on residual bootstrapping, showing significance at a 5% (**) and 10% (*) level. The outcome was “Total Deforestation (direct or after degradation)” relative to the undisturbed tropical moist forest in the year 2000. In Columns 1 and 3, the included area is the entire province. In Columns 2 and 4, each observation reflects one concession area plus a non-concession area group. Demographic and cash-crop commodity variables were excluded due to lack of subnational relevance. Wildfires are expected to affect Sangha primarily. Policy dummies were included for the national REDD+ strategy and the Brazzaville declaration (2018), the CAFI Letter of Intent (2019), and the updated Forest Code (2020). These were positive and significant in Sangha, suggesting higher-than-expected deforestation during 2018–2020, but partly negative and significant in Likouala. For the province-level Sangha model, stronger regularization was applied to improve 2 and sparsity without altering coefficient stability. prices in Sangha suggested that the elasticity of deforestation with respect to wood prices was larger in concession areas than in non-concession areas (or in concession areas before the concession was active). For Likouala, we found a contrary effect. The elasticity of wood prices was lower in concession areas than in non-concession areas. 8 Oil Rents, the Dutch Disease and Deforestation The RoC’s growth is highly dependent on oil production and total GDP closely follows global oil prices. This means that the RoC faces a challenging economic transition towards a growth model that is less dependent on oil revenue. While the forest economy will play an important part in diversifying the RoC’s economic base, it is important that economic development happens sustainably and is not accompanied by large-scale deforestation. This section shows that there is an empirical basis for concern. Historically, increased revenue from high oil prices has in the past crowded out economic activities in other sector and consequently eased the pressure on forests. The importance of oil rents for GDP means higher oil prices indirectly lead to lower deforestation through the economic phenomenon known as the Dutch disease (van der Ploeg, 2011; Venables, 2016). A boom in the oil sector boosts the economy and increases labor demand in urban and infrastructure sectors because of government expenditure, decreasing profitability in the agriculture sector and thus halting deforestation. At the same time, currency appreciation because of oil production also decreases the export competitiveness of Congolese agricultural products. A similar case can be made for Gabon, another HFLD country in the Congo Basin (Wunder, 2003). Reducing RoC’s dependence on oil revenue may lead to an unintended increase in deforestation pressures due to higher demand for (agricultural) land. While broadening RoC’s economic base is an important step 33 Figure 21: Global Oil Prices and Oil Rents in the Republic of Congo (a) Global Oil Prices and GDP (b) Oil Rents as a Share of GDP Source: World Development Indicators, FRED. towards a sustainable growth model and a stated objective of the government (World Bank 2023), economic diversification could also increase deforestation pressure if it strengthens agriculture, leading to high land demand. This risk could be mitigated if economic diversification prioritizes manufacturing or services, which are less land-intensive than agriculture. Figure 21 shows that GDP and oil prices are highly connected and that in 2021 oil rents accounted for around 35 percent of total GDP. However, because oil production mostly happens offshore and does not contribute to additional demand for land, it is not directly related to deforestation. We, therefore, excluded oil prices from the baseline model in the empirical analysis.34 However, oil revenue influences deforestation indirectly through its link with GDP. Extending our baseline model to incorporate this link allowed us to provide analytical support to the Dutch disease argument in the preceding paragraphs and provided some indication, and better predictions, of how deforestation pressure will develop under the foreseeable structural transition to diversify the economy. Including GDP directly in the model is not advisable, because it introduces a degree of endogeneity and theoretically makes the methodology more vulnerable to manipulation. GDP is not an external factor in the same way as global commodity prices or climatic conditions are. This means that the resulting benchmark would reflect a mix of external deforestation pressure and government policy. The latter, however, should be part of the KPI and not controlled for by adjusting the benchmark. In practice, it is unlikely that a government would change its ambition regarding GDP growth to manipulate a forest finance benchmark. Nevertheless, this section shows how to include GDP in a more robust way, by relaying its variation to changes in oil prices in a two-stage estimation design.35 To assess how a diversifying of RoC’s growth model away from oil could impact deforestation rates, we estimated the correlation between GDP and deforestation rates. We first split RoC’s GDP into two components: (1) the first related GDP to oil production through the global price for oil; and (2) the second component captured the part of GDP that is not related to oil production. Equation 2 shows the model estimated using ordinary least squares regression. Here, W is ROC’s GDP per capita adjusted for purchasing power (World Bank 2022, ID: NY.GDP.PCAP.PP.CD), Z is the global crude oil price (FRED, ID: POILWTIUSDM) and X is the real effective exchange rate (lagged by up to two years). = 0 + 1 + 2 + (2) Table 7 shows that higher oil prices are significantly positively associated with an increase in GDP. To construct a measure of GDP due to oil production, we extracted the fitted values of that regression using 34 In robustness tests where we included oil prices, they were not selected as a significant predictor of deforestation outcomes. 35 Notethat while the estimation design resembles that of an instrumental variable estimation, our focus was not on the effect that oil could have on deforestation ( 1 in equation 2) but instead the effect of non-oil-related GDP ( 2 ). 34 Table 7: First Stage Results to Split GDP into Oil and Non-Oil Component Variable Model 1 Global WTI oil price (%Δ), t 0.237** Global WTI oil price (%Δ), t–1 0.093** Real effective exch. rate (%Δ), t–1 0.437** Real effective exch. rate (%Δ), t–2 –0.810** Adjusted 2 0.42 Note: Significance levels were based on analytical standard errors of the OLS regression, showing significance at a 5% (**) and 10% (*) level. The adjusted 2 accounts for the number of included regressors. Table 8: Correlation between Deforestation, Oil-Related, and Non-Oil-Related GDP Deforestation rate GDP (oil component) GDP (non-oil comp.) Deforestation rate 1.000 –0.142 0.014 GDP (oil component) –0.142 1.000 0 (by construction) GDP (non-oil component) 0.014 0 (by construction) 1.000 the estimated coefficients in Table 7. The resulting prediction represents the share of GDP that can be attributed to oil prices, an arguably exogenous variable with respect to the Congolese economy. We used the unexplained variation, that is GDP minus the share of oil-related GDP, as a measure of non-oil GDP. The correlation between deforestation rates and oil and non-oil GDP confirmed that higher GDP from oil production is associated with lower deforestation while higher GDP from non-oil components is associated with higher deforestation rates. See Table 8. The correlation between oil GDP was negative (-0.142), suggesting that the Dutch disease may have contributed to lower deforestation rates in RoC by channeling resources towards the (offshore) oil sector away from more land intensive sectors. The correlation between non-oil GDP and deforestation was positive (albeit small). This emphasizes that a diversification of the economy needs to consider that alternative income channels may be more land intensive, such as export- oriented agriculture, and that may increase the demand for land and lead to more deforestation pressure. To estimate the roles of both GDP components on deforestation, we included them into our baseline model. See Table 4. Equation 3 shows the estimated specification which differed from the baseline specification only through the inclusion of the fitted and residual GDP components, ˆ −1 and ˆ −1 , respectively, which were both included with one lag. As in Model 4, was the intercept, was the autocorrelation coefficient of lag . − were the candidate variable that proxy factors influencing deforestation and | ≥ were dummies that controlled for singular events. The estimation was conducted similarly to Section 4 with regularization to restrict the number of variables ultimately included in the model and the confidence intervals calculated through bootstrapping. − = + − + − + 1 ˆ −1 + 2 ˆ −1 + | ≥ + (3) =1 =1 =1 ∈ Table 9 shows that an increase in non-oil GDP is associated with higher deforestation rates. Including the residuals from the first stage (non-oil GDP) described above as a measure of GDP not related to the oil price shows a positive association with deforestation rates with a magnitude about half as strong as the effect of global wood prices. An increase in oil GDP on the other hand is not significantly related to an increase in deforestation rate. Additionally, the estimated coefficients from the baseline model are unchanged when including a measure of oil and non-oil GDP. 35 Table 9: 2-Stage Model of Deforestation Pressure with Oil and Non-Oil GDP Variable Model 1 Model 1 (IV) Model 4 Model 4 (IV) Baseline IV Baseline IV Lagged relative deforestation rate, t–1 0.234** 0.197** 0.099** 0.028 Climatological factors SPEI-Index (12-month), t –0.267** –0.116** –0.225** –0.192** Economic factors Real effective exch. rate (%Δ), t–1 –0.230** –0.172** –0.197** –0.277** Real effective exch. rate (%Δ), t–2 n.s. –0.003 n.s. –0.027 Round- and sawnwood price (US$), t–1 0.145* 0.145** n.s. –0.041 Round- and sawnwood price (US$), t–2 0.341** 0.200** 0.219** 0.230** Oil-related GDP per capita, t–1 . –0.001 . n.s. Non-oil-related GDP per capita, t–1 . 0.100** . 0.141** Cash-crop prices (US$), t–1 . . n.s. 0.033** Cash-crop prices (US$), t–2 . . 0.569** 0.702** Demographic factors Urbanization rate, t . . –0.095** –0.146** Urbanization rate, t–1 . . –0.102** –0.150** Urbanization rate, t–2 . . –0.107** –0.153** Rural population growth rate, t . . n.s. n.s. Rural population growth rate, t–1 . . 0.039** n.s. Rural population growth rate, t–2 . . 0.053** n.s. Forest-related policies and one-off events El Niño / wildfires 2016, t . . n.s. n.s. N2 Road Project 2010 onwards, t . . n.s. n.s. Policy dummies No No Yes (n.s.) Yes (n.s.) Adjusted 2 0.55 0.40 0.62 0.75 Note: n.s. = not selected: The variable was included as a candidate but not selected during the regularization step. Significance levels were based on residual bootstrapping, showing significance at a 5% (**) and 10% (*) level. Columns 1 and 3 repeat the results from Table 3 for comparison. Oil and non-oil GDP reflect ROC’s total GDP split into two components via a first-stage regression using the global price of oil as an instrument. The fitted values represent oil-related GDP per capita, and the residuals represent non-oil-related GDP per capita. Model 4 allows for policy dummies for the REDD+ strategy and Brazzaville Declaration (2018), the CAFI Letter of Intent (2019), and the updated Forest Code (2020), none of which were selected in the regularization step. The adjusted 2 accounts for the number of included regressors. The dependent variable is “Total deforestation (direct or after degradation)” from TMF (European Commission 2024). 9 Mobilizing Financing Instruments for Forests 9.1 Risk-Based Approach versus Historical Reference Lines The global financial system is increasingly exploring ways to fund HFLD countries, including RoC, for their valuable forests. Recent assessments show that public and private finance for forests remains far below the estimated needs of US$460 billion per annum for meeting global goals to halt and reverse deforestation by 2030 (Forest Declaration Assessment Partners, 2023). Most of the currently available forest finance comes from bilateral and philanthropic grants, which lack scale and predictability. There is an urgency to increase the scale of financing, increase the leverage and certainty from global instruments and mechanisms, and mobilize specific financing instruments to meet these gaps. The Government of Singapore has initiated a series of high-level roundtables focused on the future of HFLD finance, engaging academic and technical partners. The TFFF is also exploring innovations in large scale financial incentives for tropical forest countries to conserve and increase forest cover. The RoC seeks to diversify its economy and pursue economic growth, but significant financing is needed. The NDP (2022-2026) aims to foster a green economy, promote eco-tourism, expand timber production, and enhance forest-based industries, aiming to create 5,000 to 7,000 forestry jobs. ROC’s forests contribute US$99.5 billion per year to global carbon value, essential for climate stability (World Bank 2025, forthcom- ing). Their domestic value is US$110 million per year highlighting a disparity between national and global benefits (Forest Declaration Assessment 2023). Although the forest value of RoC and HFLD countries is acknowledged globally, support remains limited due to performance assurance challenges. Without a 36 concerted strategy and external assistance, RoC as a lower-middle-income country risks compromising its forest assets through increasing levels of forest degradation and deforestation. Innovative financing mechanisms, with robust KPIs can incentivize finance flows to HFLD countries, like the RoC. Debt-for-nature swaps, SLLs, SLBs, and results-based climate finance provide support for conservation and sustainable management, offering funds to build capacity, and improving access to climate finance. Risk-based KPIs create incentives that protect forests, countering historical deforestation and addressing gaps in sustainability. It is crucial that forest and climate finance from various national and international sources, including climate funds and private foundations, unlocks sustainability. Climate or forest finance should not only focus on ‘avoided deforestation as a KPI’ but also support sustainable growth, job creation, and sustainable forest management. 9.2 Role of KPIs in Unlocking Performance-Linked Finance KPIs serve as critical tools for driving results-based and outcome-based financing. These indicators create a framework of measurable and verifiable benchmarks that instill confidence in investors. When robust and transparent systems of KPIs are established, they not only foster trust but also encourage investors to extend grants or concessional financing. The clarity and accountability offered by KPI systems are essential for cultivating a supportive investment environment, ultimately leading to more impactful financial part- nerships. While securing performance around KPIs is key, emphasis should also be on sustainable growth, job creation, and forest management. Risk-based KPIs can help scale up outcome-based payments that support environmental outcomes. Pay- ments (such as grants, reduced interest, and couple payments) help facilitate the implementation of inter- ventions that result in further environmental gains such as forest preservation, biodiversity conservation, and avoided carbon emissions. These KPIs are built on quantification, measuring how much forest is at risk of being lost, and establishing measurable indicators of avoided risk. They can help guide the disburse- ment of resources that are provided with no expectation of repayment, where funding is contingent upon the achievement of specific, predefined outcomes. Risk-based KPIs improve internationally recognized frameworks for managing grants—such as the Global Environment Facility (GEF) project cycle and the Organization for Economic Co-operation and Development (OECD) Development Assistance Committee (DAC) principles—and help enhance effectiveness by helping establish baselines, assess additionality, and determine disbursement triggers. KPIs related to financial terms can enhance concessional financing instruments like SLLs and SLBs. These instruments connect financial terms, like interest rates or coupon payments, to the borrower’s achievement of sustainability targets. The funds are not designated for specific projects. National and international financial institutions can provide this financing. For example, Congo is eligible for concessional funding from the International Development Association (IDA). KPIs should appropriately adjust interest rates or coupon payments. The Loan Market’s Association (LMA) SLL Principles and Green Loan Principles, along with the SLB Principles, offer a framework for using risk-based KPIs to modify financing terms based on outcomes. In project- and program-based financing (such as grants, green loans, and green bonds) KPIs play a subordinate yet significant role. They are primarily used to monitor the implementation of sustainability interventions delineated in grant agreements, loan contracts, or bond prospectuses. For grants disbursed in tranches, KPIs are especially valuable for tracking progress and ensuring compliance prior to subsequent disbursements. 9.3 Readiness to Mobilize Financing Risk-based KPIs are essential for using financial instruments in HFLD countries. The RoC must meet requirements for each instrument. This paper’s KPI development approach aims to define outcomes and results for sustainable forestry, lowering financing barriers. Prerequisites include institutional capacity, relevant experience, favorable credit ratings, and a positive business environment. Financing instruments include grants, debt-for-nature swaps, SLLs, and SLBs and each has specific char- 37 Figure 22: Evaluation Financing Options acteristics and conditions at jurisdictional level. The country’s macroeconomic context influences these instruments’ ability to attract projects and program-based, outcome and results-based finance, FDI, and adherence to regional and global agreements. In addition, these conditions include the country’s ability to establish institutions that channel finance and foster donor trust. Attracting results-based finance, grants, and philanthropic contributions requires demonstrating appealing outcomes to funders through robust KPIs, effective MRV systems, and credible enforcement mechanisms. Figure 22 outlines the framework for a preliminary qualitative analysis of the readiness of RoC to mobilize a specific financing instrument. This is followed by a discussion on the three sets of conditions: (1) country context and policies; (2) financing source and motivation for public and private investors; and (3) international principles, frameworks and processes. 9.3.1 Country context and policies The first set of conditions in assessing the viability and associated risks of deploying a range of financial mechanisms analyzes the country’s context. The use of capital from instruments like green bonds, SLLs, or debt-for-nature swaps depends heavily on current policy, legal, and institutional frameworks. The country context analysis evaluates the stage of development, forest management practices, and climate finance mobilization, aligning these assessments with local priorities and challenges. This includes a consideration of monitoring vulnerabilities in the financial sector and the banking sector’s exposure to government debt. The macro-fiscal landscape provides critical insights into debt sustainability and fiscal capacity. These are essential for determining the country’s capability to assume new financial commitments or allocate resources for initiatives in sustainable forestry and climate adaptation. A compromised macro-fiscal stance can curtail access to market-based financing options, amplifying the necessity for concessional funding. The investment climate encapsulates the overall risk environment for private capital, while the enabling environment assesses the presence of requisite technical, institutional, and financial infrastructures to successfully implement and oversee financed initiatives. Factors such as regional dynamics, governance quality, and climate vulnerability play a pivotal role in shaping risk mitigation strategies and the structuring 38 of these financial instruments. A holistic approach to a country’s context helps to develop effective risk management strategies, boosting the chances of successful financial outcomes, and promoting sustainable economic growth. These factors, when considered together, shape the choices made regarding the selection, design, and pricing of financial instruments from the perspective of the host country and that of donors and investors. 9.3.2 Public and private investors The second set of conditions analyzes the structural design of financial instruments from the perspectives of public and private investors. These groups have different incentive structures and risk tolerances, especially in forest conservation in the RoC. Their strategies affect the choice, structure, and governance of financing instruments for forest protection and its benefits (both monetizable and nonmonetizable). Financial instruments must balance public goals with private risk-return requirements, using transparent KPIs and MRV systems to meet both investors’ needs. Public investors usually prioritize environmental and developmental outcomes. Public investors, such as multilateral development banks (MDBs) and bilateral donors, focus on non-monetary goals like ecosys- tem preservation, climate mitigation, biodiversity conservation, and global public goods. They often use concessional approaches to reduce transaction risks or provide first-loss capital. Transparency on envi- ronmental outcomes, alignment with international commitments (such as the Paris Agreement and the Sustainable Development Goals), and clear governance mechanisms are crucial. In addition, public entities may emphasize capacity-building and long-term institutional strengthening. Private investors tend to seek risk-adjusted returns and clear capital protection mechanisms. Their invest- ment decisions are influenced by credit risk, regulatory stability, and enforceability of conditions for the use of funds. Green bonds offer restricted use and post-issuance reporting to boost investor confidence. In more complex deals like debt-for-nature swaps, private capital may only engage if backed by guarantees, insurance, or blended finance solutions. Impact and visionary investors may find participation in environmental markets appealing for long-term value creation. While voluntary compliance markets (VCMs) offer modest returns and attract private capital, they are currently shallow and fragmented. For the RoC, prioritizing institutional and technical capacity to engage in future environmental markets is crucial, as demand for forest carbon and ecosystem services is expected to rise. 9.3.3 International frameworks The third set of conditions examines how existing international frameworks align with the RoC’s context and how risk-based KPIs can be integrated into these frameworks to reflect the specific needs of HFLD countries. This includes assessing the applicability of instruments governed by established standards—such as the OECD guidelines for development assistance, the Green Bond Principles (GBPs), and the Sustainability- Linked Loan Principles—and determining the extent to which these can accommodate KPIs tailored to forest conservation and climate benefits. This filter evaluates three key aspects: (1) the extent to which an international framework can be opera- tionalized within RoC’s policy and institutional architecture; (2) the framework’s capacity to incorporate risk-based KPIs that reflect forest integrity, carbon storage, and climate resilience; and (3) the role of such integration in improving the financial attractiveness and risk profile of instruments deployed in the RoC. In practical terms, each financial instrument—such as grants, concessional loans, green bonds, or debt- for-nature swaps—can be reviewed against these dimensions. For example, concessional loans may be assessed for their conformity with OECD and DAC criteria and the ease with which climate- and forest- related KPIs can be embedded into disbursement conditions. Similarly, green bonds can be evaluated based on alignment with GBPs and the capacity of national systems to support post-issuance reporting and KPI verification. The output of this assessment can inform a structured prioritization of financial instruments, ensuring that those selected for implementation are not only aligned with global standards but also tailored to RoC’s institutional and macro-fiscal realities. These principles could also include alignment with the 39 Paris Agreement and evolving opportunities through specific initiatives like the ART-TREES standards for HFLDs, the TFFF process, and initiatives from the Government of Singapore. Specifically, this filter considers how well the updated international framework enhanced by KPIs proposed in this paper fits RoC’s institutional, policy, and financing environment. 9.4 Readiness for Financial Innovation Donors and the host country must trust that the selected financial instruments for concessional finance can be effectively implemented. This depends on the country’s specific circumstances and the interventions supported by the funding. Requirements include clear environmental strategies, institutional and technical capacity to design and implement projects, robust MRV systems, and alignment of climate policies with development objectives. KPIs are crucial in the design and delivery of finance flows. For RoC, there are obvious barriers to using innovative financial instruments: low credit ratings and insufficient fiscal space complicate the use of debt instruments. However, the experience with the FCPF creates a foundation to participate in the VCM and utilize outcome- and result-based grants. The use of green and sustainability-linked financial instruments in the RoC faces constraints due to debt sustainability, macroeconomic stability, and strong governance. Debt sustainability and fiscal space affect the country’s ability to accommodate new financing and its creditworthiness. Macroeconomic stability, including growth outlook, inflation control, and external shock vulnerability, influences investor risk per- ception. Governance strength, including transparency, regulatory efficiency, and reform implementation, shapes the investment climate and execution of complex financial instruments. Expected improvements in debt management, including limiting external financing, enforcing fiscal dis- cipline, high oil prices, and recent debt restructuring, will likely enhance external liquidity and solvency indicators in RoC by 2026. Furthermore, consistent high oil prices and government reforms are anticipated to lower the public debt-to-GDP ratio and prevent new domestic arrears. In addition, the government aims to transform the country into an emerging economy through initiatives that improve investment transparency and reduce bureaucratic hurdles. These reforms should target attracting diverse investments beyond the oil sector and promoting sustainable growth.36 However, governance weaknesses, macroeconomic instability, and limited financial capacity remain key challenges in RoC. Strengthening institutions, diversifying economies, and improving governance are essential to attract investors and implement innovative financing mechanisms successfully. Creating a robust policy framework, institutional capacity, and technical expertise is crucial for sustainability-linked financing. Partnerships with development finance institutions, international organizations, and private sector players are vital for mobilizing resources for climate action. Public and private finance availability depends on a country’s macroeconomic and institutional context, which influences borrowing capacity and foreign investment attraction.37 Economic development and institutional maturity affect financial instruments’ applicability and reliability, including access to mul- tilateral finance (such as International Bank for Reconstruction and Development (IBRD), IDA, African Development Bank (ADB), and GEF, credit enhancement mechanisms, and political risk insurance from entities like the Multilateral Investment Guarantee Agency (MIGA). The RoC due to fiscal constraints and macroeconomic instability has a zero ceiling for non-concessional borrowing and is unable to access capital market financing. As a result, credit enhancement mechanisms, such as those provided by the MIGA and other forms of political risk insurance, are not applicable at this stage, as they depend on the underlying creditworthiness of the sovereign. The RoC struggles to attract FDI. Despite significant potential, especially in natural resources, consistent FDI requires political stability and a more investor-friendly climate. The business environment is challenging but offers opportunities in petroleum, energy, timber, mining, agriculture, information and communication technology, and tourism. Yet, the country ranks 180th out of 190 economies for ease of doing business, with a score of 39.5 since 2019. 36 See https://www.lloydsbanktrade.com/en/market-potential/congo/investment 37 See https://ida.worldbank.org/en/financing/debt/country/republic-congo 40 Table 10: Instruments, Roles, and KPI Considerations for Forest Finance in the Republic of Congo Adequate Level of Readiness, Work in Progress: Some Capacity Gaps, Critical Gaps: Not Available for the Foreseeable Future Prerequisite of analysis for Republic of Congo Instrument Sub-category Role/relevance of KPI Concessionality Description of Country context and Availability for Intl. financing Level Instrument polices Republic of Congo frameworks Grant Project based KPIs play a secondary Concessional Resources raised with ROC should have a National, bilateral, An internationally role and are useful in no expectation of program or project for multilateral, and recognized framework monitoring project repayment, designed transitioning to international for managing implementation when to support projects that sustainable forestry. organizations, climate project-based grants grants are disbursed in contribute to funds, private (such as GEF, GCF tranches, ensuring sustainable foundations, and project cycle, and the accountability development goals. non-governmental OECD DAC principle). throughout the project. organizations. Philanthropists willing to pay for avoided 41 deforestation. Outcome KPIs that encompass Concessional Resources raised with ROC should have Same as above Same framework as based various environmental no expectation of sustainability-linked above, but a risk-based outcomes, such as repayment; but where indicators. KPI methodology forest preservation, the disbursement of accepted by donors is biodiversity, and funds is contingent needed. avoided carbon upon the achievement emissions, play an of specific, predefined essential role in outcomes or results. tracking grant-supported activities. Result-based Along with MRV, KPIs Concessional Resources raised with RoC has experience Same as above Improvement of climate constitute a core no expectation of with FCPF: (to crediting methodology finance contractual element repayment, rewards implement JREDD+ for HFLD JREDD+. defining the financing country for the emission reduction National and terms. reduction of CO2 program; established jurisdictional emissions below MRV) that could be risk-based reference business-as-usual used to participate in lines for carbon scenarios (such as other programs (such emissions accepted by JREDD+) as SCALE to attract donors. result-based climate finance from other sources. Debt for Debt KPIs are used to Concessional Total or partial RoC has benefited Bilateral creditors. Customized debt Nature forgiveness monitor and assess the cancellation of a debt from previous debt RoC should have a forgiven agreement for outcomes of activities by the lender, relief initiatives. bilateral creditor RoC based on IMF and specified in the debt conditional on the Future eligibility for willing to cancel debt World Bank Group forgiveness agreement, achievement of agreed additional debt in exchange for the Framework Data for which defines the KPIs. forgiveness will achievement of Decision (D4D). conditions for depend on new sustainability-linked 42 forgiveness. international indicators. agreements or initiatives. Loan Sustainable KPIs serve a secondary Concessional, Resources that need to ROC should have a National, LMA Sustainable projects role: useful for semi- be repaid to the project portfolio to multilateral, bilateral, Loan Principles and monitoring the concessional provider, which can be invest loan principal in and international Green Loan Principles. sustainability used to finance, forest sustainable forestry. financial institutions. In addition, a previous outcomes outlined in preservation climate RoC is eligible for IDA loan agreement can the loan agreement mitigation and loans and concessional serve as a template. and play a role in adaptation projects loans from other IDBs accountability and and policies and development oversight rather than agencies. acting as direct triggers for disbursement. Sustainability- KPIs determine Concessional, Resources where the ROC currently lacks National, Same framework as linked interest rate semi- loan terms, such as the fiscal space to incur multilateral, bilateral, above, but a risk-based adjustments concessional interest rates, are additional debt but and international KPI methodology linked to the may qualify for this financial institutions. accepted by donors is borrower’s financing if it obtains RoC is in external debt needed. achievement of specific an appropriate credit distress. Non-concess. sustainability enhancement selling for 2025 is zero. performance targets, mechanism or and the funds are not improves its financial restricted to specific situation. projects Sustainable KPIs serve a secondary Non- Resources provided by ROC currently lacks Capital market LMA’s Sustainable projects role: useful for concessional private sector the fiscal space to incur investors. RoC is in Loan Principles and monitoring the institutions, which can additional debt but external debt distress. Green Loan Principles. sustainability be used to finance may qualify for this Non-concess. selling In addition, a previous outcomes outlined in climate mitigation and financing if it obtains for 2025 is zero. loan agreement can loan agreement. adaptation projects an appropriate credit serve as a template. and policies enhancement mechanism or 43 improves its financial situation. Bond Green, KPIs serve a secondary Non- Fixed-income ROC currently lacks Capital market The GBPs that offer Sustainable role: useful for concessional securities issued in the fiscal space to incur investors. Institutional voluntary guidelines monitoring the capital markets to raise additional debt but and individual for transparent and sustainability resources to finance may qualify for this investors are unlikely credible green bond outcomes outlined in projects and activities financing if it obtains to buy bonds issued by issuance and reporting. prospectus. related to sustainable an appropriate credit RoC. development enhancement mechanism or improves its financial situation. Sustainability- KPIs determine Non- Fixed-income ROC currently lacks Same as above Sustainability-linked linked coupon rate concessional securities issued in the fiscal space to incur Bond Principles adjustments capital markets, where additional debt but (SLBP). A risk-based the financial terms, may qualify for this KPI methodology such as interest rates, financing if it obtains accepted by donors is are linked to the an appropriate credit needed. borrower’s enhancement achievement of specific mechanism or sustainability improves its financial performance targets, situation. and the funds are not restricted to specific projects Debt for Debt swap KPIs are used to Non- Agreements between a ROC should have Bilateral and Customized debt Nature monitor and assess the concessional government and one or eligible debt for the commercial financial swap agreement for Swap outcomes of activities more of its creditors to debt swap transaction institutions. RoC RoC based on IMF and specified in the debt replace sovereign debt and an appropriate should find creditors World Bank Group swap agreement. with a liability that credit enhancement willing to forgive debt framework Data for entails a spending mechanism. or the sale at a Decision (D4D). A 44 commitment over time discount. risk-based KPI towards a development methodology accepted goal by donors is needed. Carbon OTS, spot and KPIs to track collateral Non- Investment in Expansion of FCPF ESG and impact Verra or ART-TREES finance options benefits concessional emissions reduction is ERP to handle private investors, corporations crediting methodology. (VCM and transactions eligible for VCM investment with SBTi that allows compli- transactions for offsets. The number ance of transactions with markets) HFLD countries is limited but increasing. Deforest. Private Same as above Non- Investment in MRV and attribution Same as above and Carbon border free investment by concessional deforestation-free of forest carbon exporter to EU adjustment mechanism private in- corporations production (supply) emissions to a Corporate interest in (EU), SBTi vestment chain. particular corporation, zero-deforestation carbon registry, and supply chains is certification. increasing, actual implementation remains limited. 9.5 Financial Instruments for the Republic of Congo RoC has various levels of preparedness and potential to utilize financing instruments for forest conservation. Table 10 provides descriptions of the main financial instruments and a qualitative assessment of their suitability for the RoC, conducted according to the assessment framework described in Figure 22. Deploying these instruments effectively requires enabling conditions like credible KPIs, institutional and technical capacity, investment-grade creditworthiness, and alignment with international frameworks. Key elements include the GBPs, SLB Principles, SLL Principles, and Paris Club debt restructuring protocols. Additionally, global macroeconomic factors such as interest rates, exchange rates, inflation, fiscal capacity, and political commitment from donor nations impact financial flows’ scale, concessionality, and predictability. 9.6 Grants Our assessment shows that the RoC has a high level of institutional readiness to access results-based and outcome-linked grants. These grants are non-repayable and not tied to credit ratings, but effective use requires certain readiness factors. For project-based grants, the recipient must present credible portfolio of investment projects. For outcome-based grants, they need to demonstrate MRV systems and transparent KPIs. Incorporating risk-based KPIs can enhance grant eligibility and performance-based disbursement. Priority should be given to developing a pipeline aligned with deforestation mitigation goals. 9.6.1 Debt-for-Nature Swaps and Debt Forgiveness Debt-for-nature swaps offer the RoC a chance to reduce debt and invest in forest conservation. As a lower- middle-income country, RoC can benefit from these financial strategies, given its debt distress and need for forest protection funds. Gabon’s US$500 million debt-for-nature swap on August 14, 2023, provides insights into reducing national debt while promoting conservation.38 With external debt around US$852.62 million , RoC could explore similar swaps by studying successful transactions in comparable debt situations.39 The readiness assessment shows creditors are not yet convinced to forgive RoC’s debt for forest preservation and sustainable forestry. The existing framework is unsuitable for HFLD countries, but the new KPI methodology proposed in this paper can be adapted to enhance the global debt-for-nature swap framework. 9.6.2 Green Loans and Sustainability-linked Loans Concessional green loans are a viable option for the RoC which is eligible for funds from institutions like the IDA and other international development banks. These debt instruments offer favorable terms—such as low interest rates, extended maturities, or grace periods—and can finance climate mitigation and adaptation initiatives, forest preservation and sustainability policies. The RoC can access concessional financing from institutions like the IDA and the African Development Bank. Accessing these funds necessitates a clear strategic justification. Given limited fiscal resources and the RoC’s competing development priorities, concessional lending must demonstrate substantial development benefits, environmental integrity, and alignment with national climate and forest strategies. KPIs are secondary. They help monitor the sustainability outcomes specified in the loan agreement and contribute to accountability and oversight, rather than serving as direct triggers for disbursement. According to our analysis, the RoC largely meets the prerequisites. The country context cell is marked yellow because the country has yet to develop a project pipeline and apply new KPIs to show the benefits of borrowing to support sustainable forestry. Non-concessional and semi-concessional SLBs and SLLs are currently inaccessible to the RoC due to limited fiscal space and high debt concerns. The RoC cannot take on more market-based debt without risking macroeconomic stability. Non-concessional green loans, issued at or near market terms, involve institutions like development finance institutions (such as the IBRD, ADB, the European Bank for Reconstruction and 38 Formore information on Gabon’s debt-for-nature swap, see: https://www.afronomicslaw.org/print/pdf/node/2687 39 Recentexamples include El Salvador and Gabon. In October 2024, El Salvador completed a US$1 billion debt buy-back, resulting in a savings of US$352 million for the Rio Lempa Conservation Program. In August 2023, Gabon finalized a US$500 million debt-for-nature swap, designating US$125 million for marine conservation efforts. 45 Development (EBRD), export credit agencies, and commercial banks with sustainable finance divisions). These generally provide loans to creditworthy entities. The IMF (2023) and IDA classify the RoC’s overall and external debt as in distress. The public debt-to- GDP ratio is projected to rise due to its declining debt-to-GDP ratios which are increasing government deposits at the Central Bank. However, access to financing may become possible with adequate credit enhancement mechanisms—such as political risk insurance, partial guarantees, blended finance structures, or improvements in fiscal and credit outlook. Assessing SLL readiness for the RoC is like assessing readiness for non-concessional green loans. Both require a strong macro-fiscal foundation, institutional capacity, and credible sustainability targets. Indica- tors of institutional readiness, fiscal space, and external financing capacity need further development, as shown in Table 10. The RoC must and improve governance, fiscal discipline, and macroeconomic stability to enhance creditworthiness and access non-concessional funding from the IBRD or other MDBs. The IDA does not offer SLLs. 9.6.3 Green Bonds and Sustainability-linked Bonds Currently, the RoC is not eligible for the issuance of sovereign green bonds. See Table 10. The most recent credit rating for RoC from Fitch is CCC+, that is vulnerable and dependent on favorable conditions. The negative outlook40 may affect investor confidence and borrowing costs. This is a downgrade from the pre- vious rating of B- in 2023 with a positive outlook. The CCC+ rating signals concerns about creditworthiness and potential default, while the negative outlook suggests further downgrades might occur if conditions do not improve. SLBs can be structured through third party entities, like IBRD, MDB, development finance institutions, or a special purpose vehicle. This setup enables the RoC to pursue sustainability without repayment obligations. The RoC is accountable for achieving verified environmental outcomes or KPIs, using MRV systems to SLLs. Without being the main obligor, sovereign creditworthiness is not needed, making it suitable for debt distressed countries. This model promotes performance-based climate finance without adding sovereign debt, attracting private capital through blended finance and de-risking strategies. Third-party issued SLBs, where proceeds do not benefit the issues, have limited financial impact. An example is the World Bank’s US$100 million, seven-year bond from January 2024, supporting plastic waste projects in Ghana and Indonesia (Turhan, 2024). Investors in this bond accepted a lower fixed coupon than standard World Bank bonds with similar maturities. The US$14 million difference in coupon payments was used to fund projects via a hedge transaction with Citibank. Investors receive annual coupons consisting of a fixed amount plus payments from the sale of plastic and carbon credits generated by the projects. This innovative structure highlights the difficulties in scaling financing models, as the US$14 million allocated to project financing is modest compared to the US$100 million principal. 9.6.4 Carbon and Climate Finance Carbon and climate finance, including grants, concessional and non-concessional debt instruments, and debt relief instruments like forgiveness and debt for nature swap. These funds can support specific programs or projects, pay for outcomes, or results-based financing. The country’s readiness is shown by its participation in the FCPF and Forest Carbon Fund (FCF) along with its emission reduction program, which supports effective MRV systems and verifies emissions reductions. Global financing provisions are limiting growth in carbon market opportunities. Forest carbon finance faces constraints due to underdeveloped markets and limited public funding impacted by economic pressures and geopolitical instability. The current framework awards forest emissions reductions based on historical baselines, disadvantaging HFLD countries like RoC with low historical emissions. Transitioning to risk- based reference lines and KPI frameworks could address this bias and should be pursued vigorously. Outcome-based finance offers payments to incentivize forest preservation and biodiversity protection. Bonds and loans leveraged by these payments give countries upfront capital, enabling them to start envi- 40 See https://www.fitchratings.com/research/sovereigns/fitch-affirms-republic-of-congo-at-ccc-28-02-2025?utm 46 ronmental initiatives sooner. KPIs are crucial for tracking progress for these results-based payments that support sustainable growth, job creation, and effective forest management. Results-based climate finance also mobilizes funding. Congo Basin countries, like the RoC, can get upfront financing to join these programs which focus on emissions reductions, forest management, and climate resilience. Early-stage funding from the FCPF and the Congo Basin Forest Fund builds systems to access global carbon markets. Strengthening MRV systems is for obtaining funds to protect their forests. 9.7 Advancing Performance-Linked Financing in the Republic of Congo Based on our analysis, we categorize the instruments into three distinct groups. 1. Readily available instruments—The RoC is well-positioned to immediately access project-based grants, outcome-based grants, and results-based climate finance (such as JREDD+). These instruments are non- repayable and benefit from established institutional readiness and effective MRV systems. 2. Instruments dependent on near-term improvements—Debt-for-nature swaps and concessional green or SLLs may be- come viable with further groundwork. These require the development of project pipelines, risk-based KPIs, and negotiations with potential donors. 3. Currently inaccessible instruments—Non-concessional instruments, such as green bonds, SLBs, and commercial loans, remain out of reach due to debt distress and a poor sovereign credit rating. De-risking and credit enhancement tools can boost forest, climate and carbon finance in the RoC. These mechanisms reduce investment risks, enhance creditworthiness, and secure future cash flows. They lower capital costs, attract private sector participation, and strengthen the viability of emissions reduction projects. Leveraging these tools is crucial for mobilizing sustainable finance and supporting the nation’s low-carbon development path. These financial instruments can increase investor confidence, lower financing costs, and attract private capital for climate mitigation and forest conservation. Risk-based KPIs are crucial for structuring these instruments, credit enhancement solutions, and risk hedging mechanisms. They enable the collateralization of expected revenues from results-based financing and environmental market transactions, helping to mobilize upfront investment capital. Upfront financing is essential for Congo Basin nations to enhance technical capacity for climate finance programs. Enhancing MRV systems and regulatory frameworks will aid in accessing global climate finance, like carbon markets and the GCF. Collaborations with institutions like the World Bank can support developing governance structures and technical skills for securing more funding. Bonds backed by future revenues from forests can provide initial investments, while forward contracts ensure stable income if KPIs are met. RoC should focus on building capacity for results-based climate finance and increasing high-integrity deforestation credits, despite limited immediate financial returns in the short-term. 10 Conclusion Successfully mobilizing climate finance to reduce deforestation outcomes first requires the establishment of a robust, sustainable financing framework. This framework must articulate: (1) what investors (and donors) are paying for; (2) how outcomes are measured; and (3) what the issuer commits to achieve. Using such a framework before mobilizing climate finance will ensure clarity and trust in the transaction. In this paper, we explain how the issuer’s performance can be isolated, by comparing observed deforestation outcomes with modeled deforestation pressure. Employing the principles of REACH will result in a KPI that is more robust to factors beyond the influence of the government. Because the benchmark of modeled deforestation pressure controls for the influence of climatological, demographic, and macroeconomic factors, these exogenous factors have less influence on the KPI that measures the government’s performance. Instead, the KPI more closely tracks the impact of the passing and successful implementation of forest-related policies by the issuer. The FAB framework helps set sustainability performance targets (SPTs) that provide an ambitious improve- ment beyond a BAU scenario while maintaining feasibility when judged against peer countries’ historical 47 performance. By forecasting a BAU scenario based on our model of deforestation pressure, we show that deforestation pressure is expected to increase significantly between 2023 and 2030 due to changing clima- tological, demographic, and global economic conditions. A benchmark based only on historical reference levels would be 46 percent lower than expected deforestation in our BAU scenario, highlighting how his- torical reference levels unfairly raise the bar for HFLD countries and would result in insufficient financial incentives to stop deforestation. With reliable KPIs and SPTs in place, issuers can explore a variety of financing forms to advance sustainabil- ity goals. SLBs and SLLs, as well as other state-contingent debt instruments, are particularly advantageous due to their ability to tie financing costs to performance outcomes. These instruments allow for adjust- ments of financial terms based on the achievement of predefined criteria. Debt-for-nature swaps can be structured such that debt relief is contingent upon meeting specific KPIs, thereby incentivizing efforts towards development and sustainability goals. Moreover, the benchmark models underpinning these KPIs can be innovatively leveraged to mobilize climate financing and address the climate funding gap through mechanisms that provide flexible investment opportunities that can be adjusted based on evolving conditions. Our paper proposes a methodology to define KPIs and SPTs and can serve as a basis for a constructive discussion about how exactly an instrument should be structured and priced—but some aspects need to be covered in more detail before a potential issuance. First, a complete sustainable financing framework needs to elaborate on the way to achieve the proposed targets, both in terms of additional policies and in terms of improved implementation of existing policies. Second, the structure and pricing of the instrument needs to be adjusted to the RoC’s ability to raise money on international capital markets. Third, the suitability of the geospatial data used in this paper’s assessment as a basis for a MRV system needs to be validated to the satisfaction of the government and investors. The application presented in this paper should be interpreted as a proof-of-concept and future research can improve the robustness of the findings. This paper focused only on deforestation pressure in the RoC and the methodology can be made more robust by conducting analytics across several HFLD countries in a panel data framework. This would also increase the number of observations and allow an empirical estimation of global deforestation leakage between deforestation hotspots. Another promising direction for future research is to disentangle the various land cover transitions that are currently grouped under the term ‘deforestation’. Advances in satellite data now allow for detailed classification of land cover types making it possible to distinguish how previously deforested land is being used today (Borlaf-Mena, Mérida- Floriano, et al., 2025; Garcia et al., 2024). This is useful to estimate the factors influencing deforestation more accurately, build better predictions and provide more actionable policy recommendations. The efficacy and impact of forest finance, through performance-linked financing, carbon crediting mecha- nisms, and sustainability-linked financing instruments are not only a function of government actions but require support from the international community. Public capital, particularly from development finance institutions and multilateral organizations, already plays a vital role in ongoing conservation projects. However, tapping into private capital is becoming increasingly important, especially with the growing emphasis on ESG investing. Public-private partnerships can be particularly effective in this context, where public capital is utilized to lower entry barriers and mitigate risks for private investors. This fosters greater private sector participation and can mobilize capital that would otherwise not be accessible. Anticipating the rising demand for nature-based solutions and the need to mobilize its financing gap, the RoC must act now to strengthen its financial and institutional capacity. Prioritizing grant-based and result-based finance while preparing for scalable concessional instruments will allow RoC to unlock forest and climate finance without exacerbating fiscal risks. 48 A Appendix A.1 Existing Policy Targets and Relevance to Forest Management National policies, projects and specific events related to deforestation can explain strong deviations from the benchmark level. Table ?? and Table ?? show national policies and law enforcements for the forest sector in Congo during the time span of our research, 2001 to 2023. Also see Figure 24. A.2 Alternative Data Sources A.2.1 Reforestation and Afforestation While deforestation and forest losses have been limited in Congo, attention could be given to recovery and forest expansion efforts. Reforestation, defined as post-deforestation recovery, can occur naturally or through human-driven efforts, resulting in secondary forests. Afforestation is entirely anthropogenic, creating tree or forest cover in areas previously occupied by other land cover types. Tracking these processes is highly relevant in rewarding agroforestry initiatives, which are a priority in Congo’s policy agenda to enhance food security, support climate mitigation, and reduce emissions without adversely affecting GDP. Measuring reforestation and afforestation is more challenging than tracking tree loss. It usually takes at least three years for young forests to become visible in space imagery, and there is uncertainty in the survival rate of new trees. Because of this complexity, GFW provides a dataset that maps tree cover gain between 2000 and 2020 rather than annual changes. This map includes all tree areas that met the 5-meter threshold in 2020 but did not meet it before, in 2000. Here, tree cover gain can be the result of regeneration after natural disturbances, human reforestation and restoration efforts, and regrowth in tree plantations. RoC gained 113,000 hectares of tree cover in this dataset. More detailed is the forest regrowth in the annual changes set of the TMF dataset. Post-deforestation recovery and afforestation have been separately reported annually since 1990. Between 2000 and 2020, RoC showed 123,000 hectares of reforestation in the tropical moist forest domain, and 139,000 hectares afforestation. A.2.2 Geospatial Data An alternative to global satellite-based datasets covering multiple countries with a harmonized methodol- ogy, is using customized data developed as part of the financial instrument’s MRV system. This has the added benefit of not only providing a reliable data source reporting on the KPI but also generates additional data to be used by stakeholders such as the government or forestry experts. It can also involve a technical assistance component supporting the government in making the most use out of the product. Such customization can be useful to address idiosyncratic issues in the global satellite data such as natural forest degradation and deforestation due to road development. For example, the paving of the N2 road from Brazzaville to Ouesso is partly picked up as deforestation in the TMF data. While this may be true (if the existing road was widened), it might be better to exclude such deforestation due to road development from the performance metric or at least specify it separately. This is possible when customizing the satellite data by, for example, excluding pixels that align with the road network. Additionally, more tailored satellite data can be used to define deforestation by its land use and land use cover change. For example, it is possible to target ‘deforestation rates due to cropland expansion’ rather than any type of deforestation. Combining satellite data with machine learning classification algorithm that are tailored to the Congolese context can help identify land covers associated to, for example, emerging plantations, mining operations or logging infrastructure. For the Congo Basin, an ongoing collaboration with the ESA is currently developing such a data product which provides sub-annual land cover transitions between 2000 and 2020. The mapped classes are seasonal and permanent water, wetlands (grassland, shrubs, open and closed forest), bare soil, shrubs, grasslands, open and closed forest, croplands and settlements. Figure A25 shows a preliminary land cover classification for six example tiles (1 x 1 degree) in December 2022. Transition matrixes can be created from these land 49 Table 11: Results of Baseline Models for Forest Degradation Pressure Model 1 Model 2 Model 3 Model 4 Variable Climate and + Agroecon. + Demographic + Policies and econ. factors factors factors events Lagged relative degradation rate, t–1 0.173** 0.084** 0.073** 0.005 Climatological factors SPEI-Index (12-month), t –0.353** –0.488** –0.226** –0.420** Economic factors Real effective exch. rate (%Δ), t–1 –0.296** –0.430** –0.172** –0.469** Real effective exch. rate (%Δ), t–2 0.037 0.013 n.s. n.s. Round- and sawnwood price (US$), t–1 0.043 0.007** n.s. n.s. Round- and sawnwood price (US$), t–2 0.171** 0.406** 0.029** 0.342** Cash-crop prices (US$), t–1 . –0.472** n.s. –0.557** Cash-crop prices (US$), t–2 . 0.274** 0.070** 0.366** Demographic factors Urbanization rate, t . . n.s. n.s. Urbanization rate, t–1 . . n.s. n.s. Urbanization rate, t–2 . . n.s. n.s. Rural population growth rate, t . . n.s. n.s. Rural population growth rate, t–1 . . n.s. –0.042 Rural population growth rate, t–2 . . n.s. n.s. Forest-related policies and singular events Improved satellite quality after 2013, t n.s. n.s. n.s. n.s. El Niño / wildfires 2016, t . . . 1.426** N2 Road Project 2010 onwards, t . . . n.s. Policy dummies No No No Yes Adjusted 2 0.27 0.46 0.13 0.77 Note: n.s. = not selected: The variable was included as a candidate but not selected during the regularization step. Significance levels were based on residual bootstrapping, showing significance at a 5% (**) and 10% (*) level. Model 4 allows for policy dummies for the Brazzaville declaration (2018), the CAFI letter of Intent (2019) and the updated Forest Code (2020) to be included in the estimation. The 2020 dummy was negative and significant, the others were not selected by the data-driven regularization step. The adjusted 2 accounts for the number of included regressors after the regularization step. The dependent variable was “Total degradation (followed or not by deforestation)” from TMF (European Commission 2024). cover classifications that show the transition of the closed tree cover class to any of the other land cover classes. A.3 Reducing and Preventing Forest Degradation In the main analysis we focused on tree losses that were sustained for a period of more than 2.5 years, after which TMF considers an area to be deforested as opposed to degraded (Vancutsem et al., 2021). In principle, our model is also suitable to assess forest degradation. However, we need to account for the fact that factors related to forest degradation may be different from factors related to deforestation. As a proof-of-concept, this section presents the results of our baseline model for forest degradation using TMF data. We included a dummy to account for the addition of the Landsat 8 satellite in 2013 which significantly improved data coverage in Central and West Africa. This means that when considering a longer time series from 2001-2022, short-duration degradation can be underestimated before 2013. The issue mostly affects short-term forest degradation and not longer-term deforestation, which is why we did not include the dummy in our baseline results. 50 Figure 23: Model Approach 51 Figure 24: BAU Projections for Deforestation with Hypothetical Reference Level Source: Authors’ calculations. The variable of interest was “Total deforestation (direct or after degradation)” from TMF (European Commission, 2024) relative to the total area of undisturbed forest cover in the year 2000. The forecast is an out-of-sample prediction based on the model of deforestation pressure presented in Section 3 and ARIMA forecasts of the climatological, demographic and economic factors that are inputs into the model. The hypothetical 5-year reference period is calculated by extrapolating the historical deforestation rates of the past five years (2019-2023) into the future without considering changing climatological, demographic and economic factors. Figure 25: Preliminary Land Cover Classification Tiles, December 2022 Visualization provided by GMV Innovating Solutions in collaboration with the authors and the ESA. Example shows preliminary output of an ongoing collaboration under ESA’s GDA program. Source: GMV, ESA 52 A.4 Additional Results for FAB Assessment Figure 26: Feasibility Intervals for Deforestation Rates The feasibility analysis shows how peer countries deforestation rates have developed in the past, in the years following similar deforestation rates as RoC has today. The feasibility intervals are based on the distribution of the peers’ trajectories. For example, the 50 percent feasibility interval means that half of all peer-paths correspond to deforestation rates within that range. Source: Authors’ calculations. The variable of interest was “Total deforestation (direct or after degradation)” from TMF (European Commission 2024) relative to the total area of undisturbed forest cover in the year 2000. Selected peer countries were Congo Basin or HFLD countries with available TMF data and that had similar deforestation rates as ROC (below 0.15 percent) at any point in the past 20 years. Feasibility intervals were calculated based on deforestation rates and translated into hectares by using RoC’s total undisturbed forest cover in the year 2000. Figure 27: Accumulated Deforestation of Peers — Worst Historical Trajectories The feasibility analysis shows how peer countries deforestation rates have developed in the past, in the years following similar deforestation rates as RoC has today. The feasibility intervals are based on the distribution of the peers’ trajectories. For example, the 50 percent feasibility interval means that half of all peer-paths correspond to deforestation rates within that range. Source: Authors’ calculations. The variable of interest was “Total deforestation (direct or after degradation)” from TMF (European Commission 2024) relative to the total area of undisturbed forest cover in the year 2000. The worst historical trajectories of peer countries were accumulated over seven years and the quantiles of the distribution plotted in the graph. Selected peer countries were the Congo Basin or HFLD countries with available TMF data and that had similar deforestation rates as RoC (below 0.15 percent) at any point in the past 20 years. Deforestation rates were translated into hectares by using RoC’s total undisturbed forest cover in the year 2000. 53 A.5 Additional Graphs Figure 28: Forest Transition Curve for Congo Basin Countries The RoC is in Stage 1 (relatively undisturbed forests), but pressure on forests is increasing significantly and has already led to a decade of higher deforestation rates in other Congo Basin countries. Reforestation efforts in Stage 4 can usually not revert the losses of ecosystem services and biodiversity from previous deforestation. Understanding factors contributing to pressure on forests coupled with climate finance is therefore needed to support countries leapfrog the high deforestation phase of economic development and prevent deforestation from increasing in the first place. Source: Megevand 2013. The orange line represents the hypothesized forest cover under the forest transition curve theory. 54 Figure 29: Share of Access to Clean Cooking between 2000 and 2022 Access to clean fuels and technologies for cooking is below five percent for the rural population and has hardly increased between 2000 and 2022. In the total population, access to clean cooking has increased more, to about 40 percent in 2022, driven by adoption in urban areas. Source: Author’s calculations based on data from the World Bank’s WDIs and TMF (European Commission 2024). The WDI database ID’s for access to clean fuels and technologies for cooking are EG.CFT.ACCS.ZS (total) and EG.CFT.ACCS.RU.ZS (rural). Deforestation was measured as “total deforestation (direct or after degradation)” from the TMF database. Figure 30: Agricultural Exports for Wood, Cocoa, Coffee, Cocoa and Other Products Source: Share of wood, sugarcane, coffee and cocoa exports relative to the total agricultural exports in Congo, 2021. Data from Atlas of Economic Complexity (Harvard Growth Lab 2024), visualization by authors. 55 Figure 31: Global Export Prices for Gold and Copper Source: Author’s calculations. Gold was the producer price index (FRED ID: WPU10210501), and copper was the global export price (FRED ID: PCOPPUSDM). Deforestation was measured as “total deforestation (direct or after degradation)” from the TMF Country Statistics (European Commission 2024). Figure 32: Cross Validation Results for Congo, Model 4 Source: Author’s calculations. The cross validation splits the time series into five folds which consecutively used a larger share of the data for training and a smaller share for validation purposes. This was done to keep the time series structure of the data intact, as a random k-fold validation would not account for the temporal autocorrelation in the time series. 56 Figure 33: Deforestation and Degradation Rates in Sangha Concession Areas Before and After the Concession Entered into Production Source: Author’s calculation based on data from TMF (European Commission 2024). Deforestation was “Total deforestation (direct or after degradation)”, degradation was “Total degradation (followed or not by deforestation)”. Rates were relative to the “Undisturbed Tropical Moist Forest” area in the year 2000. Figure 34: Deforestation and Degradation near Liouesso, Showing the N2 road and a Sideroad Towards the Liouesso Hydroelectric Power Station Source: Forest cover maps: Annual Change Collection TMF (European Commission 2024). Satellite image: EU, contains modified Copernicus Sentinel data 2024, processed with EO browser. 57 Figure 35: Observed Deforestation Rates Compared to Modeled Deforestation Pressure in Sangha Source: Author’s calculations. The modeled deforestation pressure was based on Model 1 including dummy variables to control for singular events. The bars plotted the observed deforestation rate using “Total deforestation (direct or after degradation)” from the TMF dataset. We also normalized deforestation using the undisturbed tropical moist forest cover in the year 2000. The blue distributions represent the predicted deforestation pressure due to demographic, economic and climatological factors. Confidence intervals around the predictions were based on a residual bootstrapper. The colors of the bar represent whether deforestation outcomes were higher or lower than expected given the modeled deforestation pressure. Figure 36: Observed Deforestation Rates Compared to Modeled Deforestation Pressure in Likouala Source: Author’s calculations. The modeled deforestation pressure was based on Model 1 including dummy variables to control for singular events. The bars plotted the observed deforestation rate using “Total deforestation (direct or after degradation)” from the TMF dataset. We also normalized deforestation using the undisturbed tropical moist forest cover in the year 2000. The blue distributions represent the predicted deforestation pressure due to demographic, economic and climatological factors. Confidence intervals around the predictions were based on a residual bootstrapper. The colors of the bar represent whether deforestation outcomes were higher or lower than expected given the modeled deforestation pressure. 58 References Architecture for REDD+ Transactions. (2024). ART Earns Core Carbon Principle (CCP) Approval for TREES Crediting Level from the Integrity Council for the Voluntary Carbon Market (ICVCM) [Accessed: 2025-05-23]. Beekmann, M., Gallois, S., & Rondinini, C. (2024). Uncertain future for Congo Basin biodiversity: A system- atic review of climate change impacts. Biological Conservation, 297, 110730. Beguería, S. (2022, January). sbegueria/SPEIbase: Version 2.7 [DOI: 10.5281/zenodo.5864391]. Borlaf-Mena, I., Chunet, A., Doménech, C., Fernandez-Carrillo, A., Merida-Floriano, M., Uehara, C. C. S., & Wang, D. (2025, January). Peering through the clouds in the Peruvian Amazon: uncovering sub-annual land cover transitions [Published Jan. 6, 2025; accessed July 16, 2025]. Borlaf-Mena, I., Mérida-Floriano, M., Fernandez-Carrillo, A., Domenech, C., Wang, D., & Albergel, C. (2025). Earth Observation-based Model for Accurate and Timely Land Use Land Cover Determination for Sustainability-Linked Financing Instruments [Forthcoming]. Busch, J., & Ferretti-Gallon, K. (2017). What Drives Deforestation and What Stops It? A Meta-Analysis [publisher: The University of Chicago Press]. Review of Environmental Economics and Policy, 11(1), 3–23. Busch, J., & Ferretti-Gallon, K. (2023). What Drives and Stops Deforestation, Reforestation, and Forest Degradation? An Updated Meta-analysis. Review of Environmental Economics and Policy, 17(2), 217– 250. Central African Forest Initiative & Republic of Congo. (2019, March). Letter of Intent on the establishment of a long-term partnership to implement the Investment Plan of the National REDD+ Strategy [[Online; accessed 2024-12-18]]. Climate Policy Initiative. (2024). Landscape of Climate Finance in Africa 2024 (tech. rep.) ([Online; accessed 2024-12-18]). Crezee, B., Dargie, G. C., Ewango, C. E. N., Mitchard, E. T. A., Emba B., O., Kanyama T., J., Bola, P., Ndjango, J.-B. N., Girkin, N. T., Bocko, Y. E., Ifo, S. A., Hubau, W., Seidensticker, D., Batumike, R., Imani, G., Cuní-Sanchez, A., Kiahtipes, C. A., Lebamba, J., Wotzka, H.-P., . . . Lewis, S. L. (2022). Mapping peat thickness and carbon stocks of the central Congo Basin using field data [publisher: Nature Publishing Group]. Nature Geoscience, 15(8), 639–644. Dargie, G. C., Lewis, S. L., Lawson, I. T., Mitchard, E. T. A., Page, S. E., Bocko, Y. E., & Ifo, S. A. (2017). Age, extent and carbon storage of the central Congo Basin peatland complex [publisher: Nature Publishing Group]. Nature, 542(7639), 86–90. Darvas, Z. (2012). Real effective exchange rates for 178 countries: a new database. https://www.bruegel.org/working- paper/real-effective-exchange-rates-178-countries-new-database. Darvas, Z. (2021). Timely measurement of real effective exchange rates [[Online; accessed 2024-12-11]]. https://www.bruegel.org/sites/default/files/private/wp_attachments/WP-2021-15-231221-1.pdf. Department of Public Information, Guyana. (2022, December). Sale of Carbon Credits: $157B for Investments in Low-Carbon Development Across Guyana [Accessed: 2025-05-25]. Eba’a Atyi, R., Hiol Hiol, F., Lescuyer, G., Mayaux, P., Defourny, P., Bayol, N., Saracco, F., Pokem, D., Sufo Kankeu, R., & Nasi, R. (2022, November). The Forests of the Congo Basin: State of the Forests 2021 [DOI: 10.17528/cifor/008700]. Erickson-Davis, M. (2016a). Massive Wildfire Rips Through Congo Rainforest – Is Logging to Blame? [Accessed: 2025-06-23]. Mongabay. 59 Erickson-Davis, M. (2016b, March). Massive wildfire rips through Congo rainforest – is logging to blame? [Mongabay; accessed July 16, 2025]. https://news.mongabay.com/2016/03/massive- wildfire- rips-through-congo-rainforest-is-logging-to-blame/ European Commission. (2024). Tropical forest status and dynamics of deforestation and forest degradation - Congo [[Online; accessed 2024-12-11]]. FAO. (2020). Global Forest Resources Assessment 2020 [[Online; accessed 2024-10-25]]. FCPF & World Bank Group. (2018). Emission Reductions Programm Document in Sangha and Likouala [[Online; accessed 2024-11-04]]. Fonseca, G. A. B., Rodriguez, C. M., Midgley, G., Busch, J., Hannah, L., & Mittermeier, R. A. (2007). No Forest Left Behind [publisher: Public Library of Science]. PLOS Biology, 5(8), e216. Forest Declaration Assessment Partners. (2023, October). Theme 3: Assessing progress on forest finance [Slide deck]. Forest Declaration Assessment. Funk, J. M., Aguilar-Amuchastegui, N., Baldwin-Cantello, W., Busch, J., Chuvasov, E., Evans, T., Grif- fin, B., Harris, N., Ferreira, M. N., Petersen, K., Phillips, O., Soares, M. G., & van der Hoff, R. J. (2019). Securing the climate benefits of stable forests [publisher: Taylor & Francis _eprint: https://doi.org/10.1080/14693062.2019.1598838]. Climate Policy, 19(7), 845–860. Garcia, A. S., Silvestrini, R. A., Batista, A. M., Ferreira, L., Hanusch, M., Kollenda, P., Uehara, C. C. S., & Wang, D. (2024, April). Spatiotemporal Scenarios for Deforestation in Brazil’s Legal Amazon (tech. rep.) (Technical Note; April 2024). Instituto de Pesquisa Ambiental da Amazônia (IPAM). Brasília, Brazil. Geist, H. J., & Lambin, E. F. (2002). Proximate Causes and Underlying Driving Forces of Tropical Defor- estation: Tropical forests are disappearing as the result of many pressures, both local and regional, acting in various combinations in different geographical locations. BioScience, 52(2), 143–150. Grossman, S. J., & Hart, O. D. (1980). Take-or-Pay Contracts and Performance Bonds. Quarterly Journal of Economics, 94(4), 829–849. Grossman, S. J., & Hart, O. D. (1986). The Costs and Benefits of Ownership: A Theory of Vertical and Lateral Integration. Journal of Political Economy, 94(4), 691–719. Hansen, M. C., Potapov, P. V., Moore, R., Hancher, M., Turubanova, S. A., Tyukavina, A., Thau, D., Stehman, S. V., Goetz, S. J., Loveland, T. R., Kommareddy, A., Egorov, A., Chini, L., Justice, C. O., & Town- shend, J. R. G. (2013). High-Resolution Global Maps of 21st-Century Forest Cover Change [pub- lisher: American Association for the Advancement of Science]. Science, 342(6160), 850–853. Hanusch, M. (Ed.). (2023). A Balancing Act for Brazil’s Amazonian States: An Economic Memorandum. The World Bank. Harvard Growth Lab. (2024). The Atlas of Economic Complexity. Hausmann, R., Hidalgo, C. A., Bustos, S., Coscia, M., Chung, S., Jimenez, J., Simoes, A., & Yildirim, M. A. (2011). The Atlas of Economic Complexity: Mapping Paths to Prosperity [[Online; accessed 2024-12-11]]. Hoare, A. (2015). Tackling Illegal Logging and the Related Trade. https://www.chathamhouse.org/sites/default/files/pub International Monetary Fund. African Dept. (2024). Republic of Congo. IMF Staff Country Reports, 2024(251), 1. Krutu of Paramaribo Joint Declaration on HFLD Climate Finance Mobilization (tech. rep.). (2019) ([Online; accessed 2024-10-25]). Maniatis, D. (2024). High Forest, Low Deforestation Areas: Potential Incentive Structures and Business Models (tech. rep.). The Nature Conservancy. 60 Megevand, C., Mosnier, A., Hourticq, J., Sanders, K., Doetinchem, N., & Streck, C. (2013, January). Deforesta- tion Trends in the Congo Basin: Reconciling Economic Growth and Forest Protection [DOI: 10.1596/978- 0-8213-9742-8]. The World Bank. Mittermeier, R. A., Mittermeier, C. G., Brooks, T. M., Pilgrim, J. D., Konstant, W. R., da Fonseca, G. a. B., & Ko- rmos, C. (2003). Wilderness and biodiversity conservation [PMID: 12930898 PMCID: PMC193557]. Proceedings of the National Academy of Sciences of the United States of America, 100(18), 10309–10313. Nesha, K., Herold, M., Reiche, J., Masolele, R. N., Hergoualc’h, K., Swails, E., Murdiyarso, D., & Ewango, C. E. N. (2024). An assessment of recent peat forest disturbances and their drivers in the Cuvette Centrale, Africa [publisher: IOP Publishing]. Environmental Research Letters, 19(10), 104031. Núñez del Prado, I., Blackham, G., Moura Costa, P., & Moura Costa, M. (2023). Practical guide to conducting due diligence of tropical timber products - Republic of Congo (tech. rep.) ([Online; accessed 2024-11-04]). BVRio Institute. Rio de Janeiro, Brazil. Republic of Congo. (2017). Niveau D’Emissions De Reference Pour Les Forets (Nerf) De La Republique Du Congo [[Online; accessed 2024-12-11]]. Republic of Congo. (2021). Contribution Determinee Au Niveau National (Cdn) De La Republique Du Congo [[Online; accessed 2024-12-11]]. Simon, S., Bhanti, M., O’Sullivan, R., Sohngen, B., Dyck, M., & Pearson, T. (2021). Options for Conserving Stable Forests (tech. rep.) ([Online; accessed 2025-02-07]). World Bank. Washington, D.C. Sims, M., Potapov, P., Vancutsem, C., Bourgoin, C., Achard, F., & Carter, S. (2024, January). Differences Between Global Forest Watch’s Tree Cover Loss Data and JRC’s Tropical Moist Forest Data Explained [Global Forest Watch Blog; published January 8, 2024; accessed July 16, 2025]. https://www.globalforestwatch. org/blog/data-and-tools/tree-cover-loss-and-tropical-moist-forest-data-compared/ Stern, D. I. (2004). The Rise and Fall of the Environmental Kuznets Curve. World Development, 32(8), 1419– 1439. Teo, H., Sarira, T., Tan, A., & Koh, L. (2024). Charting the Future of High Forest Low Deforestation Jurisdic- tions. Proceedings of the National Academy of Sciences, 121(37), e2306496121. Tibshirani, R. (1996). Regression Shrinkage and Selection via the Lasso [publisher: [Royal Statistical Society, Oxford University Press]]. Journal of the Royal Statistical Society. Series B (Methodological), 58(1), 267– 288. Townsend, R. M. (1979). Optimal Contracts and Competitive Markets with Costly State Verification. Journal of Political Economy, 87(6, Part 1), 1347–1372. Transparency International. (2021). Corruption Perceptions Index 2021 (tech. rep.) ([Online; accessed 2024-12- 11]). Turhan, B. (2024, March). Case Study: Plastic Waste Reduction–Linked Bond (tech. rep.) (Case Study; published 26 March 2024). International Bank for Reconstruction and Development, World Bank. Washington, DC. United Nations, Department of Economic and Social Affairs, Population Division. (2019). World Urbanization Prospects: The 2018 Revision [ST/ESA/SER.A/420]. United Nations. van der Ploeg, F. (2011). Natural resources: Curse or blessing? Journal of Economic Literature, 49(2), 336–420. Vancutsem, C., Achard, F., Pekel, J.-F., Vieilledent, G., Carboni, S., Simonetti, D., Gallego, J., Aragão, L. E. O. C., & Nasi, R. (2021). Long-term (1990–2019) monitoring of forest cover changes in the humid tropics [publisher: American Association for the Advancement of Science]. Science Advances, 7(10), eabe1603. Venables, A. J. (2016). Using natural resources for development: Why has it proven so difficult? Journal of Economic Perspectives, 30(1), 161–184. 61 Wang, D., Hanusch, M., Gurhy, B., & Kollenda, P. (2023). Could Sustainability-Linked Bonds Incentivize Lower Deforestation in Brazil’s Legal Amazon? (World Bank Policy Research Working Paper No. [forthcom- ing]) (forthcoming). World Bank Group. World Bank. (2023, April). Strengthening Public Finances for Inclusive Growth and Sustainable Development [DOI: 10.1596/39721]. Washington DC. World Bank. (2024, January). Project appraisal document on a proposed forest investment program grant [...] and credit [...] for a northern Congo agrocforestry project [[Online; accessed 2024-12-18]]. World Bank Group. (2023). Republic of Congo Country Climate and Development Report - Diversifying Congo’s Economy: Making the Most of Climate Change (tech. rep.) ([Online; accessed 2024-10-25]). World Bank. Washington, DC. World Bank Group, Central African Forest Initiative, & Climate Investment Funds. (2025). Forest Ecosys- tem Accounts for Republic of Congo 2000-2020: Ecosystem Extent Accounts, Forest Ecosystem Condition Accounts, Forest Ecosystem Services and Asset Accounts. Wunder, S. (2003). When the Dutch Disease met the French Connection: Oil, Macroeconomics and Forests in Gabon (tech. rep. No. 1406). Center for International Forestry Research. Bogor, Indonesia. Zhang, Q., Justice, C. O., & Desanker, P. V. (2002). Impacts of simulated shifting cultivation on deforestation and the carbon stocks of the forests of central Africa. Agriculture, Ecosystems & Environment, 90(2), 203–209. Zou, H., & Hastie, T. (2005). Regularization and Variable Selection Via the Elastic Net. Journal of the Royal Statistical Society Series B: Statistical Methodology, 67(2), 301–320. 62