Carbon prices forecasting in quantiles

This paper proposes two new methods (the Quantile Group LASSO and the Quantile Group SCAD models) to evaluate the predictability of a large group of factors on carbon futures returns. The most powerful predictors are selected through the dimension-reduction mechanism of the two models, while potential differences of the statistically significant predictors for different quantiles of carbon returns are carefully considered. First, we find that the proposed models outperform a series of competing ones with respect to prediction accuracy. Second, impacts of the selected predictors over the carbon price distribution are estimated through a quantile approach, which outperforms the mean shrinkage model in our case with data featured by a non-normal distribution. Specifically, the Brent spot price, the crude oil closing stock in the UK, and the growth of natural gas production in the UK are found to impact carbon futures returns only in extreme conditions with a strong asymmetric feature. Importantly, our estimators remain robust against the extreme event caused by the Covid19. Our findings reveal that the identification of appropriate carbon return predictors and their impacts hinge on the carbon market conditions, and should be of interest to various stakeholders.

Automated summary

This paper proposes two new methods, the Quantile Group LASSO and the Quantile Group SCAD models, for evaluating the predictability of a large group of factors on carbon futures returns. The models select the most powerful predictors through dimension-reduction mechanisms and consider the potential differences in statistically significant predictors for different quantiles of carbon returns. The findings show that the proposed models outperform competing ones in terms of prediction accuracy and estimate the impacts of selected predictors on the carbon price distribution using a quantile approach.