Ethanol and gasoline price in Brazil: A long-run time series and panel approach
Kormazou, Christina- Eleni- Triada
School of Science and Technology, MSc in Energy Management
This dissertation was written as a part of the MSc in Energy Management at the International Hellenic University. The aim of the study is to investigate possible integration of hydrous ethanol and gasoline price in Brazil at state level for the period 2002-2015. Initially, the historical framework of the domestic ethanol market is described towards overall understanding of the topic. Then by employing a Vector Autoregressive (VAR) model of all 27 Brazilian states, Johansen cointegration test investigates possible cointegration of both price series at state level. Impulse Response Functions are implemented within the pairwise time series analysis. A Vector Error Correction Model (VECM) provides long-run elasticities of ethanol price with respect to gasoline price for the cointegrated states. Within a panel data context panel unit root tests examine stationarity at levels and first differences. Finally, panel cointegration is examined through Pedroni and Kao panel cointegration tests and FMOLS on aggregate level.
Regarding results, Johansen test provides evidence of co-movement between the price series in 10 out of the 27 Brazilian states. The long-run elasticities of ethanol price with respect to gasoline price for the cointegrated states range from 1.07% to 1.66% providing an average value of 1.33%. Impulse response functions reveal a relatively overall higher response of log_price_ethanol to a unitary shock in log_price_gasoline than the reverse case, with a levelling-off process within the twelve-month horizon of 2016, except for the case of Tocantins. Regarding panel data analysis, IPS and ADF Fisher panel unit root tests prove that both price series are I(1) at all states and I(0) when first differenced again at all states. Pedroni and Kao panel cointegration test output is if favor of panel cointegration. Finally, FMOLS provide an aggregate long-run elasticity estimate of 1.30 that is almost identical to the value provided by the time series approach.