The European Union Emissions Trading System (EU ETS) serves as the primary market-based mechanism for driving the continent toward carbon neutrality. However, the market for EU Allowance (EUA) prices is characterized by significant historical volatility driven by regulatory shifts, geopolitical shocks, and complex energy dynamics. This inherent instability poses substantial challenges for industrial operators managing compliance costs and for regulators ensuring market credibility. This research addresses these challenges by developing a robust quantitative framework for the short-term forecasting of EUA prices to support effective risk management.

The study utilizes a comprehensive 10-year dataset of daily closing prices spanning from September 2015 to September 2025. To capture the "Energy Complex" influencing carbon demand, the analysis incorporates key energy drivers, including Natural Gas (TTF benchmark), Coal (Rotterdam API2), and Brent Crude Oil. To ensure econometric rigor, an Augmented Dickey-Fuller (ADF) test was applied to all variables, confirming that the raw price series were non-stationary. To address this, the study applied first-order differencing to achieve stationarity, focusing on daily returns rather than price levels. While cointegration and Vector Error Correction Models (VECM) were considered, a differenced Multivariate Vector Autoregression (VAR) and a Lasso Regression model were prioritized to better capture short-term dynamic interdependencies and mitigate the risks of model overfitting in high-dimensional datasets.

The evaluation protocol employs a one-day-ahead forecasting horizon based on a strict chronological out-of-sample backtesting procedure. The data is divided into an 80% training set (September 201 –2023) and a 20% test set (2024–September 2025). To reflect real-world trading conditions, the models were evaluated using an expanding window approach, ensuring that all available historical information was utilized for each prediction. The models are compared against a "Naive" Baseline, which assumes the price for T+1 is equal to the price at T, a rigorous benchmark in efficient financial markets.

The findings demonstrate that both the VAR and Lasso models significantly outperform the naive baseline in terms of Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE), confirming that the carbon market contains valuable predictive signals rather than following a random walk. Notably, the Lasso model’s feature selection capabilities identify natural gas as the most influential driver of carbon price formation, validating the critical role of fuel-switching dynamics in the European power sector. By providing a replicable methodology for isolating volatility drivers, this research offers essential tools for capital allocation and policy monitoring.