Elena Asparouhova, Peter Bossaerts, Xiaoqin Cai, Kristian Rotaru, Nitin Yadav, Wenhao Yang
Review of Finance, Volume 28, Issue 4, July 2024, Pages 1215–1244, https://doi.org/10.1093/rof/rfae007
The study explores the impact of algorithmic trading on the emergence of bubbles in financial markets, utilizing a controlled experimental setting, and for the first time, allowing participants full control over the trading robots they choose. Previous experimental studies had shown mixed results, but participants were not in charge of the automated trading.
While algorithmic trading in field markets has been associated with improved price quality and liquidity, the exact impact remains debated, with some studies recording diminishing profits, particularly in high-frequency trading (HFT). The profitability of different algorithm types varies, with liquidity-taking algorithms potentially generating above-average profits but being less frequently chosen.
Traditionally, financial bubbles, characterized by overpricing of securities over multiple periods, have been observed frequently. The introduction of execution algorithms, or trading robots, offers participants a new tool alongside manual trading, akin to the strategy method in experiments on game theory. Even participants opting out of using robots must consider their potential presence, reinforcing the need for careful strategy formulation. The study therefore hypothesizes that algorithmic trading could reduce mispricing and improve overall pricing accuracy. Against this hypothesis is the argument that the commitment required in choosing and deploying a robot may hinder adaptability.
The experiment includes various treatments, differing in the degree of commitment to robot deployment, with participants randomly assigned to each treatment. One treatment where all trade had to be manual provides a basis against which to evaluate the impact of the availability of robots.
Surprisingly, the presence of robots does not lead to reduced bubble sizes or frequency compared to manual trading. Early rounds show higher bid-ask spreads and more frequent flash crashes/surges. Participants deploying robots alongside manual trading earn more, indicating potential benefits of combining strategies. Participants heavily utilize robots for transactions, in proportions resembling those in advanced markets in the field. Notably though, a larger fraction of deployed robots is liquidity-taking, unlike in records from field markets. This can be attributed to the inability, in field markets, to identify the presence of liquidity takers when they end up not trading. Controlled experiments therefore lead to insights into market behavior and the use of trading tools that cannot be obtained in field studies.
Figure: Number of robot deployments, per session, by treatment (free robot use; full commitment once launched; compulsory launching of one robot), and by type (liquidity making – blue – or liquidity taking – yellow). Number of liquidity takers is much higher relative to liquidity makers compared to historical analysis of field transaction data.
