The growing appreciation that human stockpickers struggle to outperform their benchmark indexes has helped fuel a massive surge in assets held by passively managed exchange-traded funds (ETFs). Now, some companies are hoping to show that artificial intelligence (AI) can finally give them an edge.
The technology is fast-evolving but at least two fund managers, San Francisco-based EquBot Inc. and Qraft Technologies Inc. of South Korea, running dedicated AI-powered ETFs are claiming early success, even though some of their AI models’ decisions might have required strong nerves to implement.
For example, the team at Qraft, which offers four AI-powered ETFs listed on NYSE Arca, witnessed its technology build a weighting of 14.7 per cent in Tesla Inc. TSLA-Q in its Qraft AI-Enhanced U.S. Large Cap Momentum ETF AMOM-A in August last year, but when it rebalanced a month later on Sept. 1, it sold it all.
The ETF began buying Tesla’s stock again in November, amassing a stake of 7.6 per cent by January of this year, but in the February rebalancing it sold the entire holding once again. In each case, when it sold, it anticipated a sharp decline in the price of Tesla and was able to profit from the subsequent rise when it bought back in.
“Alpha [excess returns above the market] is getting harder and harder to find,” says Francis Oh, managing director and head of AI ETFs at Qraft, pointing out that humans can become attached emotionally to certain stocks, impeding their portfolio returns. “Our model has no human bias.”
Academic research has certainly shown that humans tend to be reluctant to crystallize losses, while conversely they feel driven to realize gains – sometimes too early.
However, it’s arguable whether the AI systems Qraft and EquBot use can really be said to eliminate human bias because both are supported by large teams of data scientists who are constantly enhancing their models – EquBot has teamed up with IBM Watson and Qraft has its own dedicated team in South Korea.
“The machine only has historical data. It sees opportunities according to the rules it has been programmed for,” points out Greg Davies, head of behavioural science at consultancy Oxford Risk in London.
Chris Natividad, chief investment officer at EquBot, agrees.
“The system only knows what it knows, and it’s historical,” he says, adding that in addition to humans deciding what new information the self-learning system should be given, data scientists also needed to check the outcomes “so we can explain it to investors.”
Similar successes have been notched up by Qraft’s suite of AI-powered vehicles, but the outperformance narrative is less clear when compared to the plain vanilla SPDR S&P 500 ETF SPY-A so far this year.
Qraft AI-Enhanced U.S. Large Cap Momentum ETF and Qraft’s other ETF – Qraft AI-Enhanced Large Cap ETF QRFT-A, Qraft AI-Enhanced U.S. High Dividend ETF HDIV-A and Qraft AI-Enhanced U.S. Next Value ETF NVQ-A delivered returns of between 15.3 per cent and 20.8 per cent in the eight months ended Aug. 31. While respectable, none of them quite matched the 21.6 per cent return of SPDR S&P 500 ETF over the same period.
EquBot’s AI Powered Equity ETF also just undershot SPDR S&P 500 ETF, notching up a 21.3 per cent gain in the same period, while AI Powered International Equity ETF delivered only 12.2 per cent.
Despite the unremarkable returns this year, Mr. Oh and Mr. Natividad remain convinced that their models have much to offer.
“The velocity, variety and volume of data is exploding,” says Mr. Natividad, adding that bringing in new data sources was a bit like adding more pixels to an online image. “You get a clearer picture.”
He says asset managers and index providers were embroiled in an arms race for data.
Mr. Oh says value and momentum factors were becoming so short-lived and fragmented that AI systems helped to find opportunities.
EquBot scours news, social media, industry and analyst reports, and financial statements to build predictive models. It also looks at things such as job posts. Qraft also uses a variety of so-called structured and unstructured data sources to drive its models.
Still, despite the promise of the technology, assets under management (AUM) for the ETFs at both companies remain modest. EquBot’s ETFs have less than US$200-million in AUM between them, although Mr. Natividad says a partnership with HSBC Holding PLC, which was using the two EquBot indexes, meant that there was US$1.4-billion tracking the strategies.
Qraft’s ETFs have attracted less than US$70-million across all four vehicles, although Mr. Oh says the business model was once again focused on business-to-business advisory asset-allocation modelling.
Rony Abboud, ETF analyst at TrackInsight SAS, says investors probably wanted to know more. He emphasizes the importance of due diligence, noting that “the more data points used, the higher probability of having an error. So, where do they get their data from and how accurate is it?”
Despite misgivings, adoption of AI techniques in the investment world is increasing.
“Natural language processing (NLP) is certainly a growing area,” says Emerald Yau, head of equity index product management for the Asia-Pacific region at FTSE Russell, which has made its first foray into NLP-powered offerings with its launch of a suite of innovation-themed indxes.
However, Oxford Risk’s Mr. Davies warns that while algorithms were good at finding arbitrage opportunities they could not deal with ambiguity.
“The problem with the investing world is the rules are not static,” he says, adding that humans still retained an edge. “If humans learn one thing in one context, they can transfer it to another.”
© The Financial Times Limited 2021. All Rights Reserved. FT and Financial Times are trademarks of the Financial Times Ltd. Not to be redistributed, copied, or modified in any way.