Three essays about the fuel prices at the pump in Brazil (gasoline, ethanol, diesel, and Liquefied Petroleum Gas): (univariate and multivariate) forecasting and volatility

Loading...
Thumbnail Image

Journal Title

Journal ISSN

Volume Title

Publisher

Universidade Federal de Juiz de Fora (UFJF)

Abstract

This thesis consists of three essays that investigate retail fuel prices in Brazil, focusing on forecasting and volatility dynamics of gasoline and diesel prices. Given the importance of fuel prices for inflation, economic activity, and public policy, the essays employ different econometric and statistical approaches to provide new evidence on the behavior of the Brazilian fuel market. The first essay evaluates the forecasting performance of the non-parametric Singular Spectrum Analysis (SSA) method for retail gasoline and diesel prices across Brazil’s 27 states. The predictive accuracy of SSA is compared with ARIMA, Exponential Smoothing, and Neural Networks using monthly data from July 2001 to August 2023. The results show that SSA achieved the best in-sample performance, whereas ARIMA outperformed the benchmark models in the out-of-sample analysis. However, SSA provided complementary forecasting information for long-term horizons. The second essay examines the transmission of volatility from international crude oil prices to domestic gasoline and diesel prices using a multivariate GARCH model with the BEKK specification. The results indicate that gasoline prices are more responsive to international oil price shocks than diesel prices. Moreover, volatility spillovers intensified during the Import Parity Pricing (IPP) policy, when domestic fuel prices became more closely linked to international oil markets. The third essay introduces the Envelope Matrix Autoregressive (EMAR) model as an alternative approach for forecasting retail fuel prices in Brazil. By incorporating envelope dimension reduction, EMAR jointly models multiple fuel types and states while estimating fewer parameters than conventional multivariate models. Its forecasting performance is comparable to that of the Matrix Autoregressive (MAR) model, providing a more parsimonious specification. Overall, the thesis contributes to the literature by advancing forecasting methodologies and improving the understanding of retail fuel price dynamics and volatility transmission in Brazil, offering insights for researchers, policymakers, and market participants.

Description

Citation

Endorsement

Review

Supplemented By

Referenced By

Creative Commons license

Except where otherwise noted, this item's license is described as Acesso Aberto