By Dewi Habraken, senior director, institutional capital services at JTC Group (New York)
In his book published in 1864, Charles Babbage, inventor of the first digital programmable computer, found himself obliged to respond to questions about the quality of input: If you put wrong figures into the machine, will the right answers come out? In reply, Babbage wrote, “I am not able rightly to apprehend the kind of confusion of ideas that could provoke such a question.”
More succinctly, Babbage anticipated, by nearly a century, the American computer programming dictum, ‘garbage in, garbage out’ (GIGO), and in case anyone today believes, that artificial intelligence/ machine learning has overcome that problem, here is AI’s own answer: “Poor-quality inputs, such as bad training data or vague prompts, will always result in flawed, inaccurate, or generic AI outputs. No matter how advanced the AI model is, it cannot fix bad foundations.”
In short, as the Arctic Monkeys, always praised for their authenticity, sang, “Snap out of it”!
In the investment world, one area where the advent of AI is having an increasing impact is fund administration which involves a series of repetitive tasks, which AI is ideally suited to undertake, such as calculating net asset value, managing fund accounting, handling financial reporting, plus more complex duties such as handling regular capital calls (commonly for private capital or alternative investment funds) and regulatory compliance.
Unsurprisingly, AI’s ability to cope with these and a range of other tasks, quickly, accurately, and economically, is changing the face of fund administration, placing increasing emphasis on individuals and teams capable of handling the unexpected (such as the series of high margin calls occasioned in July by the hits taken by shorts during the AI stock sell-off).
However, as the CIO of Schroders recently pointed out: “There is some productivity gain because these tools can do things that humans cannot. Our colleagues can now use these tools, but this is not at all about replacing people.”
This evolution within third party administrators such as JTC Group is simply a neutral response to a changing investment environment, but other changes occasioned by the presence of AI should not be viewed with equanimity.
See also: Asset managers to expand AI use for decision-making and data collection
For example, while fund managers, working in a highly competitive industry, are pleased that AI is potentially driving administration costs down, it is questionable whether managers taking the next step – abandoning external administrators altogether in favour of building their own inhouse AI-driven systems – are simply offering their investors a cost advantage or whether they are, additionally, taking on more risk.
It is here that the ‘bad foundations’ referred to above, may become apparent. The great strength of AI is its ability to develop via machine learning, turning massive amounts of data into numbers and finding mathematical patterns. To achieve a positive outcome from this process requires, as stated ‘massive amounts of data’, but, more particularly, massive amounts of diverse data – which may not be available from the experience of a single manager.
Moreover, it’s arguable that it’s better for investment managers to focus on the analysis of investment opportunities and leave (AI-driven) fund admin with the fund administrator. That’s where additional value for their investors will more likely be created. The fund administrators will invest in AI to improve fund admin (stay in your lane).
By comparison, AI-based systems built from the data inherent in long-established administrators, with hundreds, perhaps thousands, of investment management clients, are by definition more robust, and imbued with intelligence arising from handling with a multitude of different administrative needs arising from the employment of a wide range of investment strategies across a wide range of equity, fixed income, commodities and other markets across the globe.
A second concern arising from inhouse-built administrative systems based on AI, is the external confirmation of the numbers and the formal reports produced. Without independent, third party verification of, say, the efficiency and accuracy of the systems generating a fund’s net asset value, it is impossible for investors to place absolute faith in the price.
The verification issue is also a problem for internal auditors responsible for reviewing and maintaining a manager’s daily processes, but even more so for external auditors who are obliged to follow formal accounting rules to check, for example, a management company’s financial records and accounts to ensure they are accurate, legal, and honest. An auditor’s reputation depends on faithfully discharging this obligation – a difficult, perhaps impossible task when ‘poor-quality’ inputs are either suspected or, at least, possible.
A critical, albeit qualitative role undertaken by an external administrator, is the maintenance of the relationship between investment manager and investor which is, in large part, based on trust – particular in the case of active managers.
But such trust is easily and quickly eroded by, for example, the inaccurate reporting of figures, and it is therefore the responsibility of the administrator to ensure this does not occur but, at the same time, strive, in today’s fast-moving markets, to report and publish numbers as quickly as possible.
Speed is also critical for managers’ analysis of investment opportunities, and this requirement is proving to be one of the major drivers of AI uptake in the investment industry with a recent study reported by Institutional Investor saying 68% reported extensive or moderate AI use in investment research and decision making and more than eight in ten (83%) expect their use of AI in investment research to increase over the next two years.
One AI factor that is, however, negative as far as the investment management industry is concern, is a perceived tendency of investors to use similar models to interpret information and come to conclusions that may also be similar.
Just as input diversification is key to the building of robust and effective AI models, so, according to AI specialist David Trainer of New Constructs, is diversification of opinion. He says that when investors’ views are no longer independent and become the same, “that’s what causes manias and bubbles.”
This risk remains. As the Arctic Monkeys questioned, “Have you no idea you’re in deep?”
