融合政策虚拟变量的碳排放预测研究

Carbon Emissions Forecasting Based on Policy Dummy Variable Integration

摘要: 近年来，碳排放已成为全球可持续发展的核心约束因素。中国在“双碳”目标战略下亟需科学的碳排放预测工具，现有研究多基于时间序列或回归方法，或通过多方法融合以提升预测精度，然而普遍缺乏对政策变量的显式建模，对政策影响的考察多停留于情景分析，难以揭示政策因素对碳排放演化的结构性影响。为此，本文基于STIRPAT模型，提出一种融合分段政策虚拟变量与岭回归的预测框架。通过遗传算法优化断点设置与正则化参数，用于刻画政策冲击的结构性效应。为检验引入政策虚拟变量的扩展模型的有效性，本文采用1980~2019年数据进行模型训练，并利用2020~2022年数据开展样本外测试。研究结果显示，扩展模型在训练集与测试集上的预测误差均低于基础模型。由此可见，政策虚拟变量的引入不仅增强了模型的拟合能力，同时有效改善了其在样本外的推广性能，体现出更强的稳健性与适用性。

Abstract: In recent years, carbon emissions have become a core constraint on global sustainable development. Under China’s “dual-carbon” strategy, accurate forecasting tools are urgently needed. Existing studies primarily rely on time-series and regression models, or combine multiple methods to improve prediction accuracy, yet they generally lack explicit modeling of policy variables, with policy impacts often limited to scenario analyses. To address this gap, this study develops an extended STIRPAT-based forecasting framework that integrates segmented policy dummy variables and ridge regression. Genetic algorithms are employed to optimize breakpoint selection and regularization parameters, enabling structural characterization of policy shocks. Using data from 1980~2019 for training and 2020~2022 for out-of-sample testing, the results show that the extended model achieves lower prediction errors than the baseline model in both datasets. The inclusion of policy dummy variables not only enhances model fitting accuracy but also improves its generalization performance, demonstrating stronger robustness and policy interpretability.