Análise de preços de commodities agrícolas brasileiras utilizando o método grafo de visibilidade horizontal

Abstract

Agricultural commodities are directly associated with agribusiness and exert a fundamental role on the Brazilian economy. These goods can be considered to be among the most important, as their price has a significant effect on the population's life. The objective of this work was to analyze the temporal price variation, and to compare the price dynamics of 11 Brazilian agricultural commodities with that of ethanol, whose price variation is directly related to sugar prices, and thus contribute to the development and validation of theoretical and computational tools for forecasting prices of that market. The daily price series were analyzed, as well as the series of returns and volatility recorded from January 2010 to December 2019. We used the Horizontal Visibility Graph (HVG) method, based on the complex network theory, which maps time series into graphs through a geometric visibility criterion that associates each time series data point to a node in the visibility graph. For each series, the Clustering Coefficient, the λ Coefficient (slope of the semi-logarithmic line of the probability distribution of the degree of the network nodes), and the Average Length of the Shortest Path was calculated, representing indexes used to describe the network topology. The results demonstrate that the chicken and pork price series generated less connected networks, and specifically the chicken price series generated a more integrated network. The cotton, chicken and wheat price series are governed by chaotic processes, indicating that their prices are less predictable in the analyzed period, while the other price series (9 out of 12), represent correlated stochastic processes. Both the return series and the volatility series correspond to networks with similar connection levels, except for chicken (a less connected network). Regarding the level of integration, the returns for live cattle and calf prices correspond to more integrated networks, and the volatility of chicken and coffee prices to less integrated netwoks. In addition, we observed that the series of soybean price returns appear to exhibit an uncorrelated series behavior. It was also found that half of the series of returns and price volatility are generated by chaotic processes. Thereby, the HVG proved to be efficient in studying the structure and classification of the processes that govern the financial series of Brazilian agricultural commodities.