Limitations On The Use Of Artificial Intelligence
By Ambrose Tan, Head of Dealing – Asia Pacific, Aberdeen Standard Investments
The successful application of the latest technologies to the trade execution process requires clear and consistent rules and standards.
Diverse global regulatory regimes and contrasting trading procedures in different asset classes restrict the effectiveness of new technologies in the trading process. Custom, inertia and the difficulties in enforcing change in dynamic markets mean differences in asset class trading practices are likely to remain entrenched for some time. On the other hand, a common policy shared among regulators might facilitate a more confident deployment of the latest technologies, notably artificial intelligence (AI).
The application of AI can offer buy-side dealing desks undoubted benefits, notably by reducing the time it typically takes for a human trader to gather and analyse information required to facilitate best execution. AI can access data faster, and then recognise trading patterns in order to provide the basis for a better informed trade decision. In future, it is feasible that dealing desks may have AI specialists working alongside human traders.
However, currently, it has fundamental limitations: AI is only able to access and evaluate the data that is available, recognise patterns and factors within pre-established parameters, and therefore cannot make recommendations based on inherently incomplete material, subjective instructions and constraints. AI can suggest a “most likely” outcome, not predict a definitive result.
On the face of it, a statistically accurate assessment and decision forecast might seem attractive. But, both regulators and clients could feel justifiably uncomfortable. Regulators, especially in Europe following the introduction of Markets in Financial Instruments Directive (MiFID) II this year, might conclude that if a platform’s software is provided by a sell-side firm, and it promotes a decision-making process based on AI, then it could be an inducement to trade, which would potentially hinder best execution and not work in the interest of clients’. The MiFID II requirement for buy-side firms to provide LEI numbers designating discrete, individual accounts tightens the scrutiny even further, but promotes greater trade transparency and reporting. It is a fine line between enticement and inducement, and the advertisement of a buy or sell signal might fall on the side of the latter.
Yet, the unbundling of research provision and trade execution in itself is unambiguous, so it is important that a fund manager avoids stepping over that line, or is even suspected of doing so. The onus is on the fund management firm and its traders to be fully cognizant of brokerages that have been selected as authorised providers of (qualitative and quantitative) research material and ensure that there is no suspicion that it is being used as an inducement to trade.
Moreover, the regulator is also wary of a buy-side firm that relies on an in-house developed AI system for its trading and investment decisions, rather than a process featuring directly accountable individuals. Ultimately, humans, not machines, must be held accountable.
Meanwhile, clients too might legitimately complain that a trading decision, which is integral to a buy-side firm’s investment process, should not be the product of a machine-generated computation. After all, they are paying the firm for the skills and experience of its fund managers and traders.
Nevertheless, AI is a useful tool for helping trade some, but not all, asset classes. The main determinant is the comprehensiveness of accessible data. Information about exchange-traded equities is usually sufficiently extensive and reliable for AI systems to operate, but far less so for fixed income and foreign exchange markets where most trading occurs over-the-counter (OTC), directly with sell-side counterparties or anonymously via broker intermediaries. Executed transactions and post-trade information can be opaque. Price makers are conscious of market impact, especially during bouts of illiquidity, hence there is an emphasis on protecting market flows.
Fixed income markets are characterised by brokerages’ inventory supplies, risk capacity and niche expertise; although AI might identify trading patterns, ultimately a human trader’s networks and experience are a more reliable way of achieving best execution.
Foreign exchange markets are generally opaque, especially spot trades where price activity is determined by myriad influences. Speculative behaviour remains a powerful force, but less so now that hedge funds are under tighter regulatory scrutiny. Instead, “genuine” transactions that support or hedge investment and corporate treasury decisions dominate the spot market and AI will struggle to determine useful patterns among those trades.
MiFID II seeks to address some if not all the above issues as it aims to create a level playing field for all market participants and all markets with a push for greater disclosure and transparency. Now, transactions that involve foreign exchange products – options, non-deliverable forwards and currency and interest rate swaps – are reportable to the market within 15 minutes of trading. Trading data on various venues including Multi Trading Facilities (MTF), aka multi-dealer platforms, will also be captured. New legislation seeks to monitor not just post-trade but pre-trade analysis as well.
While these developments may spell a positive for AI trading, we go back to the fundamental issue of human accountability. Will AI create an environment in which volatility spikes up in the face of speedier data crunching?
Disparate jurisdictions and rules
A more fundamental problem is the confusingly disparate regulatory regimes throughout the world that force buy-side firms, in particular, to be circumspect about the adoption of some technologies – such as AI – that should in theory benefit the investment and trading processes.
Some financial regulators in Asia do manage to combine a firm oversight (across all asset classes), while also being receptive to global regulatory trends and open to the development of new trading technologies. Elsewhere in Asia, there are several emerging markets with diverse regulatory regimes which are a challenge to institutional investors with a global or regional mandate.
Yet, perhaps a standout example of regulatory anomaly is actually between the rules in two developed jurisdictions. In a direct contradiction to the unbundling requirements of MiFID II throughout Europe, in Japan, brokers are not allowed to charge fees directly for research, but instead bundled in transaction charges.
If the various worldwide regulatory bodies could agree on common policies and standardised rules, then they would help money managers and their clients achieve their objectives more easily. More importantly, consistency would allow a successful adoption and implementation of the new technologies that are increasingly available.
Aberdeen Standard Investments is a brand of the investment businesses of Aberdeen Asset Management and Standard Life Investments.
The views expressed in this article are the author’s and not the company’s.
We’d love to hear your feedback on this article. Please click here