4.1. Samples and Data
Guangdong Province is a major forestry province in China, with a forest area of 1.106 million hectares and a forest coverage rate of 56%. According to the available statistics, Guangdong’s annual carbon emissions exceed 300 million tons, accounting for more than one-tenth of the country’s total. It is also one of the first pilot provinces for forest carbon sink projects in the country. To achieve emission reduction targets, Guangdong Province has been committed to improving the quantity and quality of forestry resources for several consecutive years, and the degree of development of carbon sink plantations is now at the forefront of all provinces in China. In 2011, the Changlong Carbon Sink Plantation Project was implemented in the relatively underdeveloped barren hills of Guangdong Province, where the carbon sink plantation areas of Wuhua County, Xingning City, Zijin County, and Dongyuan County were 266.7 ha, 266.7 ha, 200.0 ha, and 133.3 ha, respectively. In May 2015, the first phase of the project’s emission reductions was issued, making it the first Chinese Certified Emission Reduction (CCER) Project to receive emission reductions issued by the National Development and Reform Commission. The Changlong Carbon Sink Plantation Project in Guangdong Province is the most representative because of its long implementation time and established project operation [
36]. Therefore, this study includes four counties that are implementing the Changlong Carbon Sink Plantation project as the treatment group and other counties that have not implemented the project as the control group. In view of the fact that there are individual counties (districts) in Guangdong Province that have undergone changes at the administrative level, we exclude these counties from the empirical study. At the same time, we take 2011 as the starting point for the effects of carbon sink plantation policy, and select 6 years before the policy time point, that is, 2006–2011, and 7 years after the policy time point, that is, 2012–2018, as the sample period for policy evaluation. Finally, panel data of 56 county samples from 2006 to 2018 were collected.
The data selected in this study are mainly sourced from the China County Statistical Yearbook, the Guangdong Rural Statistical Yearbook, and the Guangdong Statistical Yearbook for the years of interest.
4.2. Model Construction
The site selection of carbon sink afforestation projects has strict requirements on objective factors such as land type, property rights, and area. However, there is a positive relationship between natural conditions and economic levels, which will affect the research results of this study via factors arising from differences in the natural environment that are omitted or difficult to include in the model, resulting in inevitable endogeneity. Therefore, we tried to use the DID method to account for the presence of endogenous problems, test the net effect of the carbon sink afforestation project on the county economy, and accurately evaluate the policy effect of the project implementation. Because of the different economic development speeds and levels of the counties in Guangdong Province, the treatment group and the control group, which were obtained by matching only the implementation of the project in the total sample, may have significant differences in originality, which in turn makes the research results appear to have some degree of deviation. Therefore, before running the DID model, we used the propensity score matching (PSM) method to match the treatment group and the control group. When performing PSM, we first divided each sample county into two categories: counties where carbon sink afforestation began in 2011 are treated as the treatment group, and counties that have never participated in the carbon sink afforestation project as of the time of the study are the control group. This study used the kernel matching method to match each treatment group and the control group that is not affected by the policy (before 2011). For example, the goal is to find county A in the control group and make it as similar as possible to the observable variables of county B in the treatment group that implements carbon sink afforestation. If a county’s participation in the project and the probability of participating in the project completely depends on the abovementioned observable control variables, the two counties have a similar probability of implementing the policy. The PSM method calculates the propensity score according to the matching index, and then matches the sample of individuals included in the treatment and control groups based on the similarity of the propensity score between them, which can reduce the estimation error caused by sample selection bias and, thus, improve the credibility of the empirical study results. Then, in the new matched samples, the policy effects of carbon sink afforestation on regional economic development were examined according to the DID method.
The DID model is commonly used in policy analysis and project evaluation, and is usually used to estimate the net effect of a policy or project. It can be used to solve the aforementioned endogenous problems that cannot be quantified, and then evaluate the net effect of carbon sequestration afforestation projects. The common single difference method used in the literature compares the difference of regional economic growth before and after the implementation of the project to judge the effect of the policy on economic growth, but its conclusion may be inaccurate. In addition to many other factors that affect regional economic growth, other policies issued during the same period may also enable cities that have not implemented carbon sink afforestation to develop. These factors will affect the accuracy of the evaluation results. Therefore, the effect of carbon sink afforestation needs to be evaluated under a more scientific DID method, because the model combines temporal and spatial differences by setting up control groups and treatment groups, which mitigates the effect of other unpredictable factors. In this study, we compare the empirical results of the single-difference method and the DID method.
For the DID method, we used the following model:
In Formula (1), is a proxy variable that measures the economic development level of the county, is a constant term, “treated” is used to distinguish the treatment group from the control group, t is a dummy variable used to distinguish between before and after project implementation, the cross-product term “treated·t” is the core explanatory variable for measuring whether the carbon sink plantation project is implemented, represents the net effect of carbon sink plantation on the county’s economic development, represents the coefficient of each control variable, “control” includes the added value for the primary industry (pri), the added value for the secondary industry (sec), various loan balances of financial institutions at the end of the year (fin), general fiscal budget income (inc), and expenditure (exp), refers to controlling individual fixed effects, represents time fixed effects, and εit is a random interference term.
The above model was used to estimate the average benefit of carbon sink plantation projects. We further examined the dynamic effects of the carbon sink plantation project on economic development since the implementation of the project using the following model:
In Formula (2), is the cross-product term “treated × ”, which is a dummy variable for the kth year since the carbon sink plantation project has launched in a county. After the carbon sink plantation project is implemented in this county, it will be assigned a value of 1 only in the kth year of implementation, and a value of 0 in other years. The coefficient is used to represent the economic benefits brought by the implementation of the policy in the kth year after the implementation of the project.
To analyze the effect mechanism of carbon sink plantation projects on economic development, the following model settings were used:
Each variable of the control variables is used as an explained variable in this formula, and we perform OLS regression with policy variables cross-multiplied respectively.