Agricultural commodity futures trading based on cross-country rolling quantile return signals

Abstract

This paper formulates and examines a new type of bivariate time series trading strategy based on signals generated from cross-country quantiles of return distributions. We conduct rolling quantile trading strategies separately in the U.S. and Chinese futures markets for soybeans, wheat, corn and sugar over very short (daily, intraday and overnight) holding periods. Overall, we find that these practical strategies outperform various benchmarks and there is a large profit potential when trades follow quantile-based signals rather than focusing on the median only. The results highlight the value of cross-country trading strategies and the harnessing of information from different parts of the return distributions which have so far been neglected.

Keywords:

• Quantiles
• Trading strategies
• Spillovers
• Profitability
• Agricultural market

JEL Classification:

• C21
• C61
• G11
• G14

Acknowledgements

We would like to thank Robert Bianchi, Joëlle Miffre, Stefan Trueck, Graham Bornholt, John Hua Fan, Akihiro Omura, Liang-cheng Zhang, Charles Hyde, and Andrew Kaleel for their constructive and insightful comments, and acknowledge the comments from participants of the Griffith Alternative Investments Conference 2016, and the 1st Australasian Commodities Markets Conference 2017.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

† These trading strategies are also shown to work well in equity markets (Hong and Satchell 2015, Marshall et al. 2017, Packham et al. 2017).

† Spillover results from Jiang et al. (2016) in relation to the trading strategy are detailed in the Appendix. Subsection 3.2, paragraph 1 and 2 summarizes the cross-quantilogram methodology and significant spillovers.

‡ Jiang et al. (2016) point out that the cross-country spillovers are much stronger than auto-spillovers in the U.S. and Chinese agricultural futures markets. Therefore, the rolling quantile trading strategy based on auto-spillovers is not included in this paper for brevity.

§ The term ‘abnormal return’ is used in the paper when the returns from the strategy are higher than three benchmarks for the same commodity during the same period.

† In China, soybean and corn futures contracts are listed on the Dalian Commodities Exchange, wheat and sugar are listed on the Zhengzhou Commodities Exchange. In the US, soybeans, wheat and corn are traded on the Chicago Board of Trade, whereas sugar is traded on the Inter Continental Exchange.

‡ The data source is proprietary. The terminal of the database is available for download on http://www.agdata.cn/client/downLoadPage.html. A description on the database can be found on http://www.agdata.cn/solution/index.html.

§ The decomposition formula is identical for each commodity. For simplicity, no subscript is used to denote the commodity type.

¶ It is demonstrated by Jiang et al. (2016) that cross-country return spillovers are most pronounced from the daily returns of country i to the 12-hour-ahead daily and overnight returns in country j. Building upon on the findings of Jiang et al. (2016), the rolling quantiles in this paper focus on daily return series.

† Additional details about the spillovers are given in the Appendix.

† A brief explanation for equation (5(5) $R_{CN,t+1}^{SR}=\left\{\begin{array}{ll}-R_{CN,t+1}^{IN}& \text{if }{R}_{US,t}^D>Q_{US,t}(1-\alpha ),\\ R_{CN,t+1}^{IN}& \text{if }{R}_{US,t}^D<Q_{US,t}\left(\alpha \right),\\ 0& \text{if }{Q}_{US,t}\left(\alpha \right)<R_{US,t}^D<Q_{US,t}(1-\alpha ).\end{array}\right.$(5) ) is that, when the U.S. daily return at day t is higher than its upper rolling quantile on day t, we take the short position in the corresponding market in China and hold it for one day on day t+1 to receive the return of $-R_{CN,t+1}^{IN}$. When the U.S. daily return at day t is lower than the its lower rolling quantile on day t, we take the long position in the Chinese intraday market on day t+1 to receive the return of $R_{CN,t+1}^{IN}$. When the U.S. daily return at day t is between its upper and lower quantile pair, no trading action is taken and the strategy's return is 0.

† When calculating the sample length in years for four commodities, three decimal places are used for accuracy.

‡ Trades do not occur everyday using rolling quantile trading strategy when the quantile combinations are at the extremes and the common way of calculating the annualized mean return can be misleading, consequently, we divide the total after-cost return by the number of trading years to get the annual mean return for each of the four commodities.

§ We modified the momentum and roll-yield trading strategies in Fuertes et al. (2010, 2015) to adapt them to our time-series single commodity scenario as benchmark trading strategies. Specifically, while they deploy the strategies for a cross section of commodities using past performance or roll yield as sorting criteria, we deploy them on a single commodity basis using the same rolling windows as for the quantile based strategies. Moreover, we follow the literature and use a monthly holding period for the benchmark of momentum and roll-yield strategies, (e.g. Fuertes et al. 2010, 2015, Bianchi et al. 2015).

† Bootstrap replication of 5000 and $10000$ times present similar results as 1000 times.

‡ Window sizes of 125, 250, 375 and 500 days are used to calculate the rolling quantiles of daily returns for one country, this section only presents results from rolling quantiles of 125-day window. Trading strategies returns on rolling window of 250, 375 and 500 days show similar and consistent pattern to that of 125-day, these results are available upon request.

§ Similar results are found for 125-, 250-, 375- and 500-day rolling windows.

† The returns are higher in tables 8 and 9 compared to those of tables 5 and 6, as tables 8 and 9 present the after-cost total return calculated by equation (11(11) $R_{i,total}^{SR}=\sum _{k=1}^{T_i-{\tau }_i+1}{R}_{i,k}^{SR}-C_i\times N_i^{Tdays},$(11) ) without annualization, whereas tables 5 and 6 present the after-cost annul mean return calculated by equation (12(12) $R_{i,mean}^{SR}=\frac{\sum _{k=1}^{T_i-{\tau }_i+1}{R}_{i,k}^{SR}-C_i\times N_i^{Tdays}}{N^{Years}},$(12) ). Both column 3 of tables 8 and 9 and column 2 of tables 5 and 6 provide returns on the intraday market with a rolling window of 125 days. The connection between these two columns can be described as follows. For each quantile combination, the sum of the after-cost total return for each pair of consecutive rows in column 3 of table 8 divided by the sample length in years equals the after-cost annual mean return in column 2 of table 5. The same relationship applies between table 9 and 5.