Forecasting Naphtha Crack Based On Multiple Regression and System Dynamics

Naphtha prices directly depend on crude oil prices since naphtha is produced by refining crude oil. Naphtha plays an important role as one of basic petrochemicals for downstream products. Uncertainty in forecasting naphtha price has increased due to expanding price variability of naphtha affected by various factors. Therefore, forecasting naphtha crack (price difference between naphtha and crude oil) is a major requirement for decision making and planning. Several studies about price forecasting have been published. In this paper, we considered two methods: multiple regression model [1] and system dynamics model [2] for forecasting the naphtha price. This study is concerned with the derivation of a set of major parameters affecting naphtha prices and identification of the most dominating factors. Naphtha price depends mainly on Asia demand and supply of naphtha as well as naphtha substitute, margin, global economy and operational rate of oil company. The data for these factors are collected from March 2007 to November 2011 except from October 2008 to March 2009 to avoid unusual economic effect, Lehman brothers collapse. In multiple regression model based on statistical approach used data from March 2007 to December 2010 (training period). This model has been verified by comparing naphtha crack trend of actual naphtha crack and predicted naphtha crack from January 2011 to November 2011(forecasting period). In forecasting, the trend of naphtha crack is more important than the accurate value of naphtha crack. Therefore, we focus on the direction of predicted naphtha crack. Figure 1. shows actual naphtha crack and predicted naphtha crack in training period. From this period, R2  of this model is 91.8%. This means that 91.8% of the variations in naphtha crack can be explained by the major factors. Figure 2. presents actual naphtha crack and predicted naphtha crack forecasting period. The trend of naphtha crack is more important than precise naphtha crack.

Figure 1. Time series plot of actual naphtha crack and predicted naphtha crack for training.

(From March 2007 to December 2010) 

Figure 2. Time series plot of actual naphtha crack and predicted naphtha crack for forecasting.

(From January 2011 to November 2011)

Owing to 9 same predictions out of total 10 months in forecasting period, the percentage of correct predicted trend is 90%. Also, a model of forecasting naphtha crack is presented that is based on system dynamics. System dynamics is thinking model and simulation methodology. This model supports the study in complex systems. Therefore, this model is developed to support the changing petroleum markets. We draw a causal loop diagram for this model. This model forecasts naphtha crack and shows the relations among major factors. We presented two approaches for forecasting naphtha crack. Between two methods, we suggest higher percentage of correct predicted naphtha crack trend and the best approach in forecasting naphtha crack. The modeling approaches can be extended to forecast prices of other downstream chemicals such as LPG and NGL.

Keywords: Naphtha crack, Multiple regression, System dynamics, Forecasting

References

1.      W. Zhang, H. Chen, M. Wang, 2010, A forecast model of agricultural and livestock products price, Mechanics and materials 20-23, 1109-1114

2.      V. Karavezyris, K. Timpe, R. Marzi, 2002, Application of system dynamics and fuzzy logic to forecasting of municipal solid waste, Mathematics and Computers in Simulation, Vol. 60, 149-158

3.      Kim, J., Lee, Y., Moon, .I., 2008, Optimization of a hydrogen supply chain under demand uncertainty, International Journal of Hydrogen Energy, Vol.33, Issue 18, 4715-4729

4.      Kim, J., Moon, .I., 2008, Strategic design of hydrogen infrastructure considering cost and safety using multiobjective optimization, International Journal of Hydrogen Energy, Vol.33, Issue 21, 5887-5896