Forecasting U.S. State-Level Carbon Dioxide Emissions

This study explores the use of spatial models in forecasting U.S. state-level carbon dioxide emissions. We compare forecasts against empirical reality using panel data models with and without spatial effects. Understanding how to predict emissions is important for designing climate change mitigation policies. To determine if spatial econometric models can help us predict emissions, it is important to test these models to see if they are a valid strategy to describe the underlying data, in the context of forecasting. We find that a non-spatial OLS estimator performs best in all out-of-sample forecasts; however, the OLS model is not statistically distinguishable from a spatial panel data model with random effects.