Skip to main content

New ESMA report on AI – growing interest, real risks

The European Securities and Markets Authority (ESMA) has published a report analysing the growing use of artificial intelligence (AI) in EU securities markets and the potential related risks. The report, “Artificial intelligence in EU securities markets”,¹ was compiled based on interviews with wide-ranging market participants, written surveys, multistakeholder workshops, and a review of relevant literature. A cross-sectoral review of existing market intelligence, research, and policy supplemented the information. ESMA also carried out substantial data analysis – for example, the use of text mining to screen around 145,000 financial documents for evidence of AI use.

Overall, the report highlights increasing interest in how AI can be leveraged in the financial services sector and contributes to an improved understanding of how different approaches to using AI may or may not add value. However, it also concludes that elevated AI use may pose risks to firms, their clients, and the wider financial markets, which should be carefully managed.

Below is a summary of ESMA’s conclusions on the use of AI across asset management, trading and certain other participants in the financial markets and the potential related risks.

1. Use of AI in asset management


Interest in the use of AI for portfolio management is growing.

A variety of new tools that can be classified as “AI” are being used in asset management. These range from discretionary portfolio managers using AI for fundamental analysis, to quantitative funds deploying AI to optimise portfolio construction or to execute systematic investment strategies more broadly. In general, an increasing number of funds now use AI to extract information from large, varied numerical and textual databases with minimal supervision. The exact techniques they apply varies, as does their approach to integrating AI outputs into their investment decisions.

Investment funds are hesitant to publicise AI use

ESMA’s data analysis led it to conclude that a substantial proportion of investment funds using AI are unwilling to publicise the fact. The report suggests this may be because individual investors distrust AI due to a lack of transparency and accountability, and therefore, firms are wary of reputational repercussions from explicitly promoting the use of AI. ESMA’s analysis also points to the lack of a clear definition of what qualifies as AI and the auxiliary role that AI plays in the overall investment decision-making process.

Notably, the ESMA report highlights unanimous agreement among industry executives “that AI is not tantamount to autonomous decision-making without human oversight.”² They maintain that the best results are obtained when AI is combined with human judgment.

No notable increase in AI robo-advisors

Across its research, ESMA did not observe a noticeable increase in the use of AI in robo-advisors (computer programs that produce optimal portfolios tailored to investor’s risk appetites). The overarching reason offered by ESMA is that increased AI use in robo-advisors may create more problems than it solves for firms. First, it requires a more complex framework, which may, in turn, reduce explainability. This can make retail investors wary, given that explainability has been found to be a critical factor affecting consumers’ trust in automated platforms. Secondly, additional AI use may conflict with existing regulatory requirements, such as GDPR’s right to an explanation, which empowers users to inquire about the logic involved in an algorithmic decision affecting them.

2. Use of AI in trading


Firms are experimenting with AI in pre-trade analysis

In the pre-trade process, investors are leveraging AI models to analyse signals in asset prices and identify investment opportunities. These insights can either be corroborated through subsequent human review or can form part of algorithmic trading strategies which inform and execute trading decisions. While most algorithmic trading by banks and non-bank market makers is still built around rules-based models, recent accounts suggest that many large proprietary trading firms have integrated AI in the form of Machine Learning models.

Additionally, the report highlights a promising use case for AI in pricing securities lending transactions, which otherwise require lenders to address thousands of inquiries daily regarding the available securities inventory for short selling. To alleviate this challenge, lenders are using AI to set optimal securities lending prices and predict which securities will transform into ‘hard-to-borrow’ securities.³

AI adds the most value in trade execution

Accurately estimating market impact isa critical but notoriously complex activity. The report explains how AI is being used by groups like brokers and pension & hedge funds to minimise their market impacts and, thus, their transaction costs.

Additionally, we believe that firms using AI for trade execution purposes to conduct client activity should consider the extent to which the best execution factors required under MiFID II are incorporated into the AI model and the firm’s ability to evidence this through appropriate documentation.

Firms find less value for AI in post-trade processing

Based on feedback it received from central clearing counterparties and central securities depositories, most entities of this kind are not using AI. The barriers they face include the limitations of their legacy technology infrastructures, and their perceptions around the limited additional value that AI can bring. Nevertheless, ESMA argues that the ‘failed’ or ‘successful’ label assigned to each of a client’s past trades provides a promising data environment for AI to unlock value in the future.

3. Use of AI by other financial markets participants


Credit rating agencies continue to explore AI possibilities

The report describes some limited use of AI by credit rating agencies (CRAs), mostly to help with sourcing and processing large quantities of data. While some CRAs state that they do not use AI at all, most survey respondents were exploring its possibilities. Respondents reported that they were interested in AI tools and expected other entities would be looking to future digital solutions that could provide more timely and reliable data.

Proxy advisory firms focus on AI for research

In the main, ESMA found a section of proxy advisory firms were using AI to source, synthesise and process information. In particular, for the research and data they would provide to institutional investors. There was also a strong signal that the growing demand for ESG-related analysis was driving the development of AI tools in this area.

ESMA highlights AI risks


As well as creating opportunity, ESMA makes it clear that increased adoption of AI is elevating risk. The most relevant risks in the context of securities markets include:

  • Lack of explainability
  • Concentration, interconnectedness and systemic risk
  • Algorithmic bias
  • Operational risk
  • Data quality and model risk

While ESMA acknowledges that many of these risks are not inherent to AI, they tend to be amplified in AI systems, which “typically operate at greater scale, complexity, and automation than traditional statistical tools.”⁴ Some risks, like lack of explainability, may directly affect firms by deterring retail investors while increasing regulatory risk at the same time. The distribution of risk will vary across different financial services. For example, credit rating agencies placed special emphasis on model risk, operational risk, ethical concerns and reliability issues. In short, we believe that AI risk management approaches and control frameworks need to be tailored to each firm’s specific business environment to ensure they are fit for purpose.

The report also acknowledges that, should the growth of AI continue, risks could coalesce to affect the entire European securities market. For example, the small number of AI providers may cause market concentration, amplifying systemic risks arising from the majority of market participants deploying the same or similar investment strategies.

AI and securities markets: Growing opportunity, growing risks


These latest insights from ESMA support that AI use is gradually increasing across the EU securities market and that, with responsible stewardship, AI can benefit both individual firms and the EU securities market as a whole. A foundation for this is a robust AI risk control framework, which facilitates a systematic approach to identifying and mitigating risks across the entirety of the AI lifecycle.

Our algorithm and AI assurance team has market leading experience supporting financial services companies to capitalise on algorithms and AI through effective controls. To discuss the opportunities and challenges your organisation faces, please reach out.




¹ESMA50-164-6247-AI_in_securities_markets.pdf (, accessed 2 February 2023.

²Ibid, page 11.

³Per ESMA, “hard-to-borrow securities are securities whose supply is limited for short selling. Naturally, they carry higher fees when used for short selling.”

⁴ESMA, AI in Securities Markets, page 17.

Get the latest updates from the Deloitte Audit & Assurance blog