Using foreign experience in the use of artificial intelligence in the supervision of financial market participants in Russia.

Atabekov Atabek Rustamovich PhD in Economics Associate Professor of the Department of Administrative and Financial Law, Legal Institute, Peoples' Friendship University of Russia 117198, Russia, Moscow, Miklukho-Maklaya str., 6 atabekoff1@mail.ru Other publications by this author

It is well known that the modern financial market is the basis for the modern global economy. Financial markets include stock exchanges such as the ASX in Australia, NASDAQ in the USA, and RTSI in Russia, which provide platforms that support corporate fundraising and provide opportunities to facilitate trade and investment.

In his speech, Tony D'Aloisio (ex-chairman of the Australian Financial Regulator ASIC) notes that the fundamental purpose of the Corporations Act is investors' faith in the fairness of the actions of all participants in the financial markets. Actions aimed at market manipulation or illegal insider trading damage their trust and prevent them from participating in these markets^[1].

As O.M. Shevchenko notes, the level of effectiveness of regulatory measures in this area directly affects the degree of investment attractiveness of the economy of a particular state where investors are directed^[2].

A similar position is shared by Brown and Glodschmidt, who determine that effective tools for detecting unusual trading behavior are the basis for further strengthening the confidence of market participants, increasing market liquidity and reducing the cost of capital for business^[3].

The regulatory framework for preventing insider trading and market manipulation consists of the following documents:

1.                 In Australia, the Corporations Act 2001 qualifies actions related to market manipulation and insider trading as crimes (section 1043A and 1041A-H)^[4]. Within the framework of these provisions, a person who possesses insider information is prohibited from trading a financial product that this information may "significantly affect". At the same time, the key issue for classifying information as insider is the possibility for the subject of the offense to determine it as significantly affecting the price or value of a financial product. In addition, in accordance with the legislative provisions, manipulation is not required to lead to a beneficial result for the person involved^[5].

2.                 In the United States, the Securities and Exchange Commission (SEC) and the Financial Industry Regulatory Authority (FINRA) play a regulatory role in controlling financial markets. Both organizations provided guidance on market manipulation and insider trading^[6]. General approaches for identifying actions as insider trading are presented in the FINRA 2020 Rules, which states that no entity should make any transactions or encourage the purchase or sale of any security using any manipulative, misleading or other fraudulent action or device^[7]. In addition, SEC rule 10b5-1 prohibits the purchase or sale of securities of any issuer based on material classified information about these securities^[8].

3.                 In the Russian Federation, the Central Bank of Russia plays a central role in the regulation of financial markets. The fundamental regulatory framework is Federal Law No. 224-FZ of July 27, 2010 "On Countering the Misuse of Insider Information and Market Manipulation and on Amendments to Certain Legislative Acts of the Russian Federation", which regulates relations arising in the commodity, currency and stock organized markets. At the same time, detailed provisions regarding the definition of insiders (12 categories) and manipulative actions in relation to financial markets are presented (8 actions) in Articles 4 and 5, respectively.

It should be noted that the starting point for the emergence of algorithmic decision support systems for exchange trading purposes is 1998, in which the SEC approved the possibility of using trading robots^[9].

At the moment, high automation of trade, openness to data and the presence of a large information array give rise to a large list of issues for representatives of control and supervisory activities that require a predictive set of responses.

Financial institutions face a high level of risk of internal and external influence for criminal purposes requiring assistance from supervisory authorities. The IBM report records that financial institutions record a very large number of false positive notifications about potential suspicious activity, which do not lead to its subsequent fixation or registration. The reason for this is a combination of outdated technology and incomplete and inaccurate data^[10].

The volume of daily data is constantly growing and requires new models of response and analysis of the actions of market participants by the supervisory authorities^[11-12]. The NASDAQ study signals that the growth of stock market participants using trading monitoring algorithms leads to an increase in false positives of algorithms and, as a result, creates information noise for regulatory authorities, complicating their activities^[13].

The IOSCO Market Surveillance report signals that trading in financial instruments has become more dispersed across a growing number of trading platforms and is difficult to track. High-frequency asset trading without strict reference to geographical and other boundaries is carried out at speeds faster than a millisecond. This fact is actively used by traders for manipulative activities and is a difficult factor to disclose for regulatory and supervisory authorities^[14].

In this regard, it becomes logical to introduce artificial intelligence technology to ensure compliance with regulatory requirements by financial market participants who use high-precision trading algorithms and other automated decision-making and analysis systems.

In this regard, it is necessary to consider the experience of AI forecasting in the related field of law enforcement – law enforcement and judicial activities. Two software tools, ECHO and ALIBI, are particularly interesting examples of the potential use of AI for policing^[15]. ECHO uses artificial neural networks to model the judge's reasoning based on the input evidence in the case, and at the output a predictive transcript of the court decision on a specific case is formed. ALIBI is a program that simulates the accused of committing a crime. The algorithm uses a neural network to deconstruct the defendant's actions into their component parts, and then to build a defensive strategy around each of these composite actions to reduce guilt, modeling the behavior of the jury, based on the preliminary state of the evidence that is supposed to be used during the trial.

Considering similar cases with regard to control and supervisory detail, it is necessary to note the SMARTS software tool (created back in the 1990s), which monitors the activities of financial market participants represented at more than 30 sites around the world, including the Hong Kong Stock Exchange^[16]. This program analyzes the activity of traders, and then issues warnings about suspicious transactions, which are subsequently reported to the relevant authorities.  At the same time, profile employees have to track the relevant trading data, according to which a signal was received from the system not only in one area (the traded instrument), but also in all related and adjacent segments of the financial market^[17].

The Central Bank of Russia also actively uses technologies in the field of algorithmic remote monitoring of financial markets, including SupTech and RegTech technologies^[18]. Experts note the problem of unfair assessment by financial organizations of the fair value of assets due to the formalist approach of financial organizations to the establishment of criteria for an active market, and in some cases also by deliberately setting criteria corresponding to the minimum trading activity^[19].

Active implementation of machine learning (an integral element of AI, which implies an autonomous solution of the problem without active human intervention on the basis of existing methods of cognition)^[20] will reduce the previously mentioned information noise generated by analytical systems and increase the effectiveness of the control and supervisory detail of the authorities. In addition, the existing analytical systems are deprived of the possibility of analyzing the trading system, communication data (online and offline communication of traders) and other implicit sources of information.

One such practical example is IBM Financial Crimes Insight, which offers a holistic and cognitive approach to monitoring virtually all activities related to employees involved in trading and related activities with financial markets. It goes beyond the traditional detection of alerts based on algorithmic trading and actively monitors employees in real time, including emails, chat transcripts, voice recordings, trading and market data^[21].

At the same time, one of the constant problems for the use of such AI tools is access to data and confidentiality. Financial regulators, trading platforms and market participants have access to data that may be useful for identifying and investigating financial offenses, but all relevant data is usually not available to all participants. This may limit the applicability of surveillance technologies based on pooled data sources. Referring to the combination of artificial intelligence and human intelligence, IBM recognizes this risk and argues that AI is the next stage in the fight against financial crimes, which can be effectively implemented with a consolidated approach.

In practice, market manipulation is carried out by "stuffing" false information through mailing lists, news feeds, etc., which allows unreasonably raising or lowering the rate of the quoted instrument^[22], or creating other strategies inside the "glass" that create an artificial deficit or surplus of the traded instrument^[23]. In addition, the behavior of attackers when manipulating the market will depend on the trading volume of a particular financial product and the volatility of the market of this product, which will also significantly affect the AI analysis algorithms and potentially mislead it.

In this context, it is necessary to consider the experience of using digital sandboxes that allow you to create a secure environment based on real data sets together with synthetically generated scenarios that provide testing in market conditions that are not present in historical data sets. At the level of practical implementation, it is necessary to note the positive experience of the UK financial regulator, which processes over 200 databases on traded instruments within the sandbox^[24].

V.V. Mazurin in his research comes to the conclusion that one of the positive aspects of the integration of AI into the sphere of financial law and regulation of relations arising in financial markets for the purposes of supervisory activities is that AI allows you to remove a significant part of the analytical burden and collection of evidence, and is not subject to corruption risks^[25].

M.V. Anashkina notes a significant time delay in the detection of cases of manipulation of the domestic financial market, amounting to more than 3 years. At the heart of this delay, the author notes the technological lag of market analysis tools, the imperfection of legislation in this area and the lack of existing real-time analysis tools^[26].

As part of a comprehensive review of the digitalization of the public administration function, including the introduction of AI in financial supervision, A.E. Morozov notes the need to create regulatory platforms and strengthen the role of self-regulation for professional intermediaries and IT companies (including those engaged in AI development) with the direct participation of the Bank of Russia.^[27] E.V. Goryan declares a similar position on the basis of legal research, having studied the experience of implementing AI in the public and financial sector of Singapore^[28]. As a result of the analysis of the instrument developed by the Monetary Authority of Singapore for regulating processes in the field of artificial intelligence, and the practical case of DBS Bank, the need to develop flexible regulatory approaches to AI technology and the formation of a framework model for AI management is noted.

Considering the issue of separate regulation of AI in the context of countering the legalization of proceeds from crime, it is necessary to highlight and take into account the position of O.A. Yastrebov, who proposes to use the design of an electronic person to determine the actions of artificial intelligence, as well as to ensure the introduction of this category into the current legislation in order to regulate relations, one of the elements of which is artificial intelligence^[29].

Based on the above, the author of this article formulates the following proposals for the further effective use of AI in the field of control and supervisory activities of the financial markets of Russia:

1. It is necessary to form regulatory approaches that ensure the creation of digital twins and regulatory sandboxes for AI, which is planned to be implemented and applied within the framework of the supervisory activities of the Central Bank of Russia and other authorities.

2. It is necessary to legislate additional requirements for government officials implementing AI; the requirements should take into account both qualification positions (education, experience in AI development, maintenance, advanced training in AI, etc.) and minimizing the risks of bias against technological solutions in the field of AI.

3. When developing AI technology for the financial supervision of the Bank of Russia, it is necessary to legislate a direct ban on the dissemination of the AI technology being developed for participants in the financial market and related sub-sectors of the specified market.

4. It is also proposed to legislate the obligation to provide the source code and the tool for the operational shutdown of AI to the Bank of Russia in the case of the use of AI by financial market participants.