Integrating sustainability into climate finance by quantifying the co-benefits and market impact of carbon projects

Build up interactive SDG co-benefit framework

From our previous systematic literature review37, we find that a great deal of variation in co-benefits existed not only among project types but also within project type. In this section, we take one step further to assess the co-benefits of these projects in a more quantified way by drawing up studies of scoring exercise at the level of the SDG targets to better understanding the interaction between the CDM project technologies and the SDG dimensions. The two primary studies on this topic are38 and the Intergovernmental Panel on Climate Change (IPCC) special report Global Warming of 1.5 °C39. Both studies have conducted thorough research on the potential SDG targets from the deployment of mitigation options. Our paper adopts the structure of integrating the SDG targets into the mitigation options from both studies, while adding another layer of five co-benefit criteria on top of this structure. Thus, the final structure of the assessment is presented in Supplementary Fig. 1.

Under each SDG target, we assign an SDG-interaction score from this specific SDG target and the project. The SDG-interaction score is a seven-point scale score. Interaction between outcomes of the CDM projects and the SDG targets can be positive and/or negative. For the positive interaction, we have “high impact”, “medium impact”, and “limited impact” scales, and for negative interaction, we have “minor damage”, “medium damage” and “massive damage”. Additionally, we present the validity of the results in the literature by examining the quantity, quality, and consistency of the literature into four scales, limited, medium, robust, and extensive. Eventually, we assign the current level of confidence (“low”, “medium”, “high”) to each SDG interaction based on the previous two aspects. This bottom-up direction of assessing the SDG interaction scores eventually can be aggregated at the level of co-benefit criteria. Our implication assumption is that the SDG goals are weighted equally, despite those countries may have different focus areas on sustainable development based on their national development priorities.


Our data of the interactive SDG co-benefit framework is based on 84 academic peer-reviewed and grey studies conducted on the topic of carbon finance and community co-benefits from a systematic literature review search. The primary data source for economic models is the UNEP DTU CDM/JI Pipeline Analysis and Database (CDM/JI Pipeline). Additional information, such as ERPA dates, is extracted by Python from CDM documents in PDF format on the UNFCCC CDM projects site. To check the accuracy of the ERPA dates extracted by the computer, we adopted two methods to validate the data (see Supplementary Note 2: Accuracy of the ERPA Dates). We include 2259 CDM projects in our paper. The dataset covers 20 project types and two project sizes46. We present the statistical summaries of the data in Supplementary Table 5. We also plot the distribution of CER prices in Supplementary Fig. 9. Within these 2259 CDM projects, 1655 are regular CDM projects, and 64 are Gold Standard CDM projects. Detailed results segregated by project types and sizes are listed in Supplementary Table 6.

The underlying assumption of our analysis is that carbon prices (including any additional premiums from non-carbon sustainable development benefits) will compensate for the opportunity costs in the project, including transaction and implementation costs. For most of the CDM projects, additional revenue from carbon prices will help the CDM projects pass the additionality requirement. Because CDM projects need to demonstrate that projects are not viable unless carbon benefits are considered (i.e. with a non-zero carbon price). This assumption is valid for most CDM projects due to the additionality requirement they must pass, particularly when using financial additionality to meet the additionality requirement. During this additionality requirement, it is not supposed to include monetized co-benefits, which is precisely why projects with co-benefits are valued more highly by investors. However, CDM-AR projects suffer from particularly high opportunity costs, transaction costs, and implementation costs3,47,48,49,50. For example, the World Bank BioCarbon Fund, the major credit buyer holding17 out of the 66 CDM-AR projects51, reported that the transaction costs of CDM-AR projects exceeded $1 per tCO2e, higher than any other CDM project type52. As a result, the World Bank adopted a default price in financial analysis of CDM-AR projects51. Also, in the case of CDM-AR projects these costs are likely to be higher since they involve land and property rights and more complex implementation arrangements53. In addition, the high opportunity costs of land and labor are well discussed in the CDM-AR literature54,55,56,57. These challenges can complicate evaluation of the co-benefits of the CDM-AR projects and lead to the failure of these projects3,54. Therefore, in our analysis, we did not include CDM-AR projects due to these characteristics as well as the small number of registered projects (a total of 66 registered projects). As a result, we believe that the assumption of our analysis is reasonable.

Credit buyers

The CDM mechanism creates CERs as an important share of the global carbon markets. Like the regular markets, the demand side of the CERs is from carbon credit buyers. CDM credit buyers can be categorized into three groups, the first group called compliance buyers who are seeking to buy offsets for compliance in the EU ETS and other regional schemes; the second group called sovereign buyers, mainly Annex I parties, who are obtaining CERs directly to meet their quantified emission limitation and reductions obligations (QELRO) commitments under the Kyoto Protocol; the last group contains MDBs and carbon funds58. We further divide credit buyers into different categories by using two classification systems. First, credit buyers (company level) are classified into 14 industries by their primary business activities using the Bloomberg Industry Classification Systems (BICS). Second, credit buyers are also categorized into five statuses based on their profit status, e.g., local private companies, global private companies, government entities, MDBs, and foundations.

Definition of co-benefits (non-carbon benefits)

Although the idea of co-benefits has attracted increasing attention from governments, NGOs, financial institutions, and academic research in recent years, there is no consensus on a concrete definition or agreed list of what counts as a co-benefit59. The IPCC considers co-benefits as “the positive effects that a policy or measure aimed at one objective might have on other objectives, irrespective of the net effect on overall social welfare”60. In this paper, we focus on a smaller subset of co-benefits, particularly on co-benefits to local communities as a result of CDM mitigation actions (carbon projects) that are targeted at addressing global climate change. Thus, we adopted and adjusted the co-benefits description from the World Bank the Community Development Carbon Fund (CDCF) 2013 report on key community outcomes, where five broad areas are listed. These five areas capture the complex dimensions of co-benefits. The co-benefits of this paper cover the following five areas: Enhanced local infrastructure (e.g., roads, health clinics, schools, water, parks, community centers, etc.); access to cleaner and affordable energy for heating and/or cooking; improved income and employment; improved access to electricity and/or energy efficient lighting; and improved natural resource and environmental services (e.g., reduced pollution, natural resource conservation, forest protection, biodiversity).

Definitions of other terms

Climate finance is defined as the public and private financial flows used to support mitigation and adaptation action to address climate change61,62 There are currently two types of carbon markets for carbon offsets: compliance and voluntary markets. The market settings are different for the two markets. In the compliance (mandatory) market, buyers are primarily motivated to purchase offsets that can provide a more economic sense to reduce emissions to fulfill their lawful requirements, such as in a cap-and-trade regime63. The voluntary carbon market grew later compared to the compliance carbon market. It picked up in the late 2000s and kept a relatively stable trend until 2017. While in the voluntary markets, buyers (for example, companies) are primarily motivated by their social responsibility and concerns about climate change to reduce their emissions64,65. Multi-national, private, for-profit companies make the bulk of voluntary offset purchases by volume. Official Development Assistance (ODA) is defined as the aid from government entities to developing countries with a target to promote economic development and welfare66. Carbon benefits of CDM projects are defined as the anthropogenic emissions of greenhouse gases by sources being reduced below those that would have occurred in the absence of the registered CDM project activity46.

Empirical strategies

Main model

Our main model is expressed in the following regression equation:

$${Y}_{{it}}=\,{\beta }_{0}+\,{\beta }_{1-8}\left({{{\rm {Co}}}-{{\rm {benefit}}}}_{1-8}\right)+{\beta }_{9}{X}_{{it}}+\,{\gamma }_{i}+{\delta }_{i}+\,{\varphi }_{i}+{\omega }_{t}+{\theta }_{i}+{\varepsilon }_{{it}}$$


where i indicates projects, and t indicates years when the credit purchase agreement was signed. In all models, the dependent variable Yit is the CER price for each project. The variables of interest are Co-benefit1–8, with their coefficients β1–8 indicate the effect of different levels of co-benefits on the CDM projects. We also control for a group of other variables listed in Supplementary Table 7, e.g., project location fixed effects (γi), credit buyer fixed effects (δi), project type fixed effects (φi), year fixed effects (ωt), and project size dummy (θi). Finally, the error term captures unobserved factors affecting our dependent variable that changes over the year.

Hedonic model

The hedonic model is expressed in the following regression equation at credit buyers’ company level, where CER prices can be explained as a function of credit buyers and project characteristics67,68.

$${P}_{{{\rm {CERi}}}}= \ f({{{{\rm {numprojects}}}}_{i},{{\rm {location}}}}_{i},\,{{{\rm {industry}}}}_{i},\,{{{\rm {status}}}}_{i},\,{{{\rm {GS}}}}_{i}+{{{\rm {projectsize}}}}_{i}\\ +\ {{{\rm {portfolio}}}({{\rm {asicapacific}}},{{\rm {latinamerican}}},{{\rm {middleeast}}},{{\rm {africa}}},{{\rm {centraluropean}}})}_{i})+{\varepsilon }_{i}$$


where i indicates companies. In all models, the dependent variable PCERi is the average CER price paid by company i. We also control for a group of variables such as, numprojectsi is the number of offset projects under management; locationi is a categorical variable indicating the country where the credit buyer i is located, GSi is the proportion of projects that have Gold Standard certification. We also control for investment portfolio in terms of project regions. Thus, asicapacifici is the proportion of projects that company i invests in Asia and Pacific region, africai is the proportion of projects that company i invests in Africa, the same to the latinamericai, centraleuropeani, middleeasti. Finally, εi is an error term assumed to be normally distributed.


In our study, treatment is if a project receives a Gold Standard certification. The control group includes all the regular CDM projects. The rationale behind matching is to identify (based on the available covariates) a control group of projects with similar characteristics to a treated group of projects for comparison. Thus, the selection of covariates should be those variables that are thought to be related to the outcome (CER prices), but not the treatment69. Our strategy is to perform a propensity score matching at the level of five continuous variables. Beyond that, we also conduct the exact matching using two scenarios. Scenario 1 performs exact matching at the buyers’ country level, and Scenario 2 conducts exact matching at both buyers’ country and project location level. After finding good matches for the treatment group, the model will be adjusted by running a regression to control for the fixed effect from contract year, project type, project location, and buyers’ location.

Python and Stata are used jointly for data analysis. Supplementary Fig. 10 shows that there is overlap in the range of propensity scores across the treatment and comparison group, which we called the “common support”69,70. Assessing the common support condition ensures that any combination of characteristics observed in the treatment group can also be observed among the control group. Additionally, diagnostic tests for balancing of covariates are shown in Supplementary Fig. 11. We can see that matching did a quite good job at balancing the covariates across the treatment and control group, with all (except one) p-values from both the KS-test and the grouped permutation of the Chi-Square distance after matching to be >0.05.

Our model is expressed in the following regression equation:

$${Y}_{{it}}=\,{\beta }_{0}+\,{\beta }_{1}\left({{{\rm {Treat}}}}_{{it}}\right)+{\beta }_{2}{X}_{{it}}+\,{\gamma }_{i}+{\delta }_{i}+\,{\varphi }_{i}+{\omega }_{t}+{\varepsilon }_{{it}}$$


where i indicates projects, and t indicates years. In all models, the dependent variable Yit is the CER price for each project. The variable of interest is Treatit, with its coefficient β1 indicates the effect of Gold Standard on CDM projects. We also control for a group of continuous covariates listed in Supplementary Table 7, project location fixed effects (γi), credit buyer fixed effects (δi), project type fixed effects (φi), and year fixed effects (ωt). Finally, the error term captures unobserved factors affecting our dependent variable that changes over the year.

Matching techniques

Model 1 in Table 2 conducted OLS regression using the nine covariates that used to estimate the propensity to receive the treatment. That is, model 1 displays the difference in being Gold Standard CDM projects and regular CDM projects by controlling for the nine covariates. Model 2 through model 5 show results of estimates by using different matching techniques. Models 2 and 3 only used propensity score matching, while models 4 and 5 used the combined exact matching and propensity score matching technique. The difference between models 2 and 3 is the number of covariates used to obtain the results. In model 2, we perform the propensity score technique for all nine covariates, including both continuous and categorical covariates. In model 3, we only conduct the propensity score with the five continuous covariates. The models of interest are models 4 and model 5. In model 4, we perform the exact matching at the credit buyers’ country level, in order to obtain the impact of Gold Standard on projects within the same country of buyers. In model 5, we restricted our model further to conduct exact matching on both credit buyers’ country level and also the projects’ location level. Model 5 is the most restricted model among these five models. We lost some observations due to model restriction in model 5, and we only obtained 21 projects in the treatment group.

Balancing test

We adopted the standardized differences (SD) technique, which is the standardized difference of means, to assess the differences between multiple variables of the treatment and control groups71 in Supplementary Table 8. If there is no big difference between these two groups, we can conclude that there is adequate balance between these two groups of observations. Before matching Supplementary Table 8(a), the treated and untreated groups are unbalanced. When we do propensity score matching at both categorical and continuous covariates level Supplementary Table 8(b), we still did not get balanced groups. However, in the last test Supplementary Table 8(c), when we only conduct propensity score matching at the continuous covariates level, we get balanced groups.

Robustness checks for regular CDM projects

Many factors can influence the CER prices as indicated in Supplementary Table 7. One of the many factors is the 2008 financial crisis, which is the main cause of the price drop of CERs in that year. The price decreased by about 50%72. Thus, we dropped the 461 projects with a signed ERPA date of 2008, because we think that the year 2008 would have an impact on the CER prices. We re-ran the analysis with the remaining 1744 projects. We get very similar results (results are presented in Supplementary Table 9) across all four models compared to the results in Fig. 3 and all coefficient estimates of variables of interest deliver a similar increasing trend.

Robustness checks for regular Gold-Standard CDM projects

We conducted two robustness checks for our matching analysis. First, we replaced the credit buyer’s country information with the indicators representing the health of a country’s economy, such as GDP per capita, employment rate, government expenditure, and inflation rate. We get very similar results (results are presented in Supplementary Table 10) across all five models compared to the results in Table 1. All coefficient estimates of Gold Standard treatment are statistically significant. This indicates that our models are quite robust. Second, we conducted a “placebo” test by randomly selecting 50% of the data from our control group and artificially assigning them into the treatment group. By doing that, we created a “fake” treatment group, that is, a group that we know was not affected by the Gold Standard. we estimated the models by using the “fake” treatment, and the results are presented in Supplementary Table 11. All the coefficients of treatment effect are not statistically significant. Since we do not find that there is a difference in the absence of the real treatment, Gold Standard certificates, we successfully reject this falsification. This result increases the credibility of our research design.

Source link

Check Also

Paterson NJ candidates violated finance report mandates, says ELEC

PATERSON — Two losing candidates in Paterson’s 2020 ward council elections — Mohammed Akhtaruzzaman and Chauncy Brown — …

Leave a Reply




Get our latest downloads and information first.
Complete the form below to subscribe to our weekly newsletter.