Modelling an energy market with Bayesian networks for non-normal data

Energy markets are typically characterized by high complexity due to several reasons such as the large number of occurring variables, different in nature, and their associative structure. Estimating a statistical model that properly represents the dependencies among the variables is crucial for managing such a complexity. In this paper, a simple energy market influenced by hydroelectric availability is studied by using Bayesian networks. Since the variables of interest are quantitative but non Gaussian, non-parametric strategies are used to infer the Colombian energy market association structure. We propose a comparison between the UniNet learning algorithm and the Rank PC algorithm, both based on normal copula assumption and Spearman correlation measure, in order to explore differences in the estimated models. Finally, model usability for energy managers is shown through the discussion of some scenarios.