This dissertation analyzes an index representing the price of carbon traded in the European Union Emissions Trading System (EU ETS). By applying a variety of models, traditional econometric approaches (ARIMA), Machine Learning (CatBoost and Random Forest), and Deep Learning techniques (LSTM) were explored. The study utilized a comprehensive set of variables, encompassing traditional economic and financial indicators, as well as alternatives measures related to political, economic, and policy uncertainty. To avoid the risk of overfitting and to improve variable selection, the LASSO regularization technique was applied. In addition to selecting variables to reduce dimensionality, LASSO provided insights into the factors influencing carbon price formation. Among the selected uncertainty variables, the UK Economic Policy Uncertainty and the Climate Transition Risk Index (a proxy for perceived climate policy transition risk) stood out, showing relevance in explaining the dynamics of the SandP Carbon Credit EUA Index. Variable selection via LASSO yielded significant performance gains in out-of-sample tests, reducing overfitting and enhancing the models generalization capabilities. The consistency of the results was confirmed through time series adequate cross-validation and the Diebold-Mariano test, which verified whether there was a statistically significant difference between the performance of the models. The findings highlight the potential of alternative uncertainty indicators and machine learning methods for forecasting environmental asset prices, showing superior predictive performance in several key validation settings compared to the univariate ARIMA model under the metrics, tests, and validation strategies used.