Introduction

The introduction of MiFID II in 2018 brought about a new set of rules for firms utilizing algorithms to price and trade. In addition to establishing a risk management process, the regulation required thorough documentation, testing, and sign-off procedures for the introduction of new algorithms or any changes to existing ones. The goal was to ensure that firms exercised due care in the design, construction, testing, and maintenance of their algorithms, aiming to prevent the flash crashes that had plagued the preceding decades.

However, despite these regulations, incidents involving malfunctioning algorithms continued, such as the May 2022 flash crash triggered by one of Citi’s equity execution algorithms. While this led to a $13.9 million fine from BaFin, the European Securities and Markets Authority (ESMA) remained concerned that either controls were inadequate or enforcement insufficient, prompting a Common Supervisory Action in 2023 to investigate further.

To delve deeper into this issue, GreenBirch Group conducted a survey of banks in the first half of 2024 to assess the impact of these regulations on firms using algorithms. This white paper outlines our findings and highlights areas where best practices are making a tangible difference.

Scope of Regulation

What is an Algorithm?

MiFID II Article 4(39) defines algorithmic trading as “trading in financial instruments where a computer algorithm automatically determines individual parameters of orders, such as whether to initiate the order, the timing, price, or quantity of the order, or how to manage the order after its submission, with limited or no human intervention.”

One of the first ambiguities that emerged was the interpretation of “limited” human intervention. Without clear guidance, particularly concerning price quoting in response to client RFQs, firms that involved a trader between the pricing system and the market in which the price was being submitted often avoided classification as engaging in ‘algorithmic trading’ under MiFID II. A legitimate question arises as to how effective a trader can be in identifying a rogue price for a complex derivative compression trade before the computer-generated price reaches the market.

Breadth of Change and Risk Management

Our survey also revealed inconsistencies in interpreting the scope of risk management and change management for trading algorithms. The primary challenge was determining the extent to which firms should manage their algorithms: should it include only the algorithm itself, or also the trading software, venue gateways, servers, and networks on which the algorithm runs?

According to RTS 6 Recital 2, Algo Risk Management should encompass all of the above. As stated: “An investment firm should address all risks that may affect the core elements of an algorithmic trading system, including risks related to the hardware, software, and associated communication lines used to perform trading activities.”

However, when it came to Algo Change Management, the regulation’s wording was less clear. RTS 6 Article 5.1 states: “Prior to the deployment or substantial update of an algorithmic trading system, trading algorithm, or algorithmic trading strategy, an investment firm shall establish clearly delineated methodologies to develop and test such systems, algorithms, or strategies.” The term “Algorithmic Trading System” remains undefined, leading to inconsistencies in interpretation across the firms we surveyed.

Operational Delays and Efficiency Gaps

Time for Change

One of the most notable impacts of these regulations has been the increased time required to approve and implement new algorithms or make material changes to existing ones. Our survey highlighted significant discrepancies among firms in this regard. Some firms managed to approve new algorithms within a few weeks, while others measured their approval processes in months. Similar timescales were observed for material changes.

Non-material changes generally saw faster approval times, especially for firms utilizing workflow tools. These tools helped streamline the approval process by providing transparency on where delays occurred and ensuring that all necessary information was shared upfront, preventing rejections downstream. In fact, 80% of firms using such tools approved non-material changes within five days or less.

Some firms also employed a practice of “pre-approving” changes, conducting due diligence and testing in advance so that they could quickly switch between algorithm configurations when necessary.

Inconsistent Risk Management Approaches

Our survey revealed substantial differences in how firms define “material” and “non-material” changes to algorithms. Product changes were universally seen as material, but opinions varied when it came to changes in trading venues or pricing models. For instance, some firms viewed adjustments to pricing models as non-material, while others required full risk management reviews for even minor changes to pricing curves.

Involvement of Multiple Stakeholders

The number of stakeholders involved in the approval process for new algorithms or changes also varied widely among firms. Between five and ten individuals were typically involved in the approval of new algorithms, with fewer involved for material and non-material changes. Interestingly, the number of stakeholders did not always correlate with the time taken to approve changes, suggesting that the effective use of workflow tools can mitigate delays even with a larger group of reviewers.

Opportunities for Improvement

Streamlining Approval Processes

To improve efficiency, firms could reduce the number of people involved in the approval process, particularly for non-material changes. Additionally, running algorithm management committees more frequently—rather than on a monthly or quarterly basis—could prevent delays for urgent changes.

Expanding the Use of Workflow Tools

Our survey provided clear evidence that firms using automated workflow tools were better equipped to manage algorithmic changes efficiently. These tools proved particularly effective for non-material changes, with automated testing further enhancing efficiency.

Pre-Approval

By pre-approving certain algorithm configurations, firms can bypass lengthy approval processes for non-material changes, reducing regulatory overhead when a swift response is required.

Conclusion

The introduction of MiFID II has had a profound impact on how firms manage their algorithmic trading operations. While the regulation has improved transparency and risk management, it has also introduced operational challenges, particularly in terms of slowing down the implementation of new or modified algorithms.

However, firms can increase efficiency and reduce approval times by adopting workflow management tools, standardizing definitions around materiality, and holding more frequent algorithm management committee meetings. Reducing the number of stakeholders involved in non-material changes can also streamline the process, enabling firms to respond more quickly to changing market conditions.

The firms we surveyed showed few signs of the regulatory gaps ESMA might be looking for, but clearer guidance on materiality and “limited human intervention” would certainly benefit the industry.

For more information on our algorithm management optimization programs, contact us at [email protected] .