Dynamic Adjustment of Crude Oil Price Spreads.

• Author: Ghoshray, Atanu

1. INTRODUCTION

   Standardised crude oil contracts are traded at the major exchanges such as the New York Mercantile Exchange (NYMEX) and Intercontinental Exchange (ICE) Futures Europe in London. Additionally, a large over-the-counter (OTC) market makes contracts with non-standard terms available to suppliers, producers and financial institutions. The most [w.sub.i]dely traded crudes are WTI (40[degrees]), Brent (38[degrees]) and Dubai/Oman (1) (32[degrees]); they are considered benchmarks and are often used to hedge against price fluctuations of crudes for which there is no liquid market. The effectiveness of such hedging strategies will depend on the behaviour of the price differential between the utilised benchmark crude and the crude of interest, which often possess different characteristics.

   The quality of a crude oil stream is defined by its natural properties, namely sulphur content and API gravity, (2) which determine the usefulness of the crude oil in the production of valuable petroleum products. Light crudes, with a gravity greater than 38[degrees], are usually sweet (low in sulphur) and produce a higher proportion of gasoline and diesel after the refining process. They are bench-marked to WTI (40[degrees]) (3) and Brent (38[degrees]) which are believed to reflect the market conditions in North America and North Sea respectively. Medium crude oils have a gravity between 22[degrees] and 38[degrees] and are benchmarked to Dubai/Oman (32[degrees]). These crude oils are more difficult to separate into chemical elements and yield a lower percentage of valuable products such as diesel. Heavy crude oils, which are benchmarked to Maya (22[degrees]), have a gravity of 22[degrees] or less, and are usually sour (high in sulphur). The extraction of these crude oils is becoming commercially viable due to the recent developments in technology, however their refinery presents special challenges and not all refineries are equipped to work with them.

   Substitution of a lighter crude with a heavier crude may be associated with high investment in the adaptation of refineries and the possible setting up of new transportation routes or storage; accordingly, lighter crudes are typically traded at a premium to heavier crudes. However, the relative prices of crude oils reflect more than just their natural properties and the equilibrium price differential is not always maintained. For instance, WTI (40[degrees]) usually trades at a premium to Brent (38[degrees]), however the WTI (40[degrees])-Brent (38[degrees]) differential turned negative in April 2007 due to an instant reduction in the refining capacity around Cushing where WTI is stored (Bloomberg, 2007). (4) Similarly, differentials between crudes can widen or narrow if crude oils respond differently to geopolitical events.

   To date, most of the evidence on the dynamic behaviour of crude oil spreads has been obtained using the differentials between Brent (38[degrees]), WTI (40[degrees]) and Dubai/Oman (32[degrees]). The results were sometimes used to approximate the behaviour of spreads for non-benchmark crudes on the assumption that they follow the lead of the light/medium benchmarks. This assumption may not hold. Over the recent years the range of traded crudes expanded rapidly-the 2010 edition of the International Crude Oil Handbook describes over 200 types of crude oil in the global markets ranging from very light to very heavy, an increase from 160 types of crudes recorded in the 2006 edition. Price differentials between benchmarks inevitably reflect their unique characteristics and it is possible that they are not representative of differentials between other crudes. If this is the case, then the results obtained using the benchmark pairs may be less useful for understanding behaviour of spreads than previously thought.

   The key contribution of this paper is to offer an analysis of short run and long run dynamic adjustment of differentials between a range of benchmark and non-benchmark crudes of different quality, in pursuit of the following objectives: to qualify whether the behaviour of the benchmarks is representative of non-benchmark crudes; to confirm whether or not the asymmetry found in the price adjustment of benchmark crudes holds for other crudes; and to establish whether observed quality characteristics of crudes affect the path of adjustment of crude oil spreads. We tackle these objectives using the Momentum-Threshold Autoregressive (M-TAR) model due to Enders and Granger (1998) along with its more powerful version, the Generalised Least Squares Momentum-Threshold Autoregressive (GLS-M-TAR) model proposed by Cook (2004). The evidence presented in this paper shows that using benchmarks as a proxy for other crude oils in economic modelling and forecasting can lead to omission of vital information about the oil markets and may provide biased outcomes.

   This paper is organised as follows: the review of the key literature is presented in Section 2; a description of the dataset is presented in Section 3; the econometric methodology is explained in Section 4; the results are reported in Section 5; and finally Section 6 concludes.

2. LITERATURE REVIEW

   The first attempt to formally analyse the relationship between crude oil prices was undertaken by Weiner (1991), after Adelman (1984) proposed that the world oil market is "like an ocean-one great pool." Weiner (1991) employed a correlation approach and a switching regression to establish whether prices for seven major crude oils traded in different geographic locations tend to move together, which would imply a unified market. Weiner (1991) finds strong evidence for regionalisation using monthly prices over the period 1980 to 1987, suggesting that any demand/supply shocks affecting a particular region are largely contained [w.sub.i]thin that region. However, subsequent studies have not confirmed these results.

   Sauer (1994) criticised the correlation method for testing market integration. He argued that correlation only allows bivariate comparisons and does not separate the effect of exogenous shocks, neither does it consider lagged responses. Using Johansen's (1991) cointegration method, Sauer (1994) finds strong evidence of unified markets, and suggested that the difference between his and Weiner's (1991) results are most likely explained by the difference in the time allowed for oil price adjustment.

   Gulen (1997) and (1999) conducted bivariate and multivariate cointegration analysis of eleven crude oils based on monthly and weekly data respectively. Cointegration analysis provided evidence for a unified market between the crude oils of both similar and different quality. This was demonstrated for both monthly and weekly data. Kleit (2001) used Gulen's (1999) dataset of weekly prices for a slightly longer time period and achieved similar results with the inclusion of an exogenous break in 1993. Kleit (2001) proceeded to estimate arbitrage costs, proposing that transaction costs deter arbitrage by eroding what can be gained from it. Ewing and Harter (2000) use the Engle-Granger (1987) method of cointegration to provide evidence of integration of Alaska North Slope (ANS, 30[degrees]) and Brent (38[degrees]) markets between 1974 and 1996. Bachmeier and Griffin (2006) estimated a Vector Error Correction Model (VECM) on daily data to test the degree of market integration both [w.sub.i]thin and between crude oil, coal, and natural gas markets. They find that the following crudes were traded in a highly integrated market between 1998 and 2004: Brent (38[degrees]), WTI (40[degrees]), ANS (30[degrees]), Dubai (32[degrees]) and Arun (54[degrees]). Bentzen (2007) used the VECM to analyse WTI (40[degrees]), Brent (38[degrees]) and OPEC price between 1988 and 2004. He found evidence of a unified market and bidirectional relationship between the three prices, suggesting that WTI (40[degrees]) and Brent (38[degrees]) are influenced by the OPEC price.

   Fattouh (2010) used Caner and Hansen's (2001) Threshold Autoregressive (TAR) model on daily price data for 7 crudes between 1997 and 2008. The study examined the behaviour of price differentials in two regimes. In one regime the price differential does not exceed transaction costs, and in the other regime the price differential does exceed transactions costs. The results from Fattouh (2010) confirm that mean reversion is only present where the price differential exceeds the cost of carry. Further, the findings show that price differentials are stationary over time for liquid crude oils with similar properties and that the threshold effects are only significant for crudes with quality differentials. We note that Caner and Hansen's (2001) TAR model implicitly considers that transaction costs are constant. However, this cost varies for each importer and is largely dependent on whether oil is transported by sea or land. As a result for the oil market where transaction costs are highly variable, the TAR model which implicitly assumes constant transaction costs, may be restrictive.

   In the paper central to our study, Hammoudeh et al. (2008) use the Enders and Siklos (2001) cointegration with M-TAR adjustment to test benchmark crude oil prices. Their results show strong evidence for asymmetry in the adjustment of price differentials between Brent (38[degrees]), WTI (40[degrees]), Dubai (32[degrees]) and Maya (22[degrees]). They further find that the WTI (40[degrees])-Dubai (32[degrees]) and WTI (40[degrees])-Brent (38[degrees]) differentials widen faster than they contract; the opposite is the case for the other two pairs, Brent (38[degrees])-Dubai (32[degrees]) and Brent (38[degrees])-Maya (22[degrees]). Additionally, Hammoudeh et al. (2008) estimate an ECM to analyse the short-run dynamics and find Granger-causality in all four pairs. The same method, Enders-Siklos (2001) and in addition Hansen-Seo (2002) have been used by Hammoudeh et al. (2010) to establish the asymmetries...