A Two-Stage SARIMA–GARCH Approach to Forecasting Global Monthly Carbon Dioxide Emissions: Trend, Seasonality, and Mitigation Implications
Year: 2026
Type: article
Abstract: Accurate near-term projections of carbon dioxide (CO₂) emissions help governments gauge progress toward the Paris temperature goals and time demand-side mitigation. Most forecasts rely on annual figures, overlook the pronounced within-year movement of emissions and their reaction to sudden shocks, and rarely model the resulting volatility. Using the IEA–EDGAR v8.0 global monthly fossil CO₂ inventory (156 observations, January 2010–December 2022), this study builds a two-stage model: a seasonal autoregressive integrated moving-average (SARIMA) mean equation selected by an exhaustive search, followed by a GARCH(1,1) variance equation when an ARCH-LM test detects heteroskedasticity. STL decomposition shows a mild upward trend with a strong annual cycle—the seasonal component accounts for 73.9% of short-term fluctuation, with December–January peaks and August–September troughs. The selected mean model is SARIMA(1,0,0)(1,1,1)12 with drift (Ljung–Box p = 0.96), and the GARCH(1,1) removes the remaining ARCH effect (p = 0.27). With MAPE of 1.01% in sample and 2.55% in a held-out 2022 test set and Theil's U of 0.76, it projects 38.86 Gt in 2023 and 39.00 Gt in 2024, a rise of only 0.38% consistent with the Global Carbon Budget's near-plateau. A plateau is far from the rapid, sustained cuts needed for 1.5 °C, so the paper closes with season- and sector-targeted mitigation measures.
Author: Jingyi Guo
Institution: Xi’an Jiaotong-Liverpool University
Language: English
Subfield: Economics and Econometrics
Domain: Social Sciences
SDG: Climate action
Open Access status: hybrid
OpenAlex ID: W7235768444
