Econ Papers Daily / 期刊论文
解构中国制造企业的能源低效：基于机器学习评估方法的证据
Deconstructing energy inefficiency in China's manufacturing enterprises: Evidence from a machine learning evaluation method
摘要
As China's manufacturing industry enters a pivotal phase of high-quality development, accurately assessing its energy efficiency (EE) has become essential for advancing green transformation. However, traditional approaches to measuring EE are prone to distortion from extreme observations, thereby limiting their effectiveness in capturing firm-level EE. Therefore, machine learning (ML) techniques are employed in this study to optimize EE assessment. The primary innovations of this paper are summarized as follows. First, ML-based decision-making units are constructed and introduced into the traditional efficiency evaluation model. Second, EE at the firm level is precisely measured using the proposed model, which improves the reliability of efficiency evaluation. Third, inefficiency is decomposed into internal and external components to identify targeted improvement directions. The empirical analysis based on the panel data of listed manufacturing firms in China yields several key insights. First, the proposed method provides a more rigorous efficiency benchmark and improves the discrimination among efficient firms. Second, firm-level EE exhibits a hierarchical distribution pattern with a small number of highly efficient firms and a large number of relatively low-efficiency firms. Third, EE varies significantly across industries and regions. Fourth, inefficiency mainly originates from either internal or external constraints, indicating heterogeneous improvement directions among firms. Accordingly, a set of differentiated policy recommendations tailored to these firms is proposed to improve their EE.
中文摘要
随着中国制造业进入高质量发展的关键阶段，准确评估其能源效率（EE）已成为推进绿色转型的必要条件。然而，传统的能源效率测度方法易受极端观测值扭曲，从而限制了其在捕捉企业层面能源效率方面的有效性。因此，本研究采用机器学习（ML）技术来优化能源效率评估。本文的主要创新点概括如下。第一，构建基于机器学习的决策单元，并将其引入传统效率评价模型。第二，利用所提出的模型精确测度企业层面的能源效率，从而提高效率评价的可靠性。第三，将无效率分解为内部和外部两个组成部分，以识别有针对性的改进方向。基于中国制造业上市公司面板数据的实证分析得出若干关键发现。第一，所提出的方法提供了更严格的效率基准，并提高了对有效率企业之间的区分度。第二，企业层面能源效率呈现分层分布格局，即少数高效率和大量相对低效率企业并存。第三，能源效率在不同行业和地区之间存在显著差异。第四，无效率主要来源于内部或外部约束，表明企业之间的改进方向具有异质性。据此，本文提出一套针对这些企业量身定制的差异化政策建议，以提升其能源效率。