Predictive AI trading: Implications for the “active market” criterion in Level 1 fair value measurement

The rise of AI marks a transformative shift across industries, with finance at the forefront. The European Union has led the way by introducing the world’s first comprehensive AI regulation, the EU AI Act, published in July 2024 and effective since 1 August 2024. This regulation establishes a risk-based framework for both developers and users of AI systems. In contrast, the United States has adopted a more fragmented approach, combining legislative initiatives, executive actions, and a Bill of Rights.

Despite differing regulatory strategies, the shared objective remains clear: to set standards that address emerging and potentially unmanageable risks associated with AI-driven processes. These frameworks, still in their infancy, allow significant flexibility, particularly in the development of predictive AI models. In the financial sector, such models leverage machine learning to process vast datasets, identify patterns, and forecast outcomes such as market movements. This enables real-time, data-informed decision-making, especially in volatile environments raising new questions about the reliability of traditional valuation mechanisms like Level 1 fair value measurement.

Predictive AI-based trading

Institutional investors, rating agencies, professional traders, and other financial market participants have long relied on algorithmic tools to automate financial and credit risk analysis. Within the stock market, predictive AI-based trading tools go a step further by leveraging historical and real-time financial data to forecast price movements and market trends.

However, the level of integration and precision we are witnessing today marks a significant evolution. The predictive AI landscape in finance particularly in stock market analysis and trading has experienced rapid expansion and is set for continued, substantial growth. This surge is driven by several key factors:

• The widespread adoption of algorithmic trading
• Advances in computational power and machine learning techniques
• The growing reliance of hedge funds on AI-driven strategies
• The increasing sophistication of quantitative trading models
• The accessibility of cloud-based AI solutions

Looking ahead, growth is expected to be fueled by a convergence of favorable conditions, including:

• Supportive regulatory environments that encourage technological innovation
• Broader global adoption among retail investors
• Continued development and enhancement of cloud-based AI services

In parallel, companies are innovating with AI-enhanced equity benchmarks, aiming to outperform traditional indexes by optimizing returns and managing risk more dynamically.

While these developments are already embedded in day-to-day market operations, their regulatory implications may not yet be fully captured. Financial analysis and decision-making, on one hand, and financial reporting, on the other, are two sides of the same coin. Predictive AI not only enhances efficiency in the former but is increasingly influencing the latter – raising important questions about how traditional valuation frameworks, such as Level 1 fair value measurement, remain reliable in an AI-driven market.

“Active market” requirement under international accounting frameworks

In financial accounting and reporting, market prices from stock exchange transactions are a primary source for fair value estimations; an approach embedded in most globally recognized accounting frameworks.

In 2009, aiming to enhance transparency around fair value measurements and liquidity risk, the International Accounting Standards Board (IASB) amended IFRS 7 – Financial Instruments: Disclosures by introducing a fair value hierarchy. This hierarchy mirrors the structure used in US GAAP under ASC Topic 820 (formerly FASB SFAS No. 157 – Fair Value Measurements).

Under IFRS 7, financial instruments measured at fair value must be classified into three levels based on the inputs used in their valuation:

• Level 1 refers to quoted prices (unadjusted) in active markets for identical assets or liabilities.
• To qualify as Level 1, an instrument must be: 
  • Quoted in an active market
  •  Valued at its unadjusted quoted price as of the reporting date
  • Based on prices that are readily observable and reflect actual, regular transactions conducted at arm’s length

In essence, a Level 1 instrument is one that is openly traded, publicly accessible, and subject to frequent, unbiased transaction activity. The responsibility for ensuring transparent and orderly trading lies with regulated market management companies, typically supervised by national authorities. These entities oversee the admission, suspension, and revocation of financial instruments and operators, thereby safeguarding investor protection.

For decades, the principle of regular, unbiased, arm’s-length transactions has been the cornerstone of fair value measurement under the “active market” definition. It has served as a sine qua non condition for the validity of Level 1 estimates.

Yet, given the remarkable progress in predictive AI technologies and the operational shifts they are driving within regulated financial markets, it is worth re-examining this foundational principle. Can we still rely on it as the bedrock of fair value measurement? Or do these technological advancements call for a re-evaluation of its relevance and robustness in today’s AI-driven trading environment?

Risks in a rapidly changing environment

The increasing integration of AI-driven trading systems is fundamentally reshaping market dynamics, raising critical questions about the continued relevance of assumptions embedded in established accounting standards. It is essential to assess whether the principle of “active market” can withstand the pace of technological change, or whether updates are needed to ensure that fair value measurements continue to reflect true market conditions in an AI-dominated landscape.

The widespread use of predictive AI tools for market analysis and execution is transforming how transactions occur, especially in terms of volume and frequency. Machine-driven trades can artificially inflate transaction volumes over compressed or extended timeframes, potentially distorting the nature of market activity. These surges may not reflect genuine shifts in investor sentiment or fundamental value, but rather the automated reactions of algorithms to data and signals.

As a result, the transparency and informational value of market prices may be compromised. Stakeholders may find it increasingly difficult to distinguish between authentic market trends and algorithm-induced fluctuations.

The principle of “actual and regular transactions,” which underpins the definition of an active market, is closely linked to the concept of arm’s length transactions those conducted between independent, unaffiliated parties, each acting in their own interest. While AI-driven markets may still satisfy the “self-interest” criterion, the independence between parties becomes more ambiguous.

When trading decisions are guided by similar or even identical algorithmic strategies, both sides of a transaction may be operating under convergent logic. This homogenization of behavior driven by the widespread adoption of comparable AI models can erode the diversity and independence of market participants. Consequently, market activity may become less representative of a broad spectrum of views and interests.

Conclusion

As predictive AI continues to permeate financial market operations, its influence on trading behavior, market structure, and price formation is becoming increasingly significant. This evolution challenges the foundational assumptions of existing accounting standards particularly the definition of active markets underpinning Level 1 fair value measurements.

To ensure that fair value continues to reflect genuine, arm’s length market activity, it may be necessary for regulators and standard setters to revisit and update the criteria used to define active markets. The traditional benchmarks based on regular, unbiased transactions between independent parties may no longer fully capture the realities of an environment where algorithmic logic increasingly drives both sides of a trade