Predicting Chinese carbon prices and influence factors: evidence from quantile shrinkage methods

This paper aims to forecast Chinese carbon prices by utilizing a variety of predictors and analyzing their impact across various carbon market conditions and emergency (“black swan†) events. Advanced models, including shrinkage methods, quantile shrinkage methods, dimension reduction methods, autoregressive models, combination forecasting versions of autoregressive models with exogenous indicators and machine learning tools, are employed. To assess their predictive capabilities, we incorporate five groups of macroeconomic variables alongside 18 technical indicators. The findings reveal that using a rolling window forecasting approach, the quantile group least absolute shrinkage and selection operator (LASSO) model exhibits superior prediction accuracy compared with other models. The robustness of the results is further validated through an alternative carbon market, a recursive window rule with varying window lengths and an economic value analysis. In addition, this study explores how the effects of statistically significant factors vary across quantiles and emergency events. This study provides market participants and policy makers with an effective tool for Chinese carbon price forecasting, helping them identify key predictors and make informed investment decisions under various market conditions.