Forecasting CO2 Emissions in the United States: A Comparative Study of Econometric Models

Details

Supervisors

Iania, Leonardo

Faculty

Louvain School of Management

Degree label

Master [120] : ingénieur de gestion, à finalité spécialisée

Abstract

As the global community faces the urgent challenges of climate change, forecasting carbon dioxide (CO2) emissions and understanding their determinants in the United States has become imperative for effective climate action. This thesis presents a comprehensive study of the performance of nine econometric models for forecasting CO2 emissions in the United States over different time horizons (4, 6, 12, 18 and 24 months). The research focuses on the comparison of models derived from dimensionality reduction techniques, including shrinkage methods and factor analysis, while employing a large dataset comprising 127 macroeconomic variables observed at a monthly frequency and CO2 emissions data covering a period from January 1973 to September 2022. Additionally, the study examines the impact of incorporating lagged values of the dependent and explanatory variables as predictors. The in-sample analysis attests to the superior performance of shrinkage methods and ARDL (Autoregressive Distributed Lag) models in fitting the data. Furthermore, the incorporation of lagged values of the dependent variable as predictors demonstrates a positive impact on the models’ predictive ability. However, the out-of-sample analysis does not yield conclusive results, as none of the models outperforms the benchmark autoregressive model of past monthly growth rates of CO2 emissions. The presence of noise in the data due to observation frequency is suggested as one potential reason for this result.