Brazilian Free-Energy Market Mid-and Long-Term Forecasting Using Multi-Source Ensemble Solution

With its vast territory and diverse population distribution, Brazil is undergoing significant developments in its energy sector as it transitions towards a liberalized market. By 2030, the country aims to open the market to all consumers, enhancing competition, and liquidity in energy contracts, and fostering a dynamic trading environment. Effective market price monitoring is essential for understanding its dynamics and guiding decision-making processes in engineering applications. This paper aims to predict medium and long-term price movements in the free energy market. Our analysis of market price curves for different time horizons offers insights into energy market dynamics, enabling energy companies to optimize portfolio strategies and mitigate risks effectively. We consider forecast horizons of one quarter and one semester ahead, providing valuable insights for stakeholders navigating Brazil's evolving energy landscape. We utilize a combination of Machine Learning (ML)-based forecasters enriched by a broad set of exogenous data sources to provide forecasts for energy contracts whose effective date is Y2, Y3, and Y4 years ahead. In this study, we explore a simple yet effective and explainable alternative to using all the exogenous time series directly as input for our models. Namely, multiple related time series are aggregated into a series of histograms. Our proposed ensemble outperforms traditional statistical-based forecasting models, other ML solutions, and strong baselines in terms of resulting financial returns, particularly in the one-quarter prediction horizon, while also achieving competitive results for one semester ahead.