ACM and FIA Publish Guidance on Algo Trading

The ACM report provides a comprehensive overview of algo trading evolution specifically on wholesale energy market as noted above across the following four themes:

1. Prevalence and Types of Algorithms;
2. Drivers of Algorithmic Trading;
3. Market Impact of Algorithmic Trading on Wholesale Energy Markets; and
4. Compliance and Regulation (REMIT & MiFID 2)

[1] Prevalence and Types of Algorithms

The report provides visual representations of the different types of algos being used (see page 17) and breaks down how each type of algo is being applied in the power and gas markets. The three types of algos currently in use are:

• Execution Algorithms. Used to execute a trading decision made outside the algorithm. Parameters of execution algorithms are set outside the algorithm, after which the algorithm places orders on the trading platform in an optimal way through a specified method, usually within certain price and volume limits.
• Signal Generators. Algorithms that – based on a set of inputs – signals trading opportunities or other supporting information for trading decisions to traders or execution algorithms.
• Trading Algorithms. Make trading decisions and executes them automatically without the manual intervention of a trader. It differs from execution algorithms in one important respect: the algorithm additionally decides whether an order should be submitted on the trading platform or not. That is the case when signal generators are directly linked to execution algorithms.

ACM analysed how algorithmic trading is used in both spot power markets and natural gas markets across the three types of algorithms:

"The results of the survey indicate that, in the natural-gas market execution algorithms are more frequently used than signal generators and trading algorithms, while, in the power market, all types of algorithms are used to the same extent."

Natural Gas Trading. Out of a total of 8 respondents that use and/or develop algorithms for gas trading:

• Six respondents report the use of execution algorithms;
• Three respondents report the use of signal generators and trading algorithms; and
• Machine learning is mainly used or developed for generating trading signals.

Spot Power Market Trading. Approximately nine out of 12 respondents use and/or develop algorithms for power trading.

• Execution algorithms, signal generators and trading algorithms are all relatively applicable to each of the nine respondents.
• Machine learning is applied to both generating trading signals and trading algorithms on the power market.

[2] Drivers of Algorithmic Trading

The shift towards renewable energy sources and the need for efficient trading processes are key drivers behind the increasing use of algorithms in energy markets. Algorithms help market participants manage the complexities of trading renewable energy, which requires constant balancing due to the unpredictable nature of production from sources like wind and solar.

• Energy Transition. The shift towards renewable energy requires constant balancing of positions as it is more difficult to predict renewable energy production thus, increasing the use of algorithms for efficiency and risk management.
• Efficiency. Algorithms automate trading processes, reduce the workload on traders, and enhance the speed and accuracy of trading decisions.
• Optimization: Algorithms optimize asset utilization and mitigate risks, making them invaluable tools for managing large and complex trading portfolios.

That being said, in certain markets, trading without algorithms is becoming increasingly difficult, due to some disadvantages related to the slower speed of manual trading.

[3] Market Impact of Algorithmic Trading on Wholesale Energy Markets

ACM continues to monitor algo trading activity and notes that there are both positive outcomes and risks associated with algorithmic trading. Algorithmic trading has a significant impact on market dynamics, particularly in terms of liquidity and price formation. While algorithms enhance market efficiency and reduce trading costs, they also introduce challenges such as increased volatility and potential disconnects between market fundamentals and trading behaviour.

Positive Outcomes

• Increased Liquidity: Algorithmic trading improves market liquidity, enabling smoother and more efficient trading processes.
• Accelerated Price Discovery: The rapid processing of information by algorithms accelerates price discovery and narrows bid-ask spreads.

Risks and Challenge

• Volatility. While algorithms can increase market volatility through rapid responses, they can also complicate price determination and reduce transparency due to their complexity.
• Transparency Issues. Frequent price movements and the complexity of algorithms can obscure price determination and transparency.
• Disconnect between Fundamental Market Information and Algo-Driven trading behaviour. Algorithmic trading may increase volatility through feedback loops and rapid response to market signals, which can amplify existing market movements.
• Explainability. While algorithms also might enhance transparency by documenting trading decisions, their complexity - especially in machine learning - they may hinder explainability, impacting transparency.

[4] Compliance and Regulation (REMIT & MiFID 2)

ACM stresses the importance of compliance and internal checks by firms as algo trading becomes more prevalent in energy markets, and reminds market participants that they must ensure that their trading activities need to adhere to regulatory standards to maintain market integrity and transparency, specifically the recently updated REMIT 2 regulation which came into effect in May 2024.

It also outlines market manipulation use cases in algo trading and the changes to surveillance methods required to monitor algo trading.

Market Manipulation in Algo Trading – ‘Robot Battles’ & Order Activity Disruption

ACM notes that businesses need to prevent the use of algorithms from resulting in illegal trading activities, such as market manipulation. Two examples of market manipulation include [a] Robot Battles and [b] Order Activity Disruption.

[a] Robot Battles

‘Robot Battles’ is where two algorithms end up bidding against each other through a series of order adjustments until one algo hits its price limit. ACM provides a description of this behaviour in further detail:

"In this scenario, two algorithms compete with each other for the best (highest) buy order through a series of order adjustments. This competition continues until the algorithm representing the ‘suspect’ party reaches its price limit. At this point, the party sells at the highest priced buy order of the other party and swiftly removes its own buy order. The concern would be that the purchase price is being manipulated, given that it is pushed up to its maximum limit through layering / spoofing behaviour, which involves the rapid removal of an order on one side of the book after a transaction on the opposite side."

[b] Order activity is so extreme that it disrupts the visibility of the order book for other market participants.

Algo improvement loops is where two or more algos compete to optimise their orders. In certain scenarios, the rapid adjustments e.g. thousands of adjustments per minute can lead to a scenario where the order book is no longer visible for other traders. ACM provides a further explanation:

"This situation can arise from the correlation between algorithms that leads to improvement loops or feedback loops. These loops manifest through upward and downward order price movements, where two or more algorithms compete to optimise their orders."

Imagine a situation where two algorithms compete to be the best (lowest) sell order in the order book. When algorithm ‘A’ adjusts its price to a level that is below the price limits of algorithm ‘B’, then algorithm ‘B’ no longer alters its price such that it is lower that algorithm ‘A’. Instead, algorithm ‘B’ changes its price to be the second-best price in the order book, just slightly better than the former second-best order in the order book. Subsequently, algorithm ‘A’ follows suit by adjusting its price which slightly betters the new second-best price in the order book of algorithm ‘B’. With the best price (of algorithm ‘A’) again being within the price limits of algorithm ‘B’, algorithm ‘B’ then pursues to have the best price again by slightly bettering the price of algorithm ‘A’.

These price adjustments continue, leading to upward and downward movements. The duration of such patterns depends on the programming of the algorithms. In specific circumstances, very rapid adjustments (imagine thousands of adjustments per minute) might lead to a situation where the order book is no longer visible for other traders.

Surveillance methods to Monitor Algo Trading

The ACM observed that the increased use of trading algorithms has led to changes in surveillance methods. Specifically, investigations into suspicious algorithmic trading behaviours now demands a more data-intensive approach compared with manual trading with a shift towards quantitative data analysis and the development of advanced detection tools. ACM notes that regulators are investing in expertise and IT infrastructure to enhance their ability to detect and analyse such trading behaviours.

Compliance and Internal Checks – Measures Taken By Market Participants

ACM notes that effective compliance and internal controls are essential for managing the risks associated with algo trading. They observed that market participants have developed a range of measures to ensure their algorithms perform as intended and comply with regulatory requirements as follows:

• Compliance Measures. All market participants interviewed and surveyed have compliance and risk measures in place for their algorithms. These include setting price and volume limits and implementing kill functionalities to halt all algorithmic trading if necessary.

1. Implementation Examples: Some market participants use kill switches to immediately stop trading in case of malfunction. Price limits prevent algorithms from making trades at prices far outside the expected range, reducing the risk of significant market impact.
2. Automated Monitoring: Many firms employ automated systems to continuously monitor algorithm performance and compliance, allowing for quick detection and resolution of issues.

• Risk Management. Effective risk management systems are crucial for mitigating the risks associated with algo trading. This includes developing and testing algorithms, storing information on algorithm performance, and monitoring trading activities.

1. Algorithm Testing: Back testing against historical data and running simulations helps identify potential weaknesses and ensures algorithms behave as expected under various market conditions.
2. Real-Time Monitoring: Continuous real-time monitoring allows firms to track algorithm performance and market conditions, making it possible to adjust strategies promptly.

• Implementation and Effectiveness. While the study did not assess the effectiveness of these measures, it is clear that market participants recognize the importance of compliance. The actual implementation of these measures varies, and their effectiveness depends on the specific controls and input values set by each participant.

1. Diverse Approaches: Firms adopt different approaches based on their trading strategies and risk appetites. For example, high-frequency traders might prioritize speed and volume controls, while market makers might focus on maintaining balanced order books.

• Trading Platforms' Conditions. Trading platforms have established conditions for the use of algorithms, primarily to ensure the stability of trading systems and the quality of price discovery. These conditions include testing algorithms before deployment and continuously monitoring their performance.

1. Platform Requirements: Platforms often require traders to demonstrate that their algorithms meet specific performance and safety standards before allowing them to trade.
2. Ongoing Surveillance: Platforms conduct ongoing surveillance to ensure that algorithms do not disrupt market integrity, including monitoring for unusual trading patterns and potential manipulation.

In addition to the above observations, ACM provides further detail on governance over specific components of algo trading.

• Development and Testing Phase. Market participants rigorously develop and test their algorithms to ensure they perform as expected under various market conditions. This phase includes back testing algorithms against historical data and simulating their performance in different market scenarios.

1. Robust Testing Protocols: Testing protocols might include stress testing, where algorithms are exposed to extreme market conditions to evaluate their resilience.
2. Validation Processes: Validation processes ensure that algorithms comply with both internal policies and external regulatory requirements before they are deployed in live markets.

• Information Storage. Storing detailed information on algorithm performance is crucial for compliance and risk management. This data allows market participants to review and adjust their algorithms as needed and provides a record for regulatory scrutiny.

1. Comprehensive Logs: Detailed logs of algorithm decisions, market conditions, and trade outcomes help in understanding and optimizing algorithm behaviour.
2. Audit Trails: Maintaining audit trails ensures transparency and accountability, facilitating investigations into any irregular trading activities.

• Controls and Limits. Implementing controls and limits, such as price and volume caps, helps prevent algorithms from executing trades that could destabilize the market. These measures are designed to mitigate the risks of excessive volatility and market manipulation.

1. Dynamic Controls: Dynamic controls adjust limits based on real-time market conditions, providing an additional layer of protection.
2. Pre-Trade and Post-Trade Controls: Pre-trade controls prevent orders that exceed set thresholds, while post-trade controls review executed trades for compliance with internal and regulatory standards.

"Most of the surveyed and interviewed market participants, who currently use algorithms or who are developing algorithms, apply many pre-trade limits. Based on the survey, the following pre-trade limits are used (ordered by descending use intensity):

• Strategy positions. 14 out of 15 market participants use limits to ensure the strategy is executed properly in terms of positions.
• Order price. 13 out of 15 market participants have limits in place around the pricing of an order.
• Order value. 12 out of 15 market participants have limits in place around the value of an order.
• Number of outstanding orders. 11 out of 15 market participants have limits in place regarding the number of outstanding orders active in the order book at any time.
• Frequency limit. 10 out of 15 market participants have specific limits in place around the amount of order placements or changes per particular time period.
• Specific tradeable instruments. 9 out of 15 market participants have limits in place regulating which specific instruments can be traded by the algorithm.
• Specific trading venues accessible. 8 out of 15 market participants have limits in place regulating which specific venues are accessible by the algorithm.
• Monitoring and Compliance Measures. Continuous monitoring of algorithmic trading activities is essential for ensuring compliance with regulatory standards. Market participants use various tools and techniques to monitor their algorithms in real-time and identify any potential issues."

1. Real-Time Dashboards: Real-time dashboards provide visual insights into algorithm performance, enabling swift action if anomalies are detected.
2. Alert Systems: Automated alert systems notify compliance officers of potential issues, ensuring timely intervention.

"The majority of the surveyed market participants who currently use algorithms or who are developing algorithms have an automated surveillance system in place (12 out of 15) and require the trader to always monitor the orders of the algorithm (12 out of 15). This also holds for the interviewed market participants. Generally, such a monitoring system generates alerts for several types of problems."

• Prevention of Adverse Behaviour. Despite the implementation of compliance measures, the risk of adverse behaviour remains. The effectiveness of these measures largely depends on the specific controls and input values set by market participants. Ensuring these values are appropriately restrictive is critical for preventing undesirable trading behaviours.

1. Behavioural Analytics: Using advanced analytics to detect unusual trading patterns and behaviours that might indicate manipulation or other forms of market abuse.
2. Regular Audits: Conducting regular audits and reviews of algorithmic trading systems to ensure compliance and adapt to evolving market conditions.

"Compliance measures minimise the occurrence of adverse behaviour, though some factors may still possibly lead to unwanted outcome.

• When algorithms get more complex, the explainability decreases.
• When an increasing number of data sources and data types are fed to the algorithm, it gets more difficult to test whether a complex algorithm reacts as expected.
• Code errors or technical problems around algorithms can last a long time when they remain undetected.
• Self-learning algorithms may take less sound decisions in extreme market conditions.
• The effectiveness of applied controls and limits depend on specific input values. In case the input values are set too high or too wide, the controls and limits may not be restrictive enough in practice.
• When designing and employing algorithms, market participants must consider the occurrence of inside information."

REMIT Legislative Developments in the EU Wholesale Energy Market

The Regulation on Wholesale Energy Market Integrity and Transparency (REMIT) has been revised (i.e. REMIT 2) to address the evolving landscape of algorithmic trading. These revisions introduce new obligations for market participants and enhance the regulatory oversight capabilities of authorities like ACM.

ACM provides an overview and reminder of new REMIT obligations for market participants engaged in algorithmic trading as follows:

• New Obligations. The revised REMIT imposes several new obligations on market participants including implementing effective risk management systems, adhering to trading thresholds, ensuring business continuity, and notifying regulatory authorities of their trading activities.

1. Risk Management Systems: Participants must have systems in place to identify, assess, and manage the risks associated with algorithmic trading. This includes measures to prevent and detect market manipulation and other forms of market abuse.
2. Trading Thresholds: Setting and adhering to trading thresholds prevents excessive trading volumes that could destabilize the market.
3. Business Continuity Plans: Ensuring operational resilience through comprehensive business continuity plans that address potential disruptions.

In addition, ACM provides an overview of its enhanced oversight and responsibilities for monitoring of algo trading alongside the Dutch financial regulator AFM as follows:

• Strengthened Oversight. The revisions to REMIT enhance ACM's oversight of algorithmic trading in the Dutch energy market. ACM is now better equipped to monitor trading behaviours and enforce compliance with the new regulations.

1. Enhanced Surveillance: ACM employs advanced surveillance techniques to monitor algorithmic trading activities, ensuring adherence to regulatory standards.
2. Enforcement Actions: ACM has the authority to take enforcement actions against non-compliant market participants, ensuring a level playing field.

• Collaboration with AFM: ACM continues to work closely with the Dutch Authority for the Financial Markets (AFM) and other regulatory bodies to ensure comprehensive oversight of algorithmic trading activities.

1. Joint Initiatives: Collaborative initiatives between ACM and AFM enhance the effectiveness of regulatory oversight and enforcement.
2. Knowledge Sharing: Ongoing knowledge sharing between regulatory bodies helps keep up with technological advancements and emerging market trends.

ACM concludes its report noting future focus areas for additional market research studies as follows:

• Self-Learning Algorithms: Investigating the implications of using self-learning algorithms in trading, including their potential benefits and risks.
• Suspicious Behaviour Detection: Enhancing techniques for detecting and analysing suspicious trading behaviours to prevent market manipulation.
• Compliance Adherence: Assessing how well market participants adhere to their stated compliance procedures and the effectiveness of these measures.