Climate finance and biodiversity: an institutional perspective

Climate change and biodiversity loss are increasingly recognized as interconnected global crises rather than separate environmental challenges (Reid, 2006; Zhou and Almond, 2025). Ecosystem degradation weakens carbon storage and natural climate regulation, while climate change accelerates habitat loss and species decline. Addressing these linked problems requires substantial financial resources for mitigation, adaptation and environmental protection (García and Moros, 2025; Waldron et al., 2017). In response, climate-related funding has expanded through market-based instruments such as carbon offset markets and institutional climate finance. Although both support climate action, they operate through different institutional structures and incentive systems. Their ecological consequences therefore need to be assessed separately rather than treating them as equivalent forms of climate-related finance.

Market-based instruments make this issue visible. Voluntary carbon offsets convert emission reductions and ecosystem services into tradable credits purchased by firms seeking to compensate for emissions. Many projects claim biodiversity co-benefits, but recent evidence questions whether these claims translate into ecological gains (Zeng et al., 2026). Zhou and Almond (2025) found that voluntary carbon offset projects are associated with a 3.7% increase in habitat disturbance, with no consistent habitat improvement across certifications, stated co-benefits, protected-area overlap, registry status or rating coverage. This does not imply that carbon storage and biodiversity are inherently incompatible; it shows that biodiversity co-benefits should not be assumed.

Institutional climate finance differs from voluntary carbon offsets in purpose and governance. It is channelled through governments, multilateral institutions and development agencies to support mitigation and adaptation in recipient countries (Giglio et al., 2021). Unlike voluntary carbon credits, which are organized around tradable compensation claims (Perino, 2024), institutional climate finance is embedded in public planning and development cooperation. However, its biodiversity effect remains an empirical question because reforestation, restoration, adaptation planning, infrastructure and renewable-energy expansion may have different ecological consequences (Poyser, 2026).

This distinction motivates our research question. While voluntary carbon offsets may worsen biodiversity outcomes (Zhou and Almond, 2025), far less is known about the ecological effects of institutional climate finance. We address this gap using a panel of 132 countries and 2,413 country-year observations from 2002 to 2020. Biodiversity is measured using two complementary indicators. The Human Influence Index (HII) captures anthropogenic pressure through land-use intensity, infrastructure, accessibility and population density (Zhou and Almond, 2025). The Red List Index (RLI), based on the IUCN Red List of Threatened Species, tracks aggregate extinction risk (IUCN, 2001).

Our baseline specifications relate climate-related finance in year $t$ to biodiversity indicators observed in $t+1$⁠. The one-year lead establishes a consistent temporal ordering within the annual panel, but it does not imply that ecosystem pressure or extinction risk adjusts within one year. Ecological changes may emerge gradually across financing channels and biodiversity dimensions. We therefore supplement the baseline with distributed-lag diagnostics over one-, three- and five-year horizons and interpret these estimates as suggestive evidence on temporal consistency rather than as identification of biological adjustment speed.

Our empirical strategy uses voluntary carbon-credit activity as a country-level market-based benchmark for interpreting the institutional climate-finance results. Voluntary carbon-credit markets are built around tradable carbon units, project certification and compensation claims, whereas institutional climate finance is embedded in public planning, development cooperation, monitoring and safeguards. Because recent project-level evidence shows that voluntary carbon offset projects are associated with higher anthropogenic pressure on ecosystems (Zhou and Almond, 2025), we examine whether carbon-credit activity, when aggregated to the country-year level, is associated with similar patterns. This benchmark provides a reference point for estimating the relationship between institutional adaptation and mitigation finance and biodiversity outcomes.

This paper makes two contributions. First, it provides cross-country evidence on the relationship between institutional climate finance and biodiversity, measured by both anthropogenic pressure (HII) and species conservation status (RLI). Second, it shows different biodiversity patterns for institutional climate finance and voluntary carbon-credit activity at the country-year level. This contrast supports the broader argument that climate-related capital is not ecologically uniform; its biodiversity implications depend on the institutional channel through which it is governed and deployed.

The remainder of the paper proceeds as follows. Section 2 reviews the literature and develops a conceptual framework linking climate finance to biodiversity. Section 3 describes the data, country-level carbon-credit comparison and empirical methodology. Section 4 presents the main results and robustness analyses. Section 5 examines heterogeneity across financing channels. Section 6 concludes with policy implications, limitations and future research.

The United Nations Framework Convention on Climate Change (UNFCCC) defines climate finance as funds from public, private and alternative sources supporting mitigation and adaptation at local, national and international levels. In public economics, these flows help correct environmental externalities and support global public goods (Flammer et al., 2025; Giglio et al., 2021). Climate stability is a global public good because its benefits are nonexcludable and nonrival (Flammer et al., 2025).

This structure creates a free-rider problem because countries may benefit from climate stability while contributing little to mitigation (Barrett, 2007). Institutional climate finance addresses this problem by transferring resources from developed to developing countries for mitigation and adaptation costs (Barrett, 2007). It therefore functions as both funding and an institutional mechanism for intergovernmental cooperation, monitoring and transparency.

Allocation is central to this mechanism. Normatively, climate finance requires developed countries to support vulnerable economies based on historical responsibility and the polluter-pays principle (Bosma et al., 2025). Empirically, however, allocation often reflects recipient need, recipient merit and donor interest (Prasad and Singhania, 2026). Grants tend to target vulnerable countries, whereas loans and equity favor countries with stronger institutions and lower implementation risk, potentially excluding fragile states (Bosma et al., 2025; Prasad and Singhania, 2026).

This nonrandom allocation creates an empirical challenge because environmental conditions, institutional quality, donor–recipient alignment and political priorities may affect both climate-finance allocation and biodiversity outcomes (Almaghrabi et al., 2025; Waldron et al., 2017). Observed relationships may therefore reflect both financing effects and allocation selection.

Biodiversity enters this framework through the idea of co-benefits. Climate finance was not designed as a biodiversity-financing system; biodiversity became relevant because climate policies may generate spillovers beyond mitigation and adaptation. In the 1990s, such spillovers were framed as co-benefits when mitigation also improved health, air quality or local development (Karlsson et al., 2020). After the 2001 IPCC Third Assessment Report defined co-benefits as positive effects on objectives beyond the primary policy target, the concept linked climate, development and conservation debates.

Biodiversity co-benefits are plausible because climate change and biodiversity loss are closely connected. These crises operate as a “twin-crises multiplier,” where ecological degradation weakens climate regulation while climate instability accelerates biodiversity loss (Giglio et al., 2025). Mitigation and adaptation finance may therefore affect biodiversity even without an explicit conservation objective. Institutional climate finance can support restoration, forest protection, nature-based adaptation, avoided deforestation, sustainable land management and climate-resilient planning (Bosma et al., 2025; Flammer et al., 2025; Seidl et al., 2024). Its development-oriented framework may also strengthen monitoring, safeguards and implementation, helping finance translate into ecological outcomes (Guo et al., 2026; Wang et al., 2025).

This institutional pathway differs from the market-based carbon-offset model. Offset projects frequently claim biodiversity co-benefits (Zhou and Almond, 2025). Yet, their ecological effects remain contested. They center on tradable mitigation claims and carbon-accounting criteria such as additionality, leakage, permanence, verification and credit quality. These rules support carbon-market integrity but do not capture biodiversity needs such as habitat integrity, species recovery, landscape connectivity and ecological resilience (Seddon et al., 2020; Zeng et al., 2026).

Zhou and Almond (2025) show that voluntary carbon offset projects are associated with higher habitat disturbance after project establishment. Their analysis suggests shifts from shrublands and sparse forests toward pastures and simplified landscapes. They find little evidence that biodiversity claims, certifications, protected-area overlap, registry rules or third-party ratings identify better habitat outcomes (Zhou and Almond, 2025).

Conservation-targeted spending is associated with lower biodiversity loss (Waldron et al., 2017), whereas broader climate finance may yield weaker benefits without explicit conservation priorities (Bosma et al., 2025). International environmental aid has been linked to lower extinction risk after substantial lags (Rogers, 2025). Some mitigation investments may create biodiversity pressures through land-use changes linked to renewable energy infrastructure and transport networks (Gonon et al., 2025). These findings suggest that climate-related capital is not ecologically uniform.

This study examines whether institutional climate finance is associated with lower human pressure and better species conservation at the country level. Voluntary carbon-credit activity serves as its market-based counterpart. Comparing the two at the country-year level allows us to assess whether biodiversity outcomes differ across institutional forms of climate-related finance rather than assuming uniform ecological effects.

We use the HII to measure human pressure on ecosystems (Zhou and Almond, 2025). The index uses global satellite data at 300-m resolution covering 2001–2020 ^[1] (Venter et al., 2016). It ranges from 0, indicating no human influence, to 64, indicating maximum pressure (Sanderson et al., 2022). The HII is suitable for this study because many climate-finance activities operate through land use, infrastructure, energy systems, forest management and ecosystem restoration (Poyser, 2026; Stoll et al., 2021). However, because the HII does not directly measure biological outcomes (Zhou and Almond, 2025), we complement it with the RLI.

The RLI captures the species-level dimension of biodiversity. It measures changes in aggregate extinction risk based on the IUCN Red List of Threatened Species and covers 243 countries over 1993–2024. Species are classified using criteria such as population size, rate of decline and geographic range (IUCN, 2001). Unlike simple threatened-species counts, the RLI tracks genuine movements across extinction-risk categories, excluding changes caused by taxonomy or improved knowledge (Butchart et al., 2006). Together, the HII and RLI examine whether climate finance is associated with lower ecosystem pressure and improved species conservation.

The main explanatory variables are the annual mitigation and adaptation finance received by each country (Lee et al., 2022). The data come from the OECD Development Assistance Committee (DAC) ^[2], which records official development assistance (ODA) and related financial flows from bilateral and multilateral donors. The data set provides annual climate-finance data for 161 countries during 2002–2024.

We separate mitigation and adaptation finance because they operate through different policy channels. Mitigation finance targets greenhouse-gas reduction, while adaptation finance supports responses to local climate risks (Michaelowa, 2001). This separation allows us to examine whether biodiversity associations differ across policy channels related to energy transition, land use, ecosystem management and resilience.

To isolate the relationship between climate finance and biodiversity, we constructed a country-year panel with controls for economic structure, energy systems, vulnerability and baseline conservation capacity. From the World Bank's World Development Indicators (WB-WDIs ^[3]), we include GDP per capita, merchandise trade (% of GDP), urban population (% of total), industry value added (% of GDP), inflation and agricultural land (% of land area). These variables capture development, structural transformation and population pressures affecting ecosystems (Poyser, 2026).

We also control for the financial development index (Svirydzenka, 2016), electricity generation from fossil fuels and from renewables, excluding hydropower (% of total), greenhouse gas emissions per capita and renewable energy consumption. These controls distinguish climate-finance effects from broader energy-transition and financial-capacity differences. Climate vulnerability is measured using the ND-GAIN index ^[4] because vulnerability can influence both the allocation of climate finance and environmental outcomes (Weiler et al., 2018). Finally, cumulative protected areas (km²) from the World Database on Protected Areas (WDPA ^[5]) are included to account for baseline conservation capacity, following Poyser (2026).

Continuous variables are transformed using ln(1+x) to reduce right skewness and retain zero observations. Because the climate-finance variables contain a few exact zeros and many very small positive values, coefficients are not interpreted as exact elasticities. We instead assess economic magnitude using standard-deviation-scaled estimates. Table A1 in the Online Appendix (OA) reports variable definitions, sources and summary statistics. Figure A1 (OA) reports distribution and concentration diagnostics for the biodiversity outcomes, and Figure A2 (OA) reports the corresponding diagnostics for the main variables of interest.

All data sets are merged by country and year, with regional aggregates excluded. Missing values in slow-moving biodiversity indicators are treated using last observation carried forward because the HII and RLI change gradually over time (Zhou and Almond, 2025). Figure A9 in the Online Appendix reports annual missingness, while Figure A12 re-estimates the main specifications without last observation carried forward. Outliers are winsorized at the 5th and 95th percentiles, with alternative thresholds reported as sensitivity checks. The final panel includes 132 countries and 2,413 country-year observations over 2002–2020. The full list of sample countries is reported in Table A13 in the Online Appendix.

All baseline models relate climate-related finance in year $t$ to biodiversity outcomes observed in $t+1$⁠. We use this one-year lead as a consistent timing convention across both specifications. The choice is not derived from a biological lag structure and does not identify when ecological adjustment begins or when it is completed.

A country-level comparison requires a market-based counterpart at the same aggregation level as institutional climate finance. We therefore construct a benchmark using country-year net credit issuances from the Voluntary Registry Offsets Database compiled by the Berkeley Carbon Trading Project ^[6]. Voluntary carbon credits are suitable for this comparison because they represent a prominent market-based channel through which climate-related capital is linked to land use, forestry, renewable energy and nature-based projects. Their association with greater habitat disturbance makes carbon-credit activity a relevant contrast with institutional climate finance (Zhou and Almond, 2025). The empirical model is as follows:

where $Biodiversityi,j,t+1$ measured one year ahead is proxied by both the HII and RLI and the $Total Creditsi,j,t$ are the country-year total credit issuances minus credit retirements. $Xi,j,t$ denotes a vector of time-varying control variables. $δj$ and $ηt$ denote region and year fixed effects. A positive association with $HIIi,t+1$ and a negative association with $RLIi,t+1$ would provide a country-level market-based benchmark against which the institutional climate-finance estimates can be compared.

We use region and year fixed effects as the baseline because the analysis aims to retain meaningful cross-country variation rather than relying only on short-run within-country changes. This is important because the HII and RLI are slow-moving indicators. In the regression sample, the within standard deviation is 0.080 for HII, compared with an overall standard deviation of 0.500 and 0.005 for RLI, compared with an overall standard deviation of 0.047 (Table A2 in the Online Appendix). With limited within-country variation, country fixed effects may absorb much of the identifying signal. Region fixed effects account for broad regional heterogeneity while preserving the cross-country variation central to this setting (Heckelman and Wilson, 2025; Auffhammer and Schlenker, 2014; Kentor et al., 2012).

After establishing the country-level market-based benchmark, we turn to the main analysis of institutional climate finance and biodiversity outcomes:

where $Climate Financei,j,t$ is measured separately by mitigation and adaptation climate finance. As in the carbon-credit benchmark, the one-year lead establishes temporal ordering without assuming that biodiversity adjusts within one year.

Robustness checks assess sensitivity to outcome measurement, omitted confounders and ecological exposure definitions. First, following Zhou and Almond (2025), we replace the baseline HII measure with alternative distributional statistics, including minimum, maximum, standard deviation and sum/area HII (aggregate intensity). Second, we add controls for environment-related technologies (Ahmad and Zheng, 2021), other ODA (Poyser, 2026), governance indices (Hussain and Dogan, 2021) and political aid cycles (Faye and Niehaus, 2012; Reid, 2006), which helps separate climate finance from broader development assistance and political allocation patterns. Third, we examine whether the institutional contrast is also visible in a more concrete land-use setting. We compare climate finance targeted to forestry with voluntary carbon credits in forestry and land-use activities, using forest area and forest rents as outcomes. Because Poyser (2026) shows that forests are a direct land-use channel through which climate finance may affect biodiversity, these tests strengthen the construct validity of the ecological interpretation.

Climate finance allocation is unlikely to be random. Donors may favor countries with greater climate vulnerability, policy capacity, mitigation potential or environmental needs. These same factors may also shape biodiversity outcomes through land use, conservation capacity and ecosystem pressure. Estimates from equation (2) may therefore reflect reverse causality, omitted-variable bias or simultaneity. To address this concern, we re-estimate equation (2) using an instrumental variables–two-stage least squares (IV-2SLS) approach.

Our main external instrument is a predetermined top-donor budget-competition instrument. The intuition is that each recipient country historically depends more on some climate-finance donors than on others. If a recipient's historically important donor reallocates a larger share of its finite climate-finance portfolio to other countries in a given year, less funding should be available for the focal recipient. This provides donor-side variation in climate finance that is less directly tied to contemporaneous recipient-country biodiversity.

For each recipient country i, we identify its top donor during the 2002–2006 preperiod. The top donor is the donor that provided the largest cumulative total climate finance to recipient i during this period. Total climate finance is defined as the sum of adaptation and mitigation finance:

where d denotes donor, i denotes recipient and t denotes year. The preperiod top donor for recipient i is defined as

Donor identity is fixed over time to avoid using current allocation patterns to define the instrument. We then construct the instrument from the predetermined donor's current allocation of total climate finance to other recipient countries. Let $diT$ denote recipient i's top donor in the 2002–2006 preperiod. The donor's total climate-finance outflow in year t is as follows:

The amount allocated by this donor to countries other than the recipient is as follows:

The donor-budget-competition instrument is therefore

This variable measures the share of recipient i's predetermined top donor's climate-finance portfolio allocated to other countries in year t. A higher value indicates stronger allocation pressure elsewhere in the donor's portfolio and should reduce the climate finance available to the focal recipient.

The 2002–2006 donor-ranking window fixes donor identity before the climate-finance system became more formally organized around biodiversity-relevant policy channels. COP13 and the Bali Action Plan in 2007 shifted global climate finance toward a more structured framework, placing mitigation, adaptation, technology transfer, financial resources, REDD+, NAMAs, MRV requirements and the Adaptation Fund closer to the center of international climate cooperation (Clémençon, 2008; Leggett, 2010). Post-2007 donor portfolios may therefore reflect climate-policy priorities that are also correlated with biodiversity outcomes, including adaptation vulnerability, mitigation potential, REDD-related opportunities, forest-carbon projects and land-use change. Using the 2002–2006 window reduces this concern by defining each recipient's historically important donor before these channels became more embedded in the climate-finance architecture. This does not make the preperiod perfectly exogenous, but it provides a more defensible historical baseline than a later period shaped by the post-Bali framework (Hall and Persson, 2018; Lyster, 2013).

The relevance condition follows from donor-side budget competition. When recipient i's pre-Bali top donor allocates a larger share of its climate-finance portfolio to other countries, less funding should be available for country i. The exclusion restriction requires the instrument to affect biodiversity outcomes only through climate finance received by the focal recipient. The design supports this condition by combining predetermined donor identity with a leave-one-recipient-out allocation measure. The identifying assumption is that, conditional on controls, fixed effects and common year shocks, the predetermined donor's allocation share to other recipients does not directly affect biodiversity outcomes in country i.

The role of COP13 is not to generate instrument variation but to justify the fixing of donor identity before climate-finance mechanisms became more explicitly tied to adaptation, mitigation, REDD+, NAMAs, MRV and the Adaptation Fund. We nevertheless treat the exclusion restriction as a conditional identifying assumption rather than a mechanical guarantee since donor allocations to other countries may still reflect broader geopolitical priorities, humanitarian crises, natural disasters or development shocks that could influence biodiversity through channels other than climate finance. We mitigate this concern by controlling for observed country characteristics, fixed effects and common year shocks and by relying on predetermined donor identity rather than current-year donor rankings.

As an additional robustness check, we applied the heteroskedasticity-based identification approach proposed by Lewbel (2012). This method constructs internal instruments from the model's own data by exploiting heteroskedasticity in the first-stage error term. It provides a complementary identification strategy and helps assess whether the estimated biodiversity effects of climate finance depend on the specific external donor-budget-competition instrument.

Table A1 in the Online Appendix reports descriptive statistics for 2,413 country-year observations over 2002–2020. All variables are transformed using ln(1+x) and winsorized at the 5th and 95th percentiles. The biodiversity indicators show moderate dispersion, with mean values of 2.379 for the HII and 0.617 for the RLI. HII and RLI capture ecosystem pressure and species conservation status, respectively.

The climate-finance variables are concentrated near zero, especially after winsorization. CF adaptation has a skewness of 1.409 and kurtosis of 3.935, while CF mitigation has a skewness of 2.221 and kurtosis of 7.126, indicating that mitigation finance is more right-skewed than adaptation finance. In the raw 2002–2020 country-year data, exact zero values are rare: 2 observations have zero adaptation finance, 10 observations have zero mitigation finance and no observation has both variables equal to zero. The high density near zero therefore reflects many very small positive finance values rather than many exact zeros. This supports the use of ln(1+x), but it also means that coefficients on climate finance should not be interpreted as exact elasticities. For this reason, the paper relies on standard-deviation-scaled estimates when discussing economic magnitude.

Table A3 in the Online Appendix reports the pairwise correlation matrix. Figure A6 (OA) reports the 10 largest absolute pairwise correlations in the regression sample. The highest correlation involving the main variables of interest is between adaptation and mitigation climate finance (⁠$r=0.724$⁠), indicating that countries receiving one type of climate finance often also receive the other. This pattern supports estimating the two flows separately to avoid conflating their biodiversity associations. For the remaining covariates, the pairwise correlations with adaptation and mitigation finance are below 0.6, suggesting that severe bivariate multicollinearity is unlikely to drive the main estimates.

Before estimating the main relationship between institutional climate finance and biodiversity, we use voluntary carbon-credit activity as a country-level market-based benchmark. This exercise, reported in Table A6 of the Online Appendix, does not validate the project-level mechanism documented by Zhou and Almond (2025) and Zeng et al. (2026). It provides a country-year benchmark for assessing whether carbon-credit activity displays a biodiversity association different from institutional climate finance.

The estimates in Table A6 indicate that higher carbon-credit activity is associated with higher one-year-ahead human pressure (HII) and weaker one-year-ahead species conservation status (RLI). This pattern is consistent with prior evidence showing that voluntary carbon-offset projects can be linked to higher habitat disturbance (Zhou and Almond, 2025). These country-level associations broadly align with project-level evidence. This aligns with concerns that carbon-focused mechanisms may create ecological trade-offs when safeguards are weak (Phelps et al., 2012). This comparison shows how a market-based climate-finance channel behaves in the same country-year setting used for institutional climate finance.

Having established a country-level benchmark for the market-based carbon-credit channel, we turn to the main institutional climate-finance results. Table 1 reports the baseline one-year-ahead associations for institutional climate finance. Adaptation and mitigation finance are linked to lower one-year-ahead human pressure and improved species conservation status. Importantly, the lead provides temporal ordering within the annual panel but does not imply that ecological adjustment occurs within one year.

Because the climate-finance variables are transformed using ln(1+x) and contain many very small positive values, we do not rely only on 1% elasticity interpretations. Instead, we report standard-deviation-scaled effects to assess economic magnitude (Figure A7 in the Online Appendix). A one-standard-deviation increase in mitigation finance is associated with a 0.144 standard-deviation decrease in HII and a 0.034 standard-deviation increase in RLI. Similarly, a one-standard-deviation increase in adaptation finance is associated with a 0.052 standard-deviation decrease in HII and a 0.074 standard-deviation increase in RLI.

These results are consistent with the view that institutional climate finance can be associated with biodiversity co-benefits when climate-related spending is embedded in public planning, safeguards and development-oriented implementation (Barrett, 2007; Giglio et al., 2021). By funding sustainable land-use planning, ecosystem restoration and nature-based adaptation, institutional climate finance can reshape the governance and safeguards that protect ecological systems (Bosma et al., 2025; Flammer et al., 2025).

The contrast between Table 1 and Table A6 shows that different climate-finance structures are associated with different biodiversity patterns. While institutional climate finance is linked to improved biodiversity status, carbon-credit activity is associated with higher human pressure and weaker species outcomes. This divergence suggests that the biodiversity effects of climate action are not mechanical; they depend on whether the financing channel prioritizes public safeguards or market incentives (Barrett, 2007; Giglio et al., 2021).

Because biodiversity responses may unfold over several years, we complement the one-year-ahead estimates with long-horizon timing diagnostics. Figure A8 in the Online Appendix reports distributed-lag estimates over horizons h = 1, 3 and 5, comparing cumulative-window effects with individual lead-lag coefficients from the same specification. The cumulative-window estimates broadly align with the baseline signs ^[7]. The individual lead-lag coefficients are imprecise and are reported only to show how the cumulative estimates are distributed across the timing window. Given the slow movement of HII and RLI, we treat Figure A8 as suggestive evidence that the baseline directional pattern is not confined to the $t+1$ specification.

Table A7 in the Online Appendix examines whether the baseline results depend on the construction of the HII, following Zhou and Almond (2025). The estimates remain consistent across alternative HII measures. Adaptation finance is negatively and significantly associated with the maximum HII, aggregate intensity and dispersion, suggesting lower pressure in highly disturbed areas and lower overall ecosystem stress. Mitigation finance shows a similar sign pattern, although with smaller magnitudes. These results indicate that the baseline HII findings are not specific to the mean HII measure.

Table A8 in the Online Appendix adds controls for environment-related technologies, other Net ODA excluding climate and biodiversity aid, biodiversity-related funding, governance indices and political-cycle variables. The climate-finance coefficients remain broadly stable across these specifications. These results do not eliminate omitted-variable concerns but show that the main associations are not absorbed by green technological capacity, broader development assistance, dedicated conservation aid, governance quality and electoral timing.

Table A9 (OA) focuses on forest outcomes to provide sectoral evidence related to the land-use interpretation. Forest ecosystems respond more directly to financial incentives and land-use policy, making them a useful sectoral check on the land-use interpretation (Poyser, 2026). Adaptation and mitigation finance are positively associated with one-year-ahead forest area and negatively associated with one-year-ahead forest rents. This pattern is consistent with expanded forest cover and lower measured dependence on forest extraction.

In contrast, forestry-related carbon credits show the opposite pattern: they are negatively associated with future forest area and positively associated with future forest rents. These results reinforce the distinction between institutional climate finance and market-based carbon-credit activity. The forest results therefore provide a concrete land-use counterpart to the broader HII and RLI findings.

We further examined whether the baseline findings were sensitive to data treatment and sample composition. Coefficient plots show that the signs for adaptation and mitigation finance remain stable across alternative winsorization thresholds, including no cutoff, 1/99, 2.5/97.5, 5/95 and 10/90 rules (Figure A5 in OA). The HII results are especially stable, while the mitigation-to-RLI estimates remain positive but less precise under the least restrictive cutoffs. Influence tests (Figure A3 in OA) and leave-one-country-out jackknife checks (Figure A4 in OA) further show that the results are not driven by extreme observations, dominant recipient countries or unusual finance years.

To address the nonrandom allocation of climate finance, we re-estimate the main models using IV–2SLS. Donors may allocate funds according to recipient vulnerability, implementation capacity or donor–recipient alignment, so baseline fixed-effect estimates may reflect both finance associations and allocation patterns. We use a predetermined top-donor budget-competition instrument: each recipient's top donor is fixed using 2002–2006 flows, before the 2007 Bali Action Plan (Clémençon, 2008) and the instrument measures the donor's annual climate-finance share allocated to other recipients. This preperiod design reduces the risk that donor identity is shaped by the post-Bali climate-finance architecture (Hall and Persson, 2018; Lyster, 2013). A larger allocation share with other recipients captures donor-side budget pressure and predicts lower finance to the focal recipient.

Table 2 supports instrument relevance. The coefficient on the top-donor other-recipient allocation share is negative and statistically significant at the 1% level in all specifications, indicating that stronger budget pressure elsewhere predicts lower adaptation and mitigation finance received by the focal country. Weak-instrument concerns are limited: Kleibergen–Paap F-statistics range from 29.808 to 52.466. Endogeneity tests are more mixed. Exogeneity is not rejected for HII, with p-values of 0.153 for adaptation and 0.518 for mitigation, but is rejected for RLI, with p-values of 0.002 and 0.000. Therefore, endogenous allocation appears more consequential for species-level outcomes.

Table 3 reports the second-stage results. Adaptation and mitigation finance are both associated with lower future HII and higher future RLI. Adaptation finance is associated with a reduction in HII (−14.517) and an increase in RLI (2.695), while mitigation finance has corresponding coefficients of −11.214 and 2.127. All estimates are statistically significant at the 1% level. In standardized terms, a one-standard-deviation increase in adaptation finance is associated with a 0.116-standard-deviation decline in future HII and a 0.229-standard-deviation increase in future RLI. For mitigation finance, the corresponding changes are 0.135 and 0.272 standard deviations. As in the baseline models, the one-year lead provides temporal ordering and should not be read as identifying the biological adjustment period. The IV estimates indicate nontrivial cross-country associations and are broadly consistent with the baseline results, suggesting that the documented relationship is unlikely to be driven solely by endogenous allocation. Nevertheless, because the exclusion restriction cannot be verified empirically, and given the potential for residual geopolitical or macroeconomic confounding, these estimates should be interpreted as supportive evidence rather than definitive causal identification.

The IV estimates are generally larger than the baseline estimates, especially for RLI and for the adaptation-finance coefficient on HII. This pattern is consistent with attenuation in the baseline models from endogenous allocation, measurement error or reverse causality. Climate finance may be directed toward countries with greater environmental stress, stronger implementation capacity or closer donor-recipient alignment, while reporting and classification inconsistencies may also attenuate conventional estimates (Prasad and Singhania, 2026; Weiler et al., 2018; Almaghrabi et al., 2025; Zhou and Almond, 2025). The larger RLI estimates are also consistent with the endogeneity tests, which reject exogeneity for the RLI but not for the HII.

We conducted four additional checks. First, because the baseline IV model is exactly identified, we add the one-year lag of the donor-budget-competition instrument as a diagnostic extension. This creates an overidentified specification and tests whether current and lagged donor-allocation measures produce consistent IV patterns. The approach is related to studies that use lagged versions of the same underlying instrument or instrumented process to expand the instrument set and report Hansen tests (Liu and Hua, 2024; Wang and Zhou, 2025). The first- and second-stage results remain stable, and Hansen tests do not reject the overidentifying restrictions. Since both instruments come from the same donor-budget mechanism, this test is interpreted as a consistency check rather than a proof of exclusion. Figure A10 (OA) reports the IV-2SLS coefficient estimates using the current instrument only, the lagged instrument only and both instruments jointly, and Figure A11 (OA) reports the corresponding diagnostic tests.

Second, we add donor-recipient exposure controls for persistent cross-country differences that may shape both climate-finance allocation and biodiversity outcomes. These controls are weighted by adaptation- or mitigation-related donor exposure and capture predetermined geographic, historical, institutional, cultural and socio-economic links between donors and recipients, as reported in Tables A4 and A5 in the Online Appendix ^[8]. The results remain stable, reducing the concern that the region fixed-effects specification is driven by omitted structural heterogeneity. While this does not make region fixed effects equivalent to country fixed effects, it provides a robustness check that directly targets persistent structural differences across countries.

Third, we exclude the 2002–2006 donor-ranking window from the estimation sample and re-estimate the IV models for the post-2007 period. This addresses the concern that donor exposure is defined using years that also appear in the baseline panel. Tables A10 and A11 in the Online Appendix show results consistent with the main IV estimates. The Kleibergen–Paap F-statistics remain above conventional weak-instrument thresholds, indicating that the instrument continues to predict climate-finance receipts after the donor-ranking window is removed from the estimation. This strengthens the timing rationale for the pre-Bali donor definition.

Fourth, Table A12 (OA) applies Lewbel's (2012) heteroskedasticity-based identification strategy, which constructs internal instruments from heteroskedasticity in the first-stage error term. The estimates remain directionally consistent with the IV–2SLS results, suggesting that the main pattern is not specific to the external donor-budget-competition instrument.

Figure 1 examines whether biodiversity responses differ across climate finance components rather than relying only on an average effect. Conservation funding often operates through specific mechanisms, while aggregate biodiversity indicators may hide sectoral differences (Poyser, 2026). Decomposing climate finance ^[9] helps show whether the aggregate pattern is concentrated in a single category or appears across several financing components.

The left panel reports component-level coefficients for Mean HII observed at t + 1. Most adaptation and mitigation components show negative estimates, indicating that higher finance in year (⁠$t$⁠) is followed by lower human pressure, except for adaptation finance for disaster, which is statistically insignificant. The right panel reports effects on the RLI in year (⁠$t+1$⁠). Most coefficients are positive, suggesting improved species outcomes, though adaptation finance for disaster, again, remains statistically insignificant. The weak result for disaster-related adaptation is not surprising. Disaster adaptation often finances emergency preparedness, protective infrastructure and postshock resilience. These activities may be essential for adaptation without producing immediate changes in HII or RLI.

Mitigation finance shows the most stable sign pattern across components. Most categories are associated with lower next-year HII and higher next-year RLI. One possible interpretation is that mitigation finance may support cleaner technologies and infrastructure, which could reduce some habitat pressures. Despite possible local trade-offs, the consistent pattern suggests that mitigation finance generally aligns with lower levels of human pressure and supports species outcomes in the following period.

This study examines whether institutional climate finance is associated with biodiversity outcomes in a panel of 132 countries from 2002 to 2020. The results show that adaptation and mitigation finance are associated with lower future human pressure and better species conservation status. This pattern suggests that biodiversity outcomes depend not only on the amount of climate-related capital mobilized but also on the institutional channel through which that capital is governed and deployed.

The findings also reveal a clear distinction between institutional climate finance and market-based carbon-credit activity. At the country-year level, voluntary carbon-credit activity is associated with higher future human pressure and lower future species conservation status, while institutional climate finance shows the opposite pattern. This contrast suggests that ecological outcomes depend not only on the volume of climate-related capital but also on how that capital is governed (Barrett, 2007; Flammer et al., 2025).

These findings have two policy implications. First, biodiversity safeguards should be part of climate-finance design rather than an external environmental add-on. Finance directed toward mitigation or adaptation is more likely to generate ecological co-benefits when project selection, monitoring and evaluation also account for land-use pressure, ecosystem integrity and conservation capacity. Second, carbon markets should not be treated as sufficient biodiversity instruments on their own; their credibility depends on stronger ecological monitoring, regulation and local accountability (Zeng et al., 2026).

This study has several limitations. The HII and RLI capture broad biodiversity conditions but cannot fully identify local ecological mechanisms. The HII summarizes aggregate human pressure without distinguishing all forms of land-use change, while the RLI measures changes in aggregate extinction risk and evolves within a particularly narrow annual range. The baseline uses outcomes observed in $t+1$ to impose a consistent temporal ordering and maintain a common annual-panel specification. This choice is an empirical convention rather than a biological assumption. The estimates therefore do not identify when ecological adjustment begins, how quickly it proceeds or when its full effect becomes observable. The distributed-lag estimates over one-, three- and five-year horizons provide suggestive evidence on temporal consistency, but their precision declines at longer horizons and they do not resolve the adjustment path. The analysis also focuses on first-order linear associations and cannot observe project boundaries, safeguard enforcement, community participation or implementation quality. Future research using longer panels, spatially explicit outcomes and project-level implementation dates could identify delayed, cumulative and heterogeneous ecological responses more precisely.

In preparing this article, we used ChatGPT to check for grammatical and typographical errors. After using this tool, we reviewed and edited the content as needed and take full responsibility for the final publication.

The authors gratefully acknowledge that an earlier version of this paper received Best Paper Awards at two international conferences: the International Scientific Conference on Resilience by Technology and Design (RTD 2026) and the International Conference on Business Sustainability (ICBS 2026), both organized by the University of Economics Ho Chi Minh City (UEH). The authors also thank the participants at both conferences for their constructive comments and valuable feedback, as well as the Editor and anonymous reviewers for their insightful suggestions, which have significantly improved the quality and clarity of the manuscript. This research was fully funded by the University of Economics Ho Chi Minh City (UEH) under Grant No. 2026-03-05–3474. The views expressed in this paper are solely those of the authors and do not necessarily reflect the views or policies of UEH. The usual disclaimer applies.

The supplementary material for this article can be found online.