Análise multifractal do mercado brasileiro de eletricidade

Abstract

In the Brazilian wholesale electricity market, long-term contract prices are established through negotiations between energy generators and marketers with medium and large consumers. Unlike conventional markets, price determination is not directly governed by market forces but by complex computational models known as Hydrothermal Dispatch Optimization Models. In addition to the centralized operation dispatch programming, these models are responsible for calculating the Short-term Settlement Price (PLD), which serves as a reference for short-term market prices for electricity. The Brazilian market is divided into four interconnected submarkets: Southeast, Northeast, North, and South. This study significantly contributes to the existing literature by investigating the multifractality of markets where prices are not formed by the interaction between supply and demand but by computational models, employing Detrended Multifractal Fluctuation Analysis (MFDFA) on time series of seasonally adjusted price returns. The goal is to characterize the multifractal properties of electricity prices, identify the underlying causes of this multifractality, and examine market efficiency over time. Historical price analysis revealed that all submarkets exhibit anti-persistent behavior, also known as mean reversion, in addition to multifractality. This result is consistent with observations in international financial markets. Notably, the South submarket exhibited the highest level of multifractality and the lowest market efficiency, while the North submarket recorded the lowest levels of multifractality and higher efficiency. Using a sliding window technique, temporal variations in the Hurst exponent and Long-term Memory Magnitude, an indicator of market inefficiency, were explored. Consistent anti-persistent behavior was observed across all submarkets, with the South submarket showing greater volatility in its inefficiency index. These conclusions provide relevant information for decision-makers and regulators aiming to promote more efficient market arrangements. In addition to examining multifractality in the Brazilian electric sector, this research advances the application of multifractality theory to validate synthetic time series produced by predictive models. The study highlights the need for a standardized method to validate synthetic series generated by statistical models and neural networks. The proposed method uses the Hausdorff distance to contrast the multifractal spectra of the synthetic series, obtained via the MFDFA method, with the multifractal spectrum of the historical series. This methodology distinguishes which synthetic series exhibit multifractal attributes similar to those of the observed series, allowing the selection of predictive models that generate more accurate and reliable results.