Intraday high-frequency pairs trading strategies for energy futures: evidence from China

ABSTRACT

We investigate the performance of pairs trading strategies based on Ornstein-Uhlenbeck (OU) process with jump-diffusion and regime-switching using minute-level data for five Chinese energy futures from 2 January 2020 to 30 November 2021 and compare them with traditional pairs trading strategies. Our results indicate that OU models can obtain an average return of 50.62% per annum and a Sharpe ratio of 2.63, which significantly exceed those of traditional pairs trading strategies. However, none of them could ‘win’ in every subperiod with diverse market conditions. Meanwhile, we find that introducing jump-diffusion indeed improves the performance (additional 25.37% annualized return and 1.12 Sharpe ratio). In contrast, considering more regimes does not always bring additional benefits. Robustness checks show that the superior performance of three-regime switching OU model (3RS-OUM) persists even under harsh trading conditions.

KEYWORDS:

• Chinese energy futures
• pairs trading
• Markov regime switching
• Ornstein-Uhlenbeck process
• high-frequency trading

JEL CLASSIFICATION:

• C1
• C3
• C6
• C63

Disclosure statement

No potential conflict of interest was reported by the authors.

Data availability statement

The data presented in the paper are downloaded from Wind Financial Terminal (www.wind.com.cn), and the access to the data is subject to third-party restrictions.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/00036846.2022.2161993

Notes

¹ The night trading hours for FU, ZC and JM futures are 21:00-23:00, for J 21:00-23:30, and for SC 21:00-02:30 the next day.

² If we had more options available, our pairs trading strategies would be more profitable.

³ in order to avoid data snooping bias, the formation period and trading period of the back-testing were determined before the study and would not change with data changes(Lo and MacKinlay 1990).

⁴ In order to reduce the impact of overnight trading data on the Bollinger band, this paper arbitrarily sets a trading adaptation time of 20 minutes. The robustness analysis of this parameter shows that if the time is set to 10 minutes or 30 minute, the results of each strategy in this paper are not significantly affected. Of course, the performance of the models would be significantly affected if the trading restrictions were completely removed. However, the t-statistic for 3RS-OUM and NRS-OUM are 3.44 and 3.63, indicating that the daily returns after transaction costs are statistically significant under no trading restrictions (related results are available from the authors upon request).