Modeling the United States crack spread: Market efficiency, persistence and the Verleger hypothesis

ABSTRACT

The price formation of crude oil and its refined products plays an essential role in the global economic system and mainly in the United States, where any shock on this market has implications for the different concerned parties. In this sense, we employ the fractionally cointegrated vector autoregressive model to analyze the long-run relationship between crude oil and each refined product and the persistence of the error term resulting, i.e., the crack spread, simultaneously. Once the cointegrating relationships between crude oil price and each refined product price are tested, we also evidence that the order of integration of the crack spread displays a long memory process. Finally, by attending to the coefficient adjustments, supply-driven market integration is given. Additionally, the Verleger hypothesis is rejected for all refined products, corroborated by the component share. This paper has important policy implications for investors, energy policymakers and refiners.

KEYWORDS:

• Crack spread
• persistence
• Verleger hypothesis
• fractional cointegration

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ WTI are the initials of the West Texas Index that picks up most of the national oil market.

² If $\beta$is equal to one, the error term can be considered as the crack spread (1:1:0 and 1:0:1 types).

³. Our model is specified with and without the presence of the unrestricted constant. The reached estimates are reasonably the same heedlessly the presence or lack of the unrestricted constant. Results are available upon request.

⁴ When the presence of unit roots are tested, the estimations were obtained using first-differenced data because the first series might be higher than 0.5. This test requires that the results are limited to the range −0.5 < d< 0.5 and then add 1 to achieve the proper estimates of d.

⁵. The whole estimates are available upon request to the authors.

⁶. This may be explained because propane production comes from two sectors: on the one hand, crude oil refining and, on the other hand, extraction and processing of natural gas, so propane prices are also affected by natural gas (Martínez, Abadie, and Fernández-Macho 2018).

⁷. According to our method, d represents the fractional order of integration of the explanatory variables, and b corresponds to the cointegrating error.

⁸. Since the relationship for the spread between WTI and Heating oil, Diesel, Kerosene and Propane, respectively, is not (1, −1), they could not be considered as the crack spread.

⁹. We have also continued with the analysis for the rest of the refined products using the CVAR model. In this case, the error term for the rest of the refined products is stationary for each pair of combinations.

¹⁰. The Verleger hypothesis is rejected by using both the FCVAR model for Conventional Gasoline (NY) and RBOB regular gasoline and CVAR model for the rest of the refined products.