Trading WTI/BRENT Spread

The WTI-Brent spread is the difference between the prices of two types of crude oil, West Texas Intermediate (WTI) on the long side and Brent Crude (Brent) on the short side. The two oils differ only in the ability of WTI to produce slightly more gasoline in the cracking ratio which causes WTI’s slight pricing margin over Brent. As both oils are very similar their spread shows signs of strong predictability and usually oscillates around some average value. It is therefore possible to use deviations from the fair spread value to bet on convergence back to fair value. The fair spread value could be calculated via moving average, regression, neural network regression or other procedures. We present moving average calculation as an example trading strategy
from the source paper.

Fundamental reason

Both oils differ in chemical compositions and they differ also in production and transportation attributes. These differences are reflected in the price spread between both futures contracts. The spread is mean reverting because most of the price shocks are only temporal so the spread moves back to its long term economical equilibrium and therefore it is possible to create a trading strategy based on this mean reversion. Caution should be only needed in utilizing parameters from the source paper as they are based on the short history and therefore could be susceptible to data mining bias.

Markets traded
commodities
Confidence in anomaly's validity
Moderately strong
Notes to Confidence in anomaly's validity
while there is no doubt about spread mean reverting, caution is needed in evaluating the moving average period used for trading in the paper as the backtest and the optimization is based on a short data sample
Period of rebalancing
Daily
Notes to Period of rebalancing
Number of traded instruments
2
Notes to Number of traded instruments
Complexity evaluation
Simple strategy
Notes to Complexity evaluation
Financial instruments
futures
Backtest period from source paper
1995-2004
Indicative performance
39.40%
Notes to Indicative performance
per annum, calculated as weighted average of in sample and out of sample period, data from table 2
Estimated volatility
21.65%
Notes to Estimated volatility
worse number from in and out of sample period, data from table 2
Maximum drawdown
-13.45%
Notes to Maximum drawdown
worse number from in and out of sample period, data from table 2
Sharpe Ratio
1.64

Keywords:

pairs trading, arbitrage, spread trading

Simple trading strategy

A 20-day moving average of WTI/Brent spread is calculated each day. If the current spread value is above SMA 20 then we enter a short position in the spread on close (betting that the spread will decrease to fair value represented by SMA 20). The trade is closed at the close of the trading day when the spread crosses below fair value. If the current spread value is below SMA 20 then we enter a long position betting that the spread will increase and the trade is closed at the close of the trading day when the spread crosses above fair value.

Hedge for stocks during bear markets

Not known - Source and related research papers don't offer insight into correlation structure of trading strategy to equity market risk, therefore we do not know if this strategy can be used as a hedge/diversification during time of market crisis. Commodities usually have a negative correlation to equities therefore proposed strategy can be negatively correlated too, but rigorous backtest is needed to asses if this is the case ...

Source Paper

Evans, Dunis, Laws: Trading Futures Spread: An Application of Correlation
http://www.palgrave-journals.com/jdhf/journal/v15/n4/full/jdhf200924a.html
http://www.ljmu.ac.uk/AFE/AFE_docs/Ben1004.PDF
Abstract:
Original motivation for this paper is the investigation of a correlation filter to improve the risk/return performance of the trading models. Further motivation is to extend the trading of futures spreads past the “Fair Value” type of model used by Butterworth and Holmes (2003). The trading models tested are the following; the cointegration “fair value” approach, MACD, traditional regression techniques and Neural Network Regression. Also shown is the effectiveness of the two types of filter, a standard filter and a correlation filter on the trading rule returns. Our results show that the best model for trading the WTI-Brent spread is an ARMA model, which proved to be profitable, both in- and out-of-sample. This is shown by out-of-sample annualised returns of 34.94% for the standard and correlation filters alike (inclusive of transactions costs).

Other Papers

Lubnau: Spread trading strategies in the crude oil futures market
http://econstor.eu/bitstream/10419/96520/1/783913591.pdf
Abstract:
his article explores whether common technical trading strategies used in equity markets can be employed profitably in the markets for WTI and Brent crude oil. The strategies tested are Bollinger Bands, based on a mean-reverting hedge portfolio of WTI and Brent. The trading systems are tested with historical data from 1992 to 2013, representing 22 years of data and for various specifications. The hedge ratio for the crude oil portfolio is derived by using the Johansen procedure and a dynamic linear model with Kalman filtering. The significance of the results is evaluated with a bootstrap test in which randomly generated orders are employed. Results show that some setups of the system are able to be profitable over every five-year period tested. Furthermore they generate profits and Sharpe ratios that are significantly higher than those of randomly generated orders of approximately the same holding time. The best results with some Sharpe ratios in excess of three, are obtained when a dynamic linear model with Kalman filtering and maximum likelihood estimates of the unknown variance of the state equation is employed to constantly update the hedge ratio of the portfolio. The results indicate that the crude oil market may not be weak-form efficient.

Donninger: The Poverty of Academic Finance Research: Spread Trading Strategies in the Crude Oil Futures Market
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2617585
Abstract:
Harvey, Liu and Zhu argue that probably most of the Cross-Section of Returns literature is garbage. One can always try an additional factor and will find a significant Cross-Sectional result with enough trial and error. Lopez de Prado argues in a series of articles in a similar vein. Theoretically scientific results are falsifiable. Practically previous results and publications are checked only in rare occasions. Growth in a Time of Depth by Reinhart-Rogoff was the most influential economic paper in recent years. It was published in a top journal. Although the paper contained even trivial Excel-Bugs it took 3 years till the wrong results and the poor methodology was fully revealed. The reviewers did not check the simple spreadsheets. This paper analyzes a less prominent example about spread trading in the crude oil futures market by Thorben Lubnau. The author reports for his very simple strategy a long term Sharpe-Ratios above 3. It is shown that - like for Reinhart-Rogoff - one needs no sophisticated test statistics to falsify the results. The explanation is much simpler: The author has no clue of trading. He used the wrong data.

Hypothetical future performance

Real-time implementation by QuantConnect