Do bilateral trade relationships influence the distribution of CDM projects?

Abstract

The use of the Clean Development Mechanism (CDM) is increasingly widespread in developing countries. However, CDM projects are still far from being an effective development activity due to the uneven distribution of these projects in a few relatively well-off economies. One potential cause of this imbalance is analysed in terms of the trade relationships between developed and developing countries. By applying a gravity model to a panel dataset, well-established export flows from developed economies towards developing countries are shown to explain why a large proportion of CDM projects are unevenly geographically distributed. This kind of lock-in effect regarding the CDM between developed and developing countries could be avoided by both enhancing the institutional framework in developing countries that host CDM projects and reinforcing compulsory rules for CDM destinations in the least-developed economies.

Policy relevance

The results presented in this article are relevant in two ways to the ongoing climate change negotiations regarding the future of the Kyoto Protocol and its various mechanisms, and more generally to the fight against climate change and its impacts on developing countries. First, in order to overcome the lock-in effect created by export flows from developed to developing countries, there should be an ad hoc policy action to redistribute CDM investments to developing countries. Second, there is also a need for the institutional framework in developing countries that host CDMs to be enhanced because it is a major factor in reducing transaction costs and the risk of uncertainty. This framework could therefore provide a stable environment for investment decisions.

Keywords:

• Clean Development Mechanism
• export flows
• gravity model
• institutional quality
• Kyoto Protocol

Acknowledgements

We would like to thank Nicola Cantore, Massimiliano Mazzanti, and three anonymous reviewers for their helpful suggestions. The usual disclaimers apply.

Funding

Financial support was received from the Roma Tre University-INEA-ENEA Consortium, the EU D.G. Research (research project ‘CECILIA2050 – Choosing efficient combinations of policy instruments for low-carbon development and innovation to achieve Europe's 2050 climate targets’, grant agreement no. 308680), and the Italian Ministry of Education, University and Research (Scientific Research Program of National Relevance 2010 on ‘Climate change in the Mediterranean area: scenarios, economic impacts, mitigation policies and technological innovation’).

Notes

1. The Herfindahl index ranges from 1/N to 1, where N is the number of host countries. To compute a fully comparable index for the selected countries, the same host countries for all investors with a common N for all have been considered.

2. Looking at distinguished host countries for the four top investors (i.e. the UK, Switzerland, Japan, and the Netherlands), it is also worth noting that the largest concentration is given by increasing projects directed towards China, with a clear crowding-out effect compared with projects directed towards all other destinations.

3. Controls for the role of past colonial relationships are also included, but it is not a robust regressor. Results are available upon request from the authors.

4. Note that the other bilateral measure represented by FDI flows is a weak statistical source in terms of country coverage and time span in comparison with trade statistics.

5. For a synthetic description of variables and data sources see the Appendix, Table A1.

6. Alternative methods are used for variables with excessive zeros (zero-inflated negative binomial regression, Hurdle models, etc.). See Cameron and Trivedi (2009) for a more comprehensive discussion.

7. The mean and variance for the dependent variable were 0.15 and 7.96, respectively. The likelihood-ratio test on the overdispersion parameter is 4511.5, where p < 0.001, thus providing strong evidence for the overdispersion. Consequently, the NBRM was preferred to the PRM.

8. The ML negative binomial mean-dispersion estimator is not consistent if the variance specification is incorrect. As an alternative estimation strategy, the basic equation was estimated with the PRM using the pseudo ML approach. This approach only requires that the conditional mean function is correctly specified and also allows consistent estimation of the coefficients if the count variable is not Poisson distributed (Wooldridge, 1999). As a further robustness check, models were estimated with the generalized method of moments (GMM). The GMM estimator is of special interest when the model includes variables that are not strictly exogenous. In both cases, results did not change substantially. Thus, the following reported results are those based on the NBRM, which in the absence of significant changes in the estimated coefficients remains the most efficient estimation method. All results based on pseudo ML and the GMM approaches are available upon request from the authors.

9. The Hausman test points out that the fixed-effects estimator is more appropriate than the random-effects estimator. As an example, the Hausman test computed for model M3 in Table 2 assumed a value of 269.00 (with a P-value of 0.00), thus rejecting the null hypothesis that difference in coefficients is not systematic.

10. It is worth mentioning that the original formulation of an NBRM model is given by an exponential function and that statistical packages often automatically transform the equation in a log-linear form, as exactly represented by equation (2). In this case, coefficients for log-transformed regressors are interpreted as elasticities, while coefficients related to variables in level (in this case all dummy variables) represent semi-elasticities.

11. Because potential multicollinearity bias may arise from considering economic and energy variables simultaneously, correlation values between GDP per capita and alternatively CO₂ emissions and all other energy-related variables were controlled for. All correlation values were below 0.30. As a further robustness check, multicollinearity bias was controlled for by computing variance inflation factor (VIF) values for all covariates included in models M3 to M5 as well as condition numbers for the whole regressions. Values obtained for VIF for single variables were always below 5.00, and mean VIFs for the whole regressions were always below 2. For condition numbers, the condition of statistics below 50.00 was always respected. For the sake of simplicity, statistics are not reported in the Tables, but are available upon request from the authors.

12. It is worth mentioning that the number of effective observations given by the empirical results is substantially lower than the potential ones. This is due to the structure of a gravity model itself, because it requires that all j countries are represented, even if they have no bilateral flows for the whole period. In this case a zero value is given and in the NBRM model these observations were automatically dropped because a log-transformed equation was estimated. As a robustness check for reduced observations, an econometric estimation of models M4 to M6 in Table 3 and models M7 to M9 in Table 4 were also developed where the dependent variable was represented by a pure binary variable, which was assigned a value of 1 if there was at least one project developed by each ith investor in each jth non-Annex I country, and a value of zero otherwise. The econometric estimator here adopted was a panel probit model, and all results on bilateral trade remained robust and statistically significant. For the sake of simplicity, results are not included here, but are available upon request from the authors.