Performance of technical trading rules: evidence from the crude oil market

ABSTRACT

This study investigates the debatable success of technical trading rules, through the years, on the trending energy market of crude oil. In particular, the large universe of 7846 trading rules proposed by Sullivan, Timmermann, and White (1999. “Data-Snooping, Technical Trading Rule Performance, and the Bootstrap.” The Journal of Finance 54 (5): 1647–1691. doi:10.1111/0022-1082.00163), divided into five families (filter rules, moving averages, support and resistance rules, channel breakouts, and on-balance volume averages), is applied to the daily prices of West Texas Intermediate (WTI) light, sweet crude oil futures as well as the United States Oil (USO) fund, from 2006 onwards. We employ the k-familywise error rate (k-FWER) and false discovery rate (FDR) techniques proposed by Romano, J. P., and M. Wolf. (2007. “Control of Generalized Error Rates in Multiple Testing.” The Annals of Statistics 35 (4): 1378–1408. doi:10.1214/009053606000001622) and Bajgrowicz, P., and O. Scaillet. (2012. “Technical Trading Revisited: False Discoveries, Persistence Tests, and Transaction Costs.” Journal of Financial Economics 106 (3): 473–491. doi:10.1016/j.jfineco.2012.06.001) respectively, accounting for data snooping in order to identify significantly profitable trading strategies. Our findings explain that there is no persistent nature in rules performance, contrary to the in-sample outstanding results, although tiny profits can be achieved in some periods. Overall, our results seem to be in favor of interim market inefficiencies.

KEYWORDS:

• Crude oil
• technical trading
• data snooping
• transaction costs
• persistence
• market efficiency

JEL Classifications:

• C12
• C15
• G11
• G14

Acknowledgements

We would like to thank participants at the Econometrics and Financial Data Science (EFDS 2017, ICMA center) workshop Forecasting Financial Markets (FFM 2016), Quantitative Finance and Risk Analysis (QFRA 2016), and WFMA Xian Jiatong-Liverpool University conferences, and at the Universities of Liverpool and Glasgow, as well as Carol Alexander, Neil Kellard, Marcel Prokopczuk, Michalis Stamatogiannis, Andrea Roncoroni, Ehud Ronn, Ernest P. Chan, Chris Florackis, Vasileios Kallinterakis, and Michael A. H. Dempster for helpful comments. Any remaining errors are ours.

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

Athanasios A. Pantelous http://orcid.org/0000-0001-5738-1471

Notes

1 Earlier studies of technical analysis and patterns in stock returns include Alexander (1961, 1964), Fama (1965, 1970), Fama and Blume (1966), Levy (1967), James (1968), Jensen and Benington (1970), and Sweeney (1988).

2 We chose this specific period in order to examine the TTRs’ performance on the same trading days for crude oil futures and the USO, given that the inception date of the USO was in April 2006.

3 The final settlement date is the 4th U.S. business day prior to the 25th calendar day of the month preceding the contract month.

4 The USO periodically ‘rolls over’ its underlying futures contracts by selling those that are approaching expiration and buying those that expire farther into the future. The investment objective of the USO is publicized on its website (http://www.unitedstatesoilfund.com/).

5 For instance, in 2008, crude oil reached its highest value, followed by a fall below $50 per barrel due to the Lehman Brothers crisis in 2009.

6 Marshal et al. (2008) evaluate the performance of the Sullivan, Timmermann, and White (1999) universe of TTRs in 15 major commodities futures series, while considering naïve methods of accounting for data snooping effects. One of these series refers to crude oil futures covering the period 1984–2005.

7 The interested reader can refer to the appendix of Sullivan, Timmermann, and White (1999) for a detailed description of each rule as well as the extra parameters used.

8 Actually, following the studies of Brock, Lakonishok, and LeBaron (1992), Sullivan, Timmermann, and White (1999), and Bajgrowicz and Scaillet (2012) who implement the ‘double-or-out’ trading strategy, a buy signal leads a trader to borrow money at the ‘risk-free’ rate in order to double the investment in the commodity portfolio, a neutral signal leads to the trader simply holding the commodity, and when a sell signal occurs the trader liquidates and exits the market.

9 Developed by Young (1991), the Calmar ratio stands for California Managed Account Reports. It is a performance measurement used to evaluate commodity trading advisors and hedge funds.

10 We acknowledge Ernest P. Chan for pointing this out to us.

11 We use as a risk-free rate the daily effective federal funds rate, in accordance with all the previous literature.

12 Most of the trading rules employed in this study are designed to capture momentum. Their effectiveness is mainly based on the existence of significant autocorrelation of returns series.

13 A studentized test statistic refers to a simple test statistic divided by the consistent estimator of its standard deviation. This helps one to compare objects in the same units of standard deviation.

14 We do not apply the Calmar ratio criterion since its formulation is based on at least a couple years of previous data, while in our persistence analysis we use a rebalancing period of six months (see Section 5).

15 The initial FDR version of Benjamini and Hochberg (1995) adopted independence across multiple hypotheses. Later, studies by Benjamini and Yekutieli (2001), Storey (2002), and Storey, Taylor, and Siegmund (2004) proved that the FDR holds under ‘weak dependence’ conditions when the number of hypotheses is very large. Also, Bajgrowicz and Scaillet (2012) explain that the Sullivan, Timmermann, and White (1999) trading rules satisfy this feature, since the rules are dependent in small blocks (within the same family) and independent across different families.

16 The block length used is equal to q = 0.1, and the number of bootstrap realizations is set to B = 1000, following previous studies.

17 Bajgrowicz and Scaillet (2012) set the value of λ just by looking for the level above which the histogram of p-values becomes fairly flat, representing the region of null p-values. There is also an automated version of this process described by Storey (2002).

18 The FDR⁻ part can be calculated in a similar way.

19 The critical value $\widehat{c_1}$ asymptotically controls the k-FWER criterion. According to the theory $c_1=c_K\left(1-a,k,P\right).$ However, the set K and the probability mechanism P are unknown. Therefore, K is replaced by the set of all rules $\left\{1,\dots ,l\right\}$ and the probability measure $\widehat{P_T}$ of the bootstrapped distribution is used instead of P.

20 Positive performance means a mean return or Sharpe ratio above zero, or a Calmar ratio above one.

21 A Calmar ratio value of 1–2 is assumed a good strategy, a value between 2 and 5 very good, and a value greater than 5 recognized as excellent (Young 1991).

22 The relevant table and results of the persistence analysis considering the subperiods given by the nonparametric change point detection approach of Ross, Tasoulis, and Adams (2011) are presented in Appendix A.

23 We should mention that the number of rules chosen varies substantially from one six-month period to the next. Sometimes, the portfolio consists of almost exclusively new rules, even after the first rebalancing.

Additional information

Funding

The authors would like to acknowledge the gracious support for this work provided by the EPSRC and ESRC Centre for Doctoral Training in Quantification and Management of Risk & Uncertainty in Complex Systems & Environments [grant number EP/L015927/1].