Predictive modeling of economic and financial monitoring of The Brazilian Electricity Regulatory Agency (ANEEL)

Abstract

According to the The Brazilian Electricity Regulatory Agency (ANEEL), the economic-financial indicators of the electricity distribution companies are fundamental to the work of supervising the management of electricity distribution in Brazil. The continuous monitoring is performed by ANEEL in accordance with Technical Note 111/2016 (June 29, 2016) in order to prevent the degradation of the regulated service and identify any issues in energy distribution administration. In this sense, a quarterly report namely Relatório de Indicadores de Sustentabilidade Econômico-Financeira das Distribuidoras comprises 11 indicators divided into 6 subareas debt, efficiency, investments, profitability, shareholder return and operating from 2011 to 2017. The objective of this paper is to quantify the indebtedness level of companies in the tensor structured data (company x indicator x year) provided by those reports. For this, tree-based, linear and polynomial regression models were fitted in which feature variables originated from dimensionality reduction methods, such as Principal Component Analysis and Autoencoder. Also, a Bayesian Structural Equation Model simultaneously promoting Confirmatory Factor Analysis (outer model) and incorporating linear relationships (inner model) between latent variables was fitted. The performance comparison of these methods was made from the predictive power in the validation set.