A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science, School of Engineering and Technology

Thesis (M.Sc.) - Asian Institute of Technology, 1984

Recent studies into the petroleum marketing system have been largely structural in nature, applying various econometric methods in the model formulation process. Their scope has been confined to the European market and to analyze the petroleum products on an aggregated basis. This thesis is concerned with the application of the Box -Jenkins time series analysis methodology to the development of a trading model for short-term forecasting of weekly Fuel Oil prices on the Singapore spot market . A univariate Autoregressive-Integrated-Moving-Average (ARIMA) model was first formulated using data of its own price series. This was then extended to include a postulated causal element by the development of a transfer-function-noise model with Arabian Light spot prices as a leading indicator. An empirical analysis of the out-of-sample one-week-ahead to twelve-weeks-ahead forecasts errors indicated that the univariate model has better predictive power than the transfer-function noise model. The mean absolute· percentage error varies between 0.6-1.6% for the univariate model a nd between 0.9-2.9% for the transfer-function-noise model. This thesis has shown that the Fuel Oil spot market is more sensitive to its own price changes with a shorter "memory" as compared to the Arabian Light spot market. The effect of the variation in the Arabian Light spot prices on the Fuel Oil spot prices is also relatively small. Both the weekly Fuel Oil and Arabian Light spot prices exhibit strong seasonality with a seasonal period of 5 and 4 weeks respectively.