Is the clean development mechanism effective for emission reductions ?

This research studies whether the Clean Development Mechanism (CDM) of the Kyoto Protocol achieves its objective of emission reductions in the host countries. It empirically investigates the impacts of CDM projects on CO2 emission reductions for 60 CDM host countries over 2005-10. This research makes use of the newly-developed econometric methods for dynamic panel data models associated with X-differencing procedure. It provides evidence in support of a decline in CO2 emissions in the CDM host countries. It has important policy implications that encourage the international community to support developing countries’ efforts towards low-carbon development via CDM projects.


The World Institute for Development Economics Research (WIDER) was established by the United Nations University (UNU) as its first research and training centre and started work in
. The Institute undertakes applied research and policy analysis on structural changes affecting the developing and transitional economies, provides a forum for the advocacy of policies leading to robust, equitable and environmentally sustainable growth, and promotes capacity strengthening and training in the field of economic and social policy making. Work is carried out by staff researchers and visiting scholars in Helsinki and through networks of collaborating scholars and institutions around the world. www.wider.unu.edu publications@wider.unu.edu

Introduction
The UNFCCC COP 17 Durban conference marks an encouraging start towards reaching an all-party-inclusive global climate agreement as successor of the Kyoto Protocol by 2015. 1 Although this Durban 'road map' is promising, given the wide divide between developed and developing nations, the international negotiation process for reaching such a deal could be the one filled with great challenges and huge difficulties. Against this background, it is, therefore, of both urgency and necessity to evaluate the effectiveness of the Kyoto market-based mechanisms such as the Clean Development Mechanism (CDM) in terms of mitigating human-induced climate change, so as to know better about how the world's collective actions against climate change should proceed after Kyoto.
Apart from setting legally binding greenhouse gas (GHG) emission reduction targets for Annex I countries, 2 one central feature and great innovation of Kyoto Protocol is that it has started the process of carbon commodification, which turns carbon dioxide emission reductions into commodities that can be market traded. Three key market-based flexibility mechanisms were set up under Kyoto Protocol: Emissions Trading (ET), 3 Joint Implementation (JI) 4 and CDM. Among others, the major purposes of the abovementioned flexibility mechanisms are to help Annex I countries to achieve their carbon emission reduction targets in a cost-effective way, to motivate developing nations onto a GHG emission mitigation path, and to attract private capital into the global emission abatement activities.
The CDM encourages Annex I countries to invest in low-cost emission-reduction projects in developing countries in return for tradable certified emission reductions credits (CERs) to help meet their national emission mitigation targets for the Kyoto compliance. 5 According to Article 12 of the Kyoto Protocol, CDM was set up with two 1 The Kyoto Protocol is an international treaty with legally binding effects on 37 industrialized countries plus the European Community to mitigate greenhouse gas emissions. For more information, please refer to the official website of UNFCCC at http://unfccc.int/kyoto_protocol/items/2830.php. 2 Annex I countries are mainly industrialized countries as listed in the Kyoto Protocol's Annex I. At present, there are in total 41 Annex I nations including European Community. Developing countries are referred to as non-Annex I countries, with no compulsory obligations of carbon emission reductions under the principal of 'common but differentiated responsibilities'. 3 Emission Trading allows countries with unused 'assigned amount units' (AAUs), which are emission allowances allocated to them under the Kyoto Protocol, to sell their excess capacity to other Annex I countries in the carbon market just as any other market commodity. 4 Joint Implementation enables Annex I countries to earn Emission Reduction Units (ERUs) by investing in emission abatement projects in another Annex I party, which can be used to count against their carbon emission targets or to be traded on the carbon market. This mechanism can on the one hand help Annex I parties to meet their Kyoto obligations in a cost-efficient way and on the other hand, bring benefits to the project hosting nations thanks to the direct foreign investment and potential technological diffusion. 5 By April 2012, CDM has grown into an enormous global market with more 7,000 validation projects and an expected value of 17 billion CERs (to 2040) (data provided by the UNFCCC).
intentions or dual objectives. 6 One goal is to promote the global GHG emission reductions at a low cost. The other goal is to facilitate the hosting countries' sustainable development thanks to the direct financial investment in emission reduction projects by and the potential low-carbon technological transfer from Annex I countries.
It is widely recognized that most of the low-cost emission mitigation opportunities lie in the developing countries (e.g. Olmstead and Stavins 2006). As the only mechanism formally involving developing countries in global emission reduction efforts, CDM is of special focus for both the academia and the practitioners (Paulsson 2009). However, in the literature, so far opinions have differed widely regarding the impacts of CDM on reducing hosting countries' carbon emissions, with some research arguing that CDM has failed to bring about real and additional GHG emission reductions (e.g. Rosendahl and Strand 2009;Schneider 2007) and other work lending considerable evidences to the positive impacts of CDM on emission abatement (e.g. Sutter and Parreño 2007;Huang and Barker 2011). (2011) is among the few work that carries out a cross country study to examine the effects of CDM project development on emission reductions. However, a dummy variable is used for the key regressor CDM in the regression, unable to sufficiently capture the time series variations in CDM development. Given the importance of this issue and the current controversial views on it, this paper revisits this topic by making use of actual CDM credits and actual investments of CDM projects.

Huang and Barker
With the recently-developed panel data econometric method due to Han et al. (2012), it aims to contribute to the field by unveiling the myth of whether or not CDM gives rise to real, additional and measurable emission reductions in the host countries.
The remainder of the paper proceeds in Section 2 by reviewing the literature. Section 3 describes the sample and data. Section 4 outlines the methodology employed. Section 5 conducts estimation and presents the results. Section 6 concludes.

Literature review
The CDM is designed to embody the principle of 'common but differentiated' responsibilities by motivating developed countries to provide funds and technology to reduce GHG emissions in non-Annex I nations in return for credits to offset the carbon emissions produced in their own countries. It not only provides the industrialized countries with some flexibility in how they meet their emission abatement targets but also helps to start off the developing countries on a low-carbon development path.
Despite its fast growth, CDM has been criticized for being ineffective in fulfilling its environmental and sustainable goals. One major concern is its additionality (e.g. Rosendahl and Strand 2009;Schneider 2007). As an offsetting mechanism, CDM projects should contribute to the global emission reductions by bringing down GHG emissions that are 'additional to any that would occur in the absence of the certified project activity' (UNFCCC 2011). That is to say, CDM projects in reality should reduce GHG emissions to the degree that they are credited for (Paulsson 2009). However, Rosendahl and Strand (2009) state that CDM, as an offset mechanism, keeps the global carbon emission level the same by shifting the emission reduction costs from Annex I countries where emission reduction costs are higher, to non-Annex I countries where emission abatement costs are much lower if there were no flaws with CDM. The assessment of additionality in reality has never been easy, either. 7 A related concern is that the integrity of CDM might be contaminated by baseline manipulation (e.g. Rosendahl and Strand 2009). 8 People are also concerned that CDM's focus on low-cost emission reductions might to some degree compromise the host countries' sustainable development (e.g. Paulsson 2009). 9 So far the existing research has a mix of evidences. On the one hand, there is research claiming that CDM has questionable additionality and has failed to bring about real and additional emission reductions to the hosting countries (e.g. Schneider 2007;Rosendahl and Strand 2009). An examination of 93 randomly selected CDM projects as well as interviews and literature review conducted by Scheider (2007) find that about 40 per cent of the CDM registered projects were questionable in terms of additionality, which compromises its environmental integrity. Based on a series of model analysis, Rosendahl and Strand (2009) argue that CDM projects do not imply full offset of GHG mitigation while increasing the likelihood of carbon leakage, which may result in a rise in the global carbon emissions.
On the other hand, there is empirical evidence for CDM's positive impacts on emission reductions (e.g. Lütken 2011;Sutter and Parreño 2007). By comparing the pre-CDM predictions of the mechanism's market size with the projections of current trajectory of potential mitigation entering the CDM pipeline, Rahman et al. (2010) find that, despite its limitations (unbalanced sector composition), CDM has been very successful in achieving emission reductions and is well on the way to reaching an average annual flow of 700 million CERs by 2012. The study by Lütken (2011) on the geographic distributions of CDM projects demonstrates that even the least developed countries (LDCs), especially in Africa, have enjoyed a reasonable share of the total of world's launched CDM projects and witnessed a boom of CDM projects. Based on the analysis of 16 officially registered CDM projects by employing the methodology of multiattributive assessment, it is found that the vast majority of examined CDM projects (72 per cent) are likely to produce real and measurable emission reductions (Sutter and Parreño 2007).
The majority of existing research in this field are based on CDM projects; accordingly, the cross country evidence is lacking. Huang and Barker (2011) is an exception in that it 7 The difficulties with the assessment of CDM's additionality or the evaluation of the actual emission reductions in host countries attributable to CDM projects include: the effects of CDM projects on the emission reductions being indirect and hard to be pinned down to certain projects; the impacts of CDM projects being long term and thus difficult to be measured within a limited crediting period; technological innovations associated with CDM projects leading to positive spill-over effects and further GHG abatement in the host countries (Schneider 2007). 8 Baseline manipulation may occur in the cases where the involving parties in CDM projects have incentives to enhance the baseline of GHG emissions so as to obtain a higher value of CERs than what should have been gained without the manipulation, which may lead to an increase of global GHG emissions (Rosendahl and Strand 2009). 9 More detailed literature review on CDM's sustainable development impacts can be found in its sister paper Huang et al. (2012).
is a cross-country study of 80 CDM host countries for the period between 1993 and 2009. It provides clear evidence for the positive effects of CDM projects on the emission reductions in the host countries. However, one limitation with that study is that due to data shortage a dummy variable is adopted to reflect the change before and after the CDM project registration, failing to sufficiently capture the rich variations in CDM development. This paper serves as a further research to Huang and Barker (2011), by using the actual numbers of CDM credits and investments with the recently developed panel data econometric method associated with X-differencing approach due to Han et al. (2012). It is of great value for this research to investigate this critical issue of whether CDM projects have led to genuine and additional emission reductions in the host countries.

The data
This section describes the variables used in the analysis and data sources. The dependent variable is CO₂ emissions per capita, denoted by CO₂. This analysis mainly makes use of total CO₂ emissions from fuel combustion by sectoral approach (MtCO₂), taken from the Enerdata's Global Energy Market Data (2012). 10 Data for total population are from the World Bank World Development Indicators Database (2012). 11 The key independent variable is the CDM project development (CDM), which could be any of the following four indicators: 12 (1) CDM credits per capita (CER_POP), the ratio of total CDM credits (kCERs) over total population in a given country; 13 (2) CDM Contribution to the Economy (CER_GDP), the ratio of total CDM credits (kCERs) over total GDP. 14 It measures the economic revenue coming out of CDM projects compared to the host country's GDP. It is a direct indicator of the relative importance of CDM projects to the host country's economy or the prominence of CDM activities relative to other economic activities; (3) CDM Actual Emissions Reductions (CER_CO₂), the ratio of total CDM credits (kCERs) over a country's actual carbon emissions. 15 It is the expected emission reductions achieved through CDM projects compared to a country's actual carbon emissions. It gives a rough idea of the domestic emission reductions 10 According to the IPCC Tier 1 Sectoral Approach, the CO₂ emissions by sectoral approach are emissions resulted from the actual fuel combustion. 11 The analysis uses a multiplication factor of 1,000,000; so the dependent variable is tCO₂ emissions per capita. 12 Based on the total numbers of CDM projects and total CDM credits (CERs), Lütken (2011) proposes four indicators: Project Generation Ability, CDM Contribution to the Economy, Investment Capacity and Actual Emissions Reductions. Since the size of CDM projects varies considerably across countries, this analysis focuses on CDM credits (CERs) and actual investments (Million US$) instead. It gives up the indicator of Project Generation Ability and uses the ratio of investment, rather than the number of projects, over GDP for the indicator of Investment Capacity. 13 The indicator is adjusted by a multiplication factor of 1,000 so that it is CERs per capita. 14 A multiplication factor of 1,000,000 is applied, so it is CERs per 1,000 units of GDP. 15 The CERs per tCO₂ emissions (sectoral approach) is used. efforts via CDM and how much the CDM projects can contribute to national emission reductions efforts in a given country; (4) CDM Investment Capability (INVEST_GDP), the ratio of total investments in CDM projects over total GDP. 16 It is an immediate indicator of green FDI via CDM projects relative to GDP, capturing a country's ability to attract external financing for emission reductions.
Data on total CDM credits (kCERs to 2040) and total investments (Million US$) at project validation are provided by UNFCCC colleagues. 17 The data source for total CO₂ emissions is the Enerdata mentioned above. The whole sample includes 60 CDM host countries over 2005-10, as listed in the Appendix Table 1. We exclude any country which has less than 2 annual observations for dependent variable or CDM credits. Descriptive statistics of all variables can be found in Appendix Table 2 and the correlations among variables are presented in Appendix Table 3.
In four separate charts, Figure 1 presents the scatter plots of CO₂ against four CDM indicators mentioned above, respectively. Apart from CDM credits per capita, CO₂ is negatively associated with the other three CDM indicators. Due to the existence of outliers, this evidence alone is not very convincing. Robust evidence will be generated from a detailed panel data econometric analysis of the relationship between CDM and CO₂ emissions in Section 5.

Methodology
This section outlines the bias-corrected estimation methods developed for a dynamic panel data model with fixed effects. This research basically applies the Panel Fully Aggregated Estimator (PFAE) using X-differencing due to Han et al. (2012), against the conventional Least Square Dummy Variables (LSDV) estimator and the bias-corrected Least Square Dummy Variables (LSDVC) estimator due to Kiviet (1995), Bun and Kiviet (2003) and Bruno (2005).
To assess the impacts of CDM projects on the reductions of CO₂ emissions in the host countries, the following model is estimated in this case: where is an unobserved time-invariant country-specific effect and can be regarded as capturing the combined effects of all the omitted variables.
is the unobserved transitory disturbance term, assumed to be independently distributed with zero mean and finite variance, and uncorrelated with individual effects.
Since we expect that reverse effects going from CO₂ emissions to some macroeconomic variables such as GDP per capital, trade openness, and governance level take considerable time, it is reasonable to assume that CDM , GDPPC , GDPPC , TRADE , KAOPEN , and WGI are strictly exogenous with respect to in the sense that these variables are uncorrelated with the earlier, current and subsequent shocks. This assumption rules out the possibility of feedbacks from the past, current and future shocks onto .
For a dynamic panel data model like this, the within-group transformation and firstdifferencing are common approaches used to eliminate any omitted variable bias created by the unobserved individual effects. Since the lagged values of are positively correlated with the omitted fixed effects, as shown by Nickell (1981), the Least Square Dummy Variable (LSDV) estimator in the autoregressive panel models is not consistent and can be badly downwards biased for short time dimension (fixed T), even as cross section dimension N goes to infinity. Starting from Kiviet (1995), a number of biascorrection methods of LSDV for finite samples have been developed. 18 Kiviet (1995) derives an approach to approximating the small sample bias of the LSDV estimator and suggests that the bias approximation can be evaluated at the estimates from some consistent estimates rather than the unobserved true parameter values, which makes bias correction operationally feasible. The Monte Carlo evidence from Kiviet (1995) and Bun and Kiviet (2003) suggest that the bias-corrected LSDV estimator (LSDVC) is more efficient than LSDV, first-differenced 2SLS due to Anderson and Hsiao (1982), first-differenced GMM due to Arellano and Bond (1991) and system GMM due to Blundell and Bond (1998) in terms of bias and root mean squared error (RMSE) for small or moderately large samples. Han et al. (2012) introduce a new estimation method for linear dynamic panel data models with fixed effects, AR (p) idiosyncratic errors and exogenous variables. First, they propose a novel form of systematic differencing, called X-differencing, which has the advantage of removing fixed effects while retaining information and signal strength in cases of persistent or near unit root dynamics. 19 They then suggest using the least square regression method to estimate the stacked and pooled system of X-differenced equations, leading to the PFAE.
( − )( − * ) = 0 for all s < t-1 and |α|<1 (5) Han et al. (2012) suggest making full use of all information by stacking the regression equations for all possible values of s. Given these orthogonality conditions, they then suggest to apply the least squares regression to estimate the full system of X-differenced equations for s=1, 2,...t-3. Han et al. (2012) show that the PFAE estimator is consistent for all parameter values, and has strong asymptotic and finite sample properties that dominate other estimation methods such as LSDVC. There is no need of bias correction for this estimator. In the unit root case, it has higher asymptotic efficiency than LSDVC, while both PFAE and LSDVC are large-T efficient for the stationary case. The asymptotic properties of this estimator hold for short or long panels as well as narrow and wide panels.

Econometric evidence
This section presents the empirical evidence for the effectiveness of CDM project development in terms of emission reductions in 60 CDM host countries over 2005-10. Four CDM indicators explained in Section 3 are examined, with the results reported in Tables 1 to 4, respectively.
Each table compares the LSDV estimates, LSDVC estimates and PFAE estimates. The conventional LSDV estimates are the OLS estimates of coefficients of a panel data regression model after removing the fixed effects via within transformation. The biascorrected LSDV estimates and LSDVC estimates are generated using the approach due to Bruno (2005) for the dynamic unbalanced panels with a strictly exogenous selection rule. 20 The PFAE estimator uses X-differencing approach to remove the fixed effects and applies the least squares regression to estimate the full system of X-differenced equations for s=1,2,3. For each model, the point estimate of the long-run effect of respective CDM variable is calculated with its standard error being approximated by using the delta method (for example, Papke and Wooldridge 2005). The key regressor, CDM credits per capita, has been found negative in the model in which PFAE estimate suggests a significantly negative effect on CO₂ emissions, not only in the short run but also in the long run. 21 This evidence clearly supports that more credits generated from CDM projects over total population in a given CDM host country are associated with more emission reductions.
For the control variables, GDP per capita (GDPPC ) and Squared GDP per capita (GDPPC ), both LSDV and LSDVC estimates imply a positive impact for both of them, although they are almost significant using the LSDV approach while insignificant using the LSDVC approach. The PFAE estimates clearly indicate that GDPPC has a significantly positive impact on CO₂ emissions while GDPPC , a negative impact on CO₂ emissions (significant at 16 per cent level). The evidence is in support of the 20 Bruno (2005) derives a bias approximation of various orders in dynamic unbalanced panels with a strictly exogenous selection rule. Essentially, Bruno (2005) adjusts the within operator to include an exogenous selection rule which only selects the observations with observable current and one-time lagged values, by which missing observations for some individuals are allowed. We use the system-GMM estimator as the initial estimator for the LSDVC since it is believed to be a more reliable estimator than the first-differenced 2SLS and first-differenced GMM estimators. 21 Both LSDV estimate and LSDVC estimate suggest a negative effect on CO₂ emissions at 13 per cent and 24 per cent level, respectively. hypothesis of Environmental Kuznets Curve in the sense that CO₂ emissions increase when the level of per capita income goes up; however, when the income reaches a certain level, CO₂ emissions decline.
For the control variables of openness, trade openness (TRADE) and financial openness (KAOPEN), both LSDV and LSDVC estimates find a positive but insignificant impact for both of them, although TRADE is more likely to have significant impacts on CO₂ emissions. Using the PFAE approach, both TRADE and KAOPEN positively enter the model in which TRADE is significant at 1 per cent level. The observed positive effects of trade openness and financial openness on CO₂ emissions are in line with the existing literature such as Frankel and Rose (2005) that it is likely that more open policies will contribute to increased CO₂ emissions.
Regarding the control variable of institutional quality, World Governance Indicator (WGI), no significant evidence has been generated using three estimation approaches. Both LSDV and LSDVC estimates suggest a positive impact while the PFAE estimate suggests a negative impact, but the latter is more plausible and consistent with the literature. Good governance can facilitate market activities, minimize environmental degradation, and enhance the emission control procedure, etc. (see Gani (2012) for a detailed review). emissions. This shows that the more importance of CDM projects for the economy is not necessarily leading to more CO₂ emission reductions. For the control variables, the pattern of the results is quite similar to those in Table 1. WGI has been found negatively associated with CO₂ emissions at 12.5 per cent significance level, as suggested by the PFAE estimate. Table 3 examines whether the higher probabilities of CDM actual emission reductions (CER_CO₂) could result in more CO₂ emissions in the host countries. All of the LSDV, LSDVC and PFAE estimates indicate a negative impact of CER_GDP on CO₂ emissions, in which the significance suggested by the PFAE is close to the valid significance level. We still do not have sufficient evidence to support that more domestic emission reduction efforts such as CDM project development in a given country will lead to more emission reductions. For the control variables, the pattern of the results is quite similar to those in Table 2. Table 4 focuses on the amount of investments put into the CDM projects (INV_GDP), rather than the CDM credits generated. It presents evidence on whether more investment in CDM projects is associated with more CO₂ emissions in the host countries. All of the LSDV, LSDVC and PFAE estimates exhibit a negative impact of INV_GDP on CO₂ emissions, in which the significant evidence has been suggested by PFAE, but not necessarily by LSDV and LSDVC. The PFAE estimate also indicates that this negative impact is likely to persist into the long run. This finding has an immediate policy implication that the more investments put into the environmentallyfriendly CDM projects or the more foreign investments or green FDI a host country could attract from Annex I countries, the more emission reductions can be expected in the host countries. For the control variable, PFAE estimates have also provided   In sum, by using different CDM indicators, this research produces significant evidence that CDM project development can contribute to CO₂ emission reductions in a given host country. It finds that higher CDM credits per capita and higher investment ratios are linked to more CO₂ emission reductions in both the short run and long run, but not necessarily the case with higher ratios of CDM credits over both total GDP and total CO₂ emissions. It also reveals significant evidence in support of the Environmental Kuznets Curve hypothesis, as well as trade openness and financial openness being the driving factors for CO₂ emissions and governance level being the limiting factor for CO₂ emissions. The findings of CDM being effective for emission reductions in the host countries have important policy implications. Developing countries and the international community should encourage more CDM projects to be established in developing countries and more investments from Annex I countries to be put into the CDM projects, which will be conducive to the global low-carbon development cause.
The results are not due to unobserved heterogeneity, and in general robust to the use of different CDM indicators, which measure different dimensions of CDM project development in a given country.

Conclusion
This paper focuses on the effectiveness of the CDM in achieving carbon emission reductions in the host countries. It demonstrates that CDM projects as a whole have brought 'real' and 'additional' emission reductions to the host developing countries. The results imply that the criticisms against CDM's additionality and the raised issue of the manipulations of and even corruptions surrounding CDM projects (e.g. baseline manipulation), are at least overstated in most cases. The results also indicate that, despite its limitations, the Kyoto Protocol has been very successful in not only setting legally binding emission abatement targets but also offering the means to help to achieve the end.
According to the IPCC report (2001), developing countries are going to overtake developed countries in terms of GHG emissions between 2010 and 2020. At the same time, it is important to note that the developing nations are still struggling to cope with the immediate development concerns. This reality means that on the one hand, the positive involvement of developing nations in the global efforts of GHG emission reductions is a necessity for the world's climate change combat to be successful, and on the other hand, to get the developing countries on board in the global emission abatement campaign, more emphasis is needed to enhance the financial and technological transfers into developing countries, for which CDM is an ideal channel.
This once again highlights that the importance of CDM's role in facilitating the Annex I countries to achieve the targeted emission reductions as set by the Kyoto should not be overlooked. By making it easier for Annex I countries to meet their emission targets, CDM along with the other two flexibility mechanisms, helps to ensure the lasting success of the Kyoto and its renewal into the next commitment period. Although improvements are required, CDM is one of such essential means that a future global climate regime cannot afford to be without.