Forecasting carbon emissions using asymmetric grouping

This paper proposes an asymmetric grouping estimator for forecasting per capita carbon emissions for a country panel. The estimator relies on the observation that a bias-variance pooling trade-off in potentially heterogeneous panel data may be different across countries. For a specific country, cross validation is used to determine the optimal country-specific grouping. A simulated annealing algorithm deals with the combinatorial problem of group selection in large cross sections. A Monte Carlo study shows that in case of heterogenous parameters, the asymmetric grouping estimators outperforms symmetric grouping approaches and forecasting based on individual estimates. Only in the case where the signal is very weak, pooling all countries leads to better forecasting performance. Similar results are found when forecasting carbon emission. The asymmetric grouping estimator leads to more pooling than a symmetric approach. Being on the same continent increases the probability of pooling, and African countries seem to benefit most from using asymmetric grouping and European countries least.