Nonparametric modeling of carbon prices

This paper constitutes the first exercise of nonparametric modeling applied to carbon markets. The framework of analysis is carefully detailed, and the empirical application unfolds in the case of BlueNext spot and ECX futures prices. The data is gathered in daily frequency from April 2005 to April 2010. First, we document the presence of strong nonlinearities in the conditional mean functions. Second, the conditional volatility functions reveal an asymmetric and heteroskedastic behavior which is dramatically different between carbon spot and futures logreturns. The results for spot prices are also robust to subsamples' decomposition. Third, we show in an out-of-sample forecasting exercise that nonparametric modeling allows reducing the prediction error by almost 15% compared to linear AR models. This latter result is confirmed by the Diebold–Mariano pairwise test statistic.