Algorithmic quoting, trading, and market quality in agricultural commodity futures markets

ABSTRACT

This paper investigates the effect of algorithmic trading activity, as measured by quoting, on the corn, soybean, and live cattle commodity futures market quality. Using the CME’s limit-order-book data and a heteroscedasticity-based identification approach, we find more intensive algorithmic quoting (AQ) is beneficial to multiple dimensions of market quality. AQ improves pricing efficiency and mitigates short-term volatility, but its effects on liquidity costs are somewhat mixed. Increased AQ significantly narrows effective spreads in the corn and soybean markets, but not in the less traded live cattle futures market. The narrowing in effective spreads emerges from a reduction in adverse selection costs as more informed traders lose their market advantage. There also is evidence that liquidity provider revenues increase with heightened AQ activity in the corn futures market, albeit the effect is not statistically significant in the soybean and live cattle futures markets. The increased revenue points to a tradeoff between the dimensions of market quality, and the need for continued assessment and monitoring of algorithmic trading activity.

KEYWORDS:

• Algorithmic trading
• commodity futures
• market quality
• identification through heteroscedasticity

JEL CLASSIFICATION:

• G13
• Q02
• Q14

Acknowledgments

This material is based upon work that is supported by the National Institute of Food and Agriculture, U.S. Department of Agriculture, Hatch under accession number 1019569, and the Office of Futures and Options Research (OFOR) at UIUC.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ Statement by Chairman Timothy Massad regarding the approval of supplement proposal to automated trading regulation on 4 November 2016. https://www.cftc.gov/PressRoom/SpeechesTestimony/massadstatement110416

² https://www.cftc.gov/PressRoom/PressReleases/7865-19

³ Prior to November 2015, the CME Globex FIX format was used. Compared to FIX format, the MDP data provide nanosecond resolution and more accurate trade information. These improvements help in reducing measurement errors when computing market quality measures, particularly for order-execution cost measures.

⁴ The Lee and Ready (1991) rule cannot be used for inferring trade directions for implied trades as they are initiated in the spread book. Additionally, realized spreads and adverse selection costs can be miscalculated as liquidity provider revenues and losses are not realized in the outright book.

⁵ Hendershott, Jones, and Menkveld (2011) interpret AQ as a proxy for the level of AT in the market. They argue this measure essentially captures changes in liquidity supply caused by trading algorithms. The measure $A{Q}_t$is similar to the message-to-fill ratio used in the CME’s Messaging Efficiency Program for inferring the degree of algorithmic quoting’s impact.

⁶ The trading hours were: 9:05 to 16:00 on Monday, 8:00 to 16:00 on Tuesday-Thursday, and 8:00 to 13:55 on Friday.

⁷ Agency algorithms are widely used by ‘buy-side’ institutions to minimize trading costs when executing large orders for portfolio rebalancing (O’Hara 2015). Since numerous positions need to be liquidated during the short rolling window (typically 5 days), commodity index funds managed by major financial institutions depend on automated trading algorithms rather than manual trades.

⁸ Corn and soybean futures were added to the CME’s messaging efficient program before the sample period. We have tried using the messaging efficiency program as an instrument for identifying the effects of AQ on market quality measures in the live cattle market. However, Stock-Yogo test results suggest the messaging efficient program is a weak instrument. Intuitively, the market efficiency program provides an upper limit to AQ rather than causing a dramatic reduction in the level of AQ.

⁹ As shown in Lewbel (2012), the selection of variables only affects the efficiency but not the consistency.

¹⁰ These results are available upon request.

¹¹ These results are available upon request.

¹² See https://www.reuters.com/article/usa-cattle-markets-idUSL2N15003S for an example.

¹³ These results are available upon request.

¹⁴ Corn and soybean futures use a split FIFO (first in, first out)/Pro-Rata based matching algorithm, while the live cattle market only uses the FIFO. The FIFO algorithm only uses time and the Pro-Rata algorithm only uses order size to determine the priority for orders at the same price. Under a split FIFO/Pro-Rata algorithm, when large orders are submitted to the LOB, a certain percentage of each matching order gets allocated FIFO and the remainder is allocated Pro-Rata. By contacting with officials in the CME group, we were informed that about 70% – 80% of the time, FIFO is used to determine order priority in crop futures markets.

¹⁵ For example https://www.rcmalternatives.com/services/futures-traders-hedgers-commercials/ag-hedging-services/.

¹⁶ These results are available upon request.

Additional information

Funding

This work was supported by the the National Institute of Food and Agriculture, U.S. Department of Agriculture [1019569] and the Office of Futures and Options Research (OFOR) at the University of Illinois at Urbana-Champaig.