Active Trading in ETFs: The Role of High-Frequency Algorithmic Trading

Abstract

In the study reported here, we explored high-frequency algorithmic trading and its effect on exchange-traded funds (ETFs). Using the cancel rate, the trade-to-order ratio, percentage odd-lot volume, and trade size as proxies for algorithmic trading, we found that more algorithmic trading in ETFs results in smaller and less persistent deviations of fund prices from their net asset values (NAVs). Arbitrage strategies adopted by algorithmic traders directly help reduce the magnitude and persistence of ETF price deviations from NAVs. Also, algorithmic trading improves ETF liquidity by lowering spreads and facilitates arbitrage.

Disclosure: The authors report no conflicts of interest.

Editor’s note:

Submitted 6 July 2020

Accepted 10 December 2020 by Stephen J. Brown

This article was externally reviewed using our double-blind peer-review process. When the article was accepted for publication, the authors thanked the reviewers in their acknowledgments. Marius Zoican and one anonymous reviewer were the reviewers for this article.

Acknowledgments

We thank Valerie Huang, Marc Lipson, Thomas McInish, Leonid Pugachev, Davide Tomio, Gulnara R. Zaynutdinova, and participants at the seminar at the University of Memphis, the seminar at Rochester Institute of Technology, the 2018 Eastern Finance Association meeting, and the 2018 Financial Management Association annual meeting for helpful comments.

Notes

¹ Algorithmic trading refers to the use of computer algorithms to automatically submit, update, and cancel orders (Hendershott, Jones, and Menkveld 2011). High-frequency algorithmic trading is a subset of algorithmic trading that involves the “use of extraordinarily high speed and sophisticated programs for generating, routing, and executing orders” (SEC 2010).

² An AP is an organization that has the right to create and redeem shares of an ETF. APs provide a large portion of the liquidity in the ETF market by obtaining the underlying assets required to create the shares of an ETF.

³ Similar to Weller (2018), we used the cancel rate, the ratio of trading volume to order volume, percentage odd-lot volume, and trade size as proxies for algorithmic trading. Higher odd lots and cancel-to-trade ratios indicate more algorithmic trading. On the contrary, higher trade-to-order volume ratios and larger average trade sizes indicate less algorithmic trading.

⁴ When the premium/discount based on midpoint prices was more than 10 percentage points greater in absolute terms than the premium/discount based on prices, we followed Broman (2016) and used prices instead of midpoint quotes to calculate the deviation.

⁵ We also computed autocorrelations of the NAV deviation (unreported). We found the equal-weighted daily autocorrelation to be only 0.23 across all funds and the average half-life of the deviations to be 0.47 day.

⁶ Order fragmentation as an AT proxy in regression analysis is discussed only for the material on robustness checking in the Supplemental Online Material.

⁷ In untabulated results, we also examined how many exchanges an ETF was traded on in a given day. We found that 94% of the ETFs were traded on at least seven exchanges on a given day.

⁸ Results based on actual data for the four proxies of algorithmic trading are reported as a robustness check and can be found in the Supplemental Online Material.

⁹ Because error terms for deviations and persistence of deviations at a given point in time may be correlated as a result of market sentiment and availability of arbitrage capital, we used time-clustered standard errors.

¹⁰ The results are qualitatively similar when we used other AT proxies, such as the trade-to-order ratio, percentage odd-lot volume, and trade size; these results are not reported here for brevity.

¹¹ For the Sobel tests, we used 1 – AT instead of AT as the independent variable because AT and spread have coefficients of opposite signs.