TL;DR
This paper introduces a novel stochastic two-stage Data Envelopment Analysis (DEA) model to assess the environmental efficiency of industrial production and waste gas treatment in China's "One Belt and One Road" regions. The model accounts for randomness in both desirable and undesirable outputs, offering a more realistic evaluation than traditional deterministic methods. Findings reveal that many regions are inefficient, with waste gas treatment being a particular bottleneck, and suggest tailored improvement strategies based on regional characteristics.
What Problem Does This Paper Address?
China's rapid industrialization has led to significant air pollution, posing serious environmental and health challenges. While many studies have evaluated environmental efficiency, few have specifically addressed the complexities of industrial production and waste gas treatment as a sequential process, especially when dealing with the inherent uncertainty in real-world data. Traditional DEA models often treat systems as "black boxes" and assume deterministic inputs and outputs, which can lead to inaccurate efficiency assessments. This paper aims to fill this gap by developing a more robust model that can handle the two-stage nature of industrial processes and the stochasticity of their outputs, particularly focusing on the "One Belt and One Road" initiative regions in China.
Methods
The core contribution of this paper is the development of a novel non-radial stochastic two-stage DEA model. This model extends existing DEA frameworks in several key ways:
- Two-Stage Structure: It explicitly models the industrial process as two sequential stages: Stage 1 (industrial production) and Stage 2 (waste gas treatment). The undesirable outputs from Stage 1 (like waste gas produced) become inputs for Stage 2.
- Stochasticity: It incorporates randomness in both desirable and undesirable outputs, acknowledging that real-world data is rarely perfectly predictable. This is achieved using chance constraints and probabilistic formulations.
- Non-Radial: Unlike radial DEA models that scale all inputs or outputs proportionally, this model allows for independent adjustments of different inputs, offering a more flexible and realistic efficiency assessment.
- Weak Disposability: It accounts for the concept of "weak disposability" for undesirable outputs, meaning that reducing pollution might require using more resources, a common scenario in environmental management.
- Transformation to Linear Programming: The complex stochastic model is mathematically transformed into an equivalent linear programming problem, making it computationally tractable.
The paper also analyzes the sensitivity of the model to a risk parameter ($\alpha$), which reflects the decision-maker's attitude towards uncertainty.
Datasets
The study utilizes data from the 2016 China Industrial Statistical Yearbook, the 2012–2016 China Environmental Yearbook, and the statistical yearbooks of 18 provinces and cities located along China's "One Belt and One Road" initiative. The data covers the year 2015.
The variables used include:
- Stage 1 Inputs: Labor, fixed assets investment, total industrial energy consumption.
- Stage 1 Desirable Output: Industrial production value.
- Stage 1 Undesirable Output (Intermediate Output): Industrial waste gas production.
- Stage 2 Input: Annual expenditure for operation.
- Stage 2 Desirable Output: Industrial waste gas removal.
Key Insights and Ideas
- Waste Gas Treatment is a Major Bottleneck: The study consistently shows that waste gas treatment (Stage 2) is significantly less efficient than industrial production (Stage 1) across most regions. This highlights a critical area for environmental improvement.
- Stochasticity Matters: The stochastic DEA model provides a more nuanced and often more realistic assessment of efficiency compared to deterministic models. Many regions that appear efficient under deterministic assumptions are revealed to be inefficient when uncertainty is considered.
- Regional Tailoring is Crucial: Efficiency scores vary significantly across regions. This implies that a one-size-fits-all approach to environmental policy is unlikely to be effective. Different regions require distinct strategies.
- Risk Attitude Influences Efficiency: The choice of the risk parameter ($\alpha$) significantly impacts efficiency scores, underscoring the importance of considering decision-makers' risk preferences when evaluating environmental performance.
Key Results
- Widespread Inefficiency: Approximately half of the 17 evaluated regions along the "One Belt and One Road" are found to be inefficient in terms of their overall environmental performance in the industrial sector.
- Stage 2 Underperformance: The efficiency scores for waste gas treatment are notably lower than those for industrial production.
- Regional Differences:
- Southeast regions (e.g., Shanghai) tend to perform well in industrial production but struggle with waste gas treatment.
- Northwest regions (e.g., Heilongjiang, Gansu) show poorer performance in industrial production but better results in waste gas treatment.
- The Southwest region generally exhibits the best overall efficiency, while the Southeast region is the poorest performer.
- Model Comparison: The stochastic two-stage model identifies fewer efficient DMUs (Decision Making Units) and generally yields lower average efficiency scores compared to a single-stage stochastic DEA model, suggesting its greater discriminating power. For instance, at $\alpha = 0.5$, only one DMU is efficient under the two-stage model, compared to nine under the single-stage model.
Limitations and Future Work
- Limited Scope of Randomness: The current model primarily focuses on randomness in outputs. Future work could extend this to include stochastic inputs as well.
- Network Complexity: The paper acknowledges that real-world production processes can be more complex than a simple two-stage structure, suggesting the need for models that can handle shared inputs and common outputs in more intricate network configurations.
- Risk Parameter Range: While the study uses $\alpha \le 0.5$ based on previous research, future work could explore the implications of a wider range of $\alpha$ values and investigate methods for determining optimal $\alpha$ values for different contexts.
- Data Granularity: The study uses provincial-level data. Future research could benefit from more granular data (e.g., city-level) to provide more targeted insights.
Summary updated