Praveen Kumar, Debashish Jena, Shrawan Kumar Trivedi, Ramakrushna Padhy and Abhijit Deb Roy
Oil refiners and Oil Marketing Companies, as major participants in the oil industry, are exposed to the uncertainties of crack spread volatility. This phenomenon is related to the price differential between crude oil and finished refined goods. The procurement managers prefer stability in the prices of crack spread to take procurements decisions. Whereas the other participants in the oil market, the hedge funds, owing to the speculative nature of their involvement, thrive in the volatility of crack spreads. While research on projecting oil prices is extensive, there is little research on anticipating crack spreads. The prevalent studies are mostly centred on modelling short term crack forecasts with a horizon of only few days ahead in order to cater to requirements of hedge funds managers and traders. The crack spread is directly reflective of the profitability of oil refineries, in order to improve the supply chain planning of oil refineries, which need crack spread forecast of 1-3 months’ time horizon, this study proposes crack spread forecasting model by using a hybrid technique of Deep learning and employing Bayesian optimization to deliver better predictive accuracies for crack price forecast. The suggested model is compared to previous hybrid models that include both linear and non-linear models as well as classic statistical models (ARIMA / ETS). The model’s performance is evaluated and compared using various known metrics such as MAPE, MSE, RMSE and is found to be superior. The improved crack spread forecasts so available will be enable Procurement managers to select and configure the most price efficient crude basket leading to increased profitability.