Algorithms put to test: Control of algorithms in securities trading through mandatory market simulations?

Abstract

In the European Union, the algorithmic trading regime introduced by MiFID II requires firms to test and simulate their algorithms before deploying them on the market. While there have been important contributions to the overall scope of the EU algorithmic trading regime, the new testing requirements have remained largely untested by scholars. Against this background, the first half of this paper examines the algo testing framework from a legal perspective. This includes the way algorithms must be tested, the testing environments, the responsibilities of the market participants and the role of the supervisory authorities. It is showed that although being generally quite prescriptive, the crucial elements for the control of algorithms remain too vague. Subsequently, the paper seeks to evaluate whether testing can help to control algorithms in securities trading from a regulatory policy perspective. To do this, three issues are explored: firstly, the possibility of firms to choose the testing process themselves (including in-house testing), leading to questions regarding self-regulation; secondly, the overall effectiveness of the testing regime, in light of the interactions algorithms have with the market; finally, implications of the use of algorithmic trading systems based on AI or ML. The central findings suggest that clearer rules regarding (self-)testing are desirable and that only network-sensitive market simulations can help to control algorithms in securities trading.