Advisers, fund managers seek balance between human insight and AI innovation
In the excitement and frenzied reporting around the 2024 United States (US) Presidential Election, two remarkable developments slipped by almost unnoticed. For the first time ever, 10 US-listed companies surpassed USD1 trillion in market capitalisation simultaneously. And M&G Investments hosted an unparalleled discussion on the practical application of artificial intelligence (AI) in portfolio management, at a level of detail this writer had not experienced before.
Investing at an innovation inflection point
Gautam Samarth, a fund manager at the global asset manager, started the discussion by describing a shift or inflection point beyond using innovation to solve existing problems towards leveraging it to do ‘genuinely new things’. “Today you are trying to use Chat GPT to deliver productivity gains such as making your PowerPoint presentation better or crafting an email; the real breakthrough will be when you are able to [use AI to] do things that you were not able to do before,” he said.
In the case of the Internet, businesses went from simply replicating what they did using facsimile to reimagining business processes and systems for an Internet that offered global connectivity and served as a single access point to an entire world of information. AI has been around for decades, as has machine learning; but the rapid progress in areas like computing power and data storage has finally brought these technologies into the mainstream.
“The AI world is extremely broad; it is really opening up other avenues in terms of the way that we think about processing data and extracting information from that data,” said Michael Cook, co-portfolio manager on the M&G Global Equity fund. He joined Samarth and Hamilton van Breda, M&G Investments Group Liaison Officer, in a three-way question and answer session titled, ‘Do not forget about the (human) intelligence in AI’.
“AI allows us to access useful information from within text, image and sound data, and that really helps us to enhance our investment decision making,” Cook said, adding that AI was useful in comparing and layering different data sets and deciphering the conditionality between them.
Three AI-centric questions for fund managers
The commercial generalised language models (GLMs) that have risen to prominence recently excel at giving general answers; but according to Samarth, these models are just a starting point. Asset managers fine-tune them using reams of data to ensure that they are informed enough to give specific, useful answers.
Van Breda then posed a three-part questions that many South African financial advisers wrestle with: First, how does a fund manager integrate AI into the investment process? Second, what AI technologies are used? And third, how does AI-backed machine learning differ from traditional investment strategies?
“AI really sits front and centre of everything that we do; our investment process is driven entirely by AI-based decision making,” Cook said, before launching into a basic explainer of the process. In broad strokes, he described taking historically representative data of the investment landscape; exposing AI models to this data; and allowing these models to learn the underlying patterns in the data. You then apply the models predictively to inform your investment choices and / or enrich your understanding of how different facets of financial markets work together.
Data plus machine learning to predict price
“There are various approaches to AI; the one that we are primarily using is a technique called machine learning (ML) based processes [which is] a supervised ML technique,” Samarth continued.
In layman’s terms, the asset manager identifies a set of data that it believes is relevant to predict outcomes; in this case share prices. “The data sets that we are using are those that fundamental analysts and investment professionals have identified as being relevant to the fortunes of the companies that we are looking to generate price predictions on,” he said. The model then makes sense of patterns between that underlying data and historic share prices, using this understanding to predict future moves.
In answering the third part of this question, the experts revealed that the much-hyped AI-based share selection machinery hinged on repeating what you get in traditional investment processes. As Samarth explained, “You first collect a lot of data on the companies that you want to potentially invest in; you model that data; you use that modelling to inform your investment decisions; and then you construct a portfolio of the best companies and ideas based on those underlying models’ forecasts into the future.”
Machine learning replicates this traditional process at scale, removing pesky biases in the process. To illustrate scale, M&G said it used to apply the above process across a universe of around 200 style-relevant companies compared to an investment universe of thousands of companies today. Of greater importance, AI allows the asset manager to run these models across this expanded universe daily, and in real time.
“Today, our investment universe is literally the world. We compare about 10 000 companies on a daily basis, [making] predictions in real time,” Samarth said. “Every one of those predictions is directly comparable to the other, creating a rank or order of preference.”
Accommodating human experience and skills
Human portfolio managers still have a part to play as a final ‘layer’ over the AI ranking, identifying any areas where the machine may have tripped up. Human intelligence is also involved in the data modelling part of the process, as Cook explained: “We use the ML techniques with supervision to basically learn all of the underlying relationships in the data; we train a model to give us a very rich view of the world and use that to inform our investment process.” The human portfolio manager still has to curate the sentiment, technical and valuation data to ensure the data fed into the model supports the right level of economic intuition.
The human versus robot debate has troubled financial advisers for some time. In fact, financial advising was once touted as the first area where AI and automation might disrupt adviser-client relationships and advice business models. Human advice has since triumphed over chatbots and other AI-backed Fintech start-ups; and it seems apparent that human investment advisers and portfolio managers will prevail too.
“From the very beginning, we emphasised that this was not a human versus machine game; the real benefits accrue when you can harness both sides of the equation,” Samarth said. Machines are good at identifying patterns and processing reams of data efficiently, without emotion; humans help by picking up blind spots in the modelling including anomalies in the data or environmental shifts due to regulation.
A rational, unemotional investment process
Asset managers still have to assess and unpack AI investment decisions before investing. According to Cook, his team has learned two important lessons over time. The first is that the model “sometimes puts on recommended trades that human traders might find slightly uncomfortable emotionally.” The second is that “models are based on a rational, unemotional investment process [that can unveil] stocks that are not on other people’s radars.” He backed up these points using the model’s recent and successful decision to initiate a position in a biotech company despite all the price history suggesting the share was in a downward trend.
Van Breda wrapped up the discussion by asking his portfolio manager colleagues to share their key AI observations with the audience. “AI is here to stay because it is a good tool to sift through the volume of data required for investment decision making,” Cook concluded. “AI needs to be handled delicately, and expertise needs to be employed at all levels of data processing to ensure optimal outcomes.” His remarks were echoed by Samarth, who also reminded the audience of the potential for diverse outcomes using AI tools on data sets.
The process (not the AI) determines the outcomes
“The process that you utilise will deliver either suboptimal or optimal outcomes; we have a five-year track record of using ML-based techniques to identify stocks that can outperform the market, and we have been very successful across all the regimes we have spoken about today,” he concluded.
Writer’s Thoughts:
Artificial intelligence (AI) can enhance an asset manager’s investment strategy; but the human element remains key in validating machine-based decisions. Are you clients comfortable with investing their cash in funds that lean heavily on AI to choose shares? Please comment below, interact with us on X at @fanews_online or email us your thoughts [email protected].
