The Algorithmic Energy Frontier: Defending Predictive Market Models Against REMIT II Manipulation Claims
Algorithmic Trading Has Become a REMIT Governance Issue
Algorithmic execution is no longer a peripheral feature of European wholesale power trading. Day-ahead and intraday desks increasingly rely on automated models to process price signals, cross-zonal capacity, order-book conditions and portfolio constraints at a speed that manual trading cannot match.
The regulatory problem is that the same speed and repetition that make an algorithm commercially useful can also produce trading patterns that resemble manipulation when viewed from the outside. Under the revised Regulation on Wholesale Energy Market Integrity and Transparency (REMIT), often referred to as REMIT II, algorithmic trading is now subject to explicit systems, controls, testing, monitoring and notification requirements.
For trading houses, this changes the compliance question. It is no longer enough to say that the model was designed to optimize a legitimate position. A desk must be able to reconstruct why the algorithm acted, which inputs triggered the action, what limits were active and how the system was tested before deployment.
This matters particularly in coupled power markets, where congestion, rapidly changing transmission capacity and short trading windows can cause an algorithm to modify or cancel orders repeatedly within minutes. A pattern that is commercially rational from inside the model can look very different in a surveillance alert.
Why Legitimate Automated Trading Can Trigger Manipulation Concerns
REMIT prohibits market manipulation and attempted market manipulation in wholesale energy markets. The prohibition is technology-neutral: an order generated by software is not treated more leniently simply because no trader clicked the button manually.
Surveillance therefore focuses on the market effect and the surrounding evidence. Repeated order placement, modification or cancellation may attract attention where the pattern could create false or misleading signals about supply, demand or price, or where it appears designed to move the market toward an artificial level.
Three features make algorithmic desks particularly exposed to that scrutiny:
High-frequency order changes during congestion
Weak evidence linking market data to execution decisions
Insufficient records of model limits, testing and overrides
The key distinction is evidence. A regulator reviewing an alert cannot infer the commercial logic that existed inside a proprietary model unless the market participant can produce it. The compliance framework therefore has to make the algorithm explainable after the event, even if the model itself operates automatically in real time.
The HUPX–OPCOM Inquiry: When Capacity Rebalancing Looked Like Spoofing
A Switzerland-incorporated energy trading house used a proprietary predictive pricing model to capture hourly spreads between the Hungarian HUPX and Romanian OPCOM power markets. The model adjusted its orders in response to price movement and changing cross-border transmission conditions.
During a period of severe grid congestion, the algorithm executed a rapid sequence of order modifications while the desk rebalanced physical cross-border volumes. The execution pattern generated a REMIT surveillance alert because parts of the activity resembled patterns commonly associated with layering or spoofing.
The relevant regional authority then required detailed information about the trading logic, order sequence and operational controls. The commercial risk was immediate: an unresolved manipulation inquiry could affect exchange and clearing relationships. For breaches of REMIT’s market-manipulation prohibition, the revised framework requires Member States to provide for maximum administrative fines for legal persons of at least 15% of total annual turnover in the preceding business year, subject to national implementation and the Regulation’s limits.
The central issue was therefore not whether the algorithm traded quickly. It was whether the desk could demonstrate that the order changes reflected real portfolio and transmission constraints rather than a strategy intended to create false or misleading market signals.
The Compliance Architecture: Turning Trading Logic into Regulatory Evidence
Algorithmic Architecture Audit
- Mapped order changes to real-time TSO and capacity inputs.
- Reconstructed the sequence between market data, model decisions and executed orders.
- Reviewed thresholds, execution limits, testing and exception controls.
- Prepared a regulator-facing explanation of why the trading pattern occurred.
AI Governance & IP Controls
- Assessed the model against the EU AI Act by function and use case rather than labelling it “high-risk” by default.
- Separated AI governance questions from the REMIT market-conduct analysis.
- Documented ownership and licensing of the proprietary source code.
- Reinforced IP protection through Swiss licensing and contractual controls.
The work produced two distinct compliance layers. The first explained the conduct of the trading system under REMIT. The second addressed governance of the predictive technology itself, including AI Act applicability, internal accountability and protection of the underlying intellectual property.
That distinction is important. The EU AI Act does not automatically classify an energy-trading model as a high-risk AI system merely because it uses machine learning or influences a commercially significant decision. Classification depends on the system’s function and the specific categories set out in the legislation. REMIT, meanwhile, applies to the trading conduct regardless of whether the decision is generated by AI, deterministic code or a human trader.
What Revised REMIT Requires from Algorithmic Energy Traders
Article 5a of revised REMIT creates an explicit control framework for market participants engaged in algorithmic trading. The requirements are operational, not cosmetic.
Trading systems must be resilient, have sufficient capacity, operate within appropriate thresholds and limits, and contain controls designed to prevent erroneous orders or disorderly market behaviour. Market participants must also maintain business-continuity arrangements and ensure that their systems are fully tested and properly monitored.
There is also a direct notification obligation. A market participant engaging in algorithmic trading in a Member State must notify the national regulatory authority of the Member State where it is registered under REMIT and ACER. The relevant national regulator may request descriptions of algorithmic strategies, trading parameters, risk controls and evidence of testing, and records covering these matters must be retained for five years.
This is why an algorithmic compliance file should exist before an investigation begins. A description assembled months after the relevant trading event is inherently weaker than contemporaneous documentation showing the model version, parameters, input data, limits and control environment that actually applied at the time.
For businesses deploying or scaling automated execution models, our Algorithmic Trading & AI Compliance practice focuses on connecting those technical controls with the legal obligations governing wholesale energy trading.
The AI Act Matters, but It Does Not Replace REMIT
The EU AI Act and REMIT regulate different risks. The AI Act creates a horizontal framework for the development and use of AI systems, while REMIT protects the integrity and transparency of wholesale energy markets.
An energy trading desk should therefore avoid using an AI classification exercise as a substitute for market-conduct controls. Even a model that falls outside the AI Act’s high-risk categories can still generate orders that trigger REMIT concerns. Conversely, a technically sophisticated AI governance programme does not prove that a particular order pattern was legitimate.
The practical approach is to map the two regimes separately and then identify the points where they interact: system documentation, change management, human accountability, data quality, testing, monitoring and incident response. Where a third-party or general-purpose AI model is incorporated into the trading stack, additional AI Act questions may arise and should be assessed according to the role of the business in that AI supply chain.
Build the Evidence Trail Before the Regulator Asks
ACER’s 44th REMIT Quarterly, covering the first quarter of 2026, specifically highlighted algorithmic trading and its implications under REMIT. That attention reflects a broader reality: automated trading is becoming a normal part of market surveillance rather than an exceptional technical issue.
A defensible control environment should make it possible to answer basic questions quickly: What strategy was running? Which version of the model was deployed? Which market and capacity data did it consume? What order and position limits applied? Who could suspend the model? What happened when the data feed failed? Were unusual execution patterns reviewed?
The strongest evidence pack normally combines technical and legal records. Model documentation, change logs, testing results and execution parameters should align with policies, exchange rules, REMIT notifications and the firm’s market-abuse controls. If the business also trades derivatives or uses financial hedges alongside physical positions, the analysis may need to extend into commodity trading and derivatives compliance.
This does not require disclosing proprietary code indiscriminately. It requires creating a controlled record that demonstrates how the system was intended to behave, how it was supervised and why a disputed execution pattern occurred.
What Algorithmic Trading Desks Should Review Now
- Map every algorithm to the responsible legal entity and market. Identify which REMIT registration, organised market place and regulatory authority sit behind each automated strategy.
- Confirm Article 5a notifications. Check whether the required notifications to the relevant national regulator and ACER have been made and remain accurate as trading activities change.
- Create a five-year regulatory evidence file. Preserve strategy descriptions, parameters, limits, testing records, model changes and material incidents in a form that can be reconstructed after the event.
- Test for market-abuse scenarios, not only profitability. Back-testing should examine whether rapid cancellations, repeated amendments or interaction between related strategies could generate misleading market signals.
- Separate REMIT, AI governance and IP protection. They interact, but they answer different legal questions and should not be collapsed into one generic “AI compliance” document.
Zero Penalties, €2.5 Million in Trading Capital, and the Real Lesson
In the HUPX–OPCOM matter, the reconstructed trading logic and compliance evidence demonstrated that the rapid order modifications were connected to genuine cross-border capacity and portfolio adjustments. The inquiry was closed without penalties, the trading desk retained market access, and the client was able to deploy more than €2.5 million in active trading capital.
The broader lesson is not that a sophisticated algorithm can be made immune from regulatory scrutiny. It is the opposite. Automated execution increases the need for a legal and evidentiary framework capable of explaining the model when trading behaviour is challenged.
For energy traders, the most valuable compliance work happens before the surveillance alert: documented strategy logic, tested controls, clear escalation procedures and records that connect automated decisions to legitimate commercial inputs.
Nykitenko Legal advises trading houses and energy-tech businesses on algorithmic energy trading, REMIT controls and AI governance, including regulatory response strategies where automated execution is already under scrutiny.