Scholarlink

<p>This study aims to forecast global CO<sub>2</sub> emissions by examining the dynamic and evolving influence of economic, geopolitical, and policy uncertainties through the application of traditional, machine learning, deep learning, and hybrid models. The analysis is conducted on a monthly dataset from 2005 to 2024, incorporating key input variables such as CO<sub>2</sub> futures prices, trade policies, geopolitical risk, financial regulation, and climate policy uncertainty. The full sample and the post-Paris Agreement period (2016–2024) are analyzed to capture shifts in the impact of external factors over time. The models applied in this study include traditional methods such as Seasonal and Trend decomposition (STL), machine learning models including Random Forest, XGBoost, and SVR, deep learning models like LSTM, GRU, and CNN, and hybrid models such as CNN-LSTM and STL-LSTM. The results indicate that in the full sample period, trade-related and financial regulatory uncertainties (TRADE, FINR) emerge as key predictors of CO<sub>2</sub> emissions, whereas geopolitical risk (GPRT) plays a secondary role. However, during the post-Paris Agreement period, financial regulation (FINR), geopolitical uncertainty (GPRT), and environmental regulation (EEREG) become more influential, reflecting a shift from traditional trade-driven dynamics to macro-financial and policy-based emission drivers. Among the models, the hybrid STL-LSTM architecture consistently outperforms other approaches by effectively leveraging trend and seasonal decomposition, which enhances long-term dependency learning and improves predictive robustness, particularly in volatile policy environments. By decomposing the CO<sub>2</sub> time series into trend and seasonal components using STL, the model reduces noise interference and enhances LSTM’s ability to learn long-term dependencies. This structure increases robustness, particularly in volatile periods. Furthermore, the study uniquely integrates geopolitical, climate policy, and financial regulatory uncertainties into a unified forecasting framework, which marks a novel contribution to the literature on emission forecasting.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Forecasting Global CO2 Emissions Under Economic, Geopolitical, and Policy Uncertainties: A Novel Hybrid Model

  • 期刊论文

  • 发表日期：

    2025年8月2日

查看全文
  • İhsan Erdem Kayral,
  • Melike Aktaş Bozkurt,
  • Tuğba Sarı,
  • Nisa Şansel Tandoğan Aktepe

摘要

This study aims to forecast global CO2 emissions by examining the dynamic and evolving influence of economic, geopolitical, and policy uncertainties through the application of traditional, machine learning, deep learning, and hybrid models. The analysis is conducted on a monthly dataset from 2005 to 2024, incorporating key input variables such as CO2 futures prices, trade policies, geopolitical risk, financial regulation, and climate policy uncertainty. The full sample and the post-Paris Agreement period (2016–2024) are analyzed to capture shifts in the impact of external factors over time. The models applied in this study include traditional methods such as Seasonal and Trend decomposition (STL), machine learning models including Random Forest, XGBoost, and SVR, deep learning models like LSTM, GRU, and CNN, and hybrid models such as CNN-LSTM and STL-LSTM. The results indicate that in the full sample period, trade-related and financial regulatory uncertainties (TRADE, FINR) emerge as key predictors of CO2 emissions, whereas geopolitical risk (GPRT) plays a secondary role. However, during the post-Paris Agreement period, financial regulation (FINR), geopolitical uncertainty (GPRT), and environmental regulation (EEREG) become more influential, reflecting a shift from traditional trade-driven dynamics to macro-financial and policy-based emission drivers. Among the models, the hybrid STL-LSTM architecture consistently outperforms other approaches by effectively leveraging trend and seasonal decomposition, which enhances long-term dependency learning and improves predictive robustness, particularly in volatile policy environments. By decomposing the CO2 time series into trend and seasonal components using STL, the model reduces noise interference and enhances LSTM’s ability to learn long-term dependencies. This structure increases robustness, particularly in volatile periods. Furthermore, the study uniquely integrates geopolitical, climate policy, and financial regulatory uncertainties into a unified forecasting framework, which marks a novel contribution to the literature on emission forecasting.

Graphical Abstract