基于因素分解与ARIMA混合模型的广州碳排放权交易价格预测

Price Forecasting of Guangzhou Carbon Emission Trading Based on a Hybrid Factor Decomposition and ARIMA Model

摘要: 在“双碳”目标背景下，碳排放权交易对企业资产配置及风险管理具有重要意义，在国家层面的宏观调控以及履约期限的影响下，碳价序列往往表现出非平稳性和异方差等问题，准确预测碳价有利于企业管理和控制碳资产，也有助于政府进一步改进碳价形成机制。基于此本文选取广州碳排放权交易市场2018年7月6日至2026年1月20日的日度数据作为研究对象，首先进行对数差分和平稳性检验后，应用赤池信息准则建立自回归移动平均模型提取其线性部分，然后由于该模型对于极端值无法给出较好的估计，所以在基础模型之上运用时间序列加法分解剔除其长期趋势、季节效应以及随机扰动项得到一个新的预测值，并将其与基础模型预测结果相结合形成混合预测的方法。结果表明广州碳价存在长记忆性以及履约期内季节性变化的特点；而且相比起基础模型而言，采用混合方法后可以使得预测均方根误差降低约9.58%，说明其在减小较大预测误差方面具有一定优势。本文也为碳金融市场上的资产价格预测提供了一种思路，在企业融资约束以及税务处理方面也给政府带来一定的启示。

Abstract: Under the background of the “Dual Carbon” goals, carbon emission trading is of great significance to corporate asset allocation and risk management. Influenced by national macro-control and compliance periods, carbon price time series often exhibit non-stationarity and heteroscedasticity. Accurately forecasting carbon prices not only benefits enterprises in managing and controlling their carbon assets but also assists the government in further improving the carbon price formation mechanism. Accordingly, this paper selects the daily trading data of the Guangzhou Carbon Emission Trading Market from 6 July 2018 to 20 January 2026 as the research object. First, after performing logarithmic differencing and stationarity tests, an autoregressive moving average model is established using the Akaike Information Criterion (AIC) to extract the linear components of the series. Subsequently, considering that the base model struggles to accurately estimate extreme values, a time series additive decomposition method is employed to strip out the long-term trend, seasonal effects, and random disturbances. These extracted components are then combined with the prediction results of the base model to formulate a hybrid forecasting approach. The empirical results demonstrate that the Guangzhou carbon price possesses characteristics of long memory and seasonal variations during the compliance period. Furthermore, compared to the single base model, the proposed hybrid approach reduces the forecasting Root Mean Square Error (RMSE) by approximately 9.58% indicating an advantage in reducing larger forecasting errors. This paper offers a novel perspective for asset price forecasting in carbon financial markets and provides insights for policymakers regarding corporate financing constraints and taxation treatments.