Carbon price returns prediction using a hybrid model

In the current context of global warming, the prediction of carbon prices has acquired a prominent
role since carbon price constitutes a powerful tool in the operation of artificial carbon markets and
the design of mechanisms oriented to mitigate climate change. A major challenge for carbon price
forecasting is the modeling of non-linear effects in the time series, for which the use of hybrid
models seems to be an appealing alternative to explore. This paper studies the performance of a
hybrid model which weights the results from the exponential smoothing model, nonlinear
autoregressive neural network, and the autoregressive integrated moving average model. These
weights are determined by (i) assuming equal weights, (ii) cross validated errors, and (iii) using a
neural network to optimize the individual weights. The results confirm the importance of
modeling the non-linear effects of time series and the capacity of hybrid models in predicting
carbon prices.
Page