����ר��

Over the last two decades, the asset management industry has been using artificial intelligence (AI) more, but to what effect? From novel research, Finance Professor Xiaowen Hu of ����ר�� Cox explains, "We document an increasing trend of mutual funds and the asset management industry using advanced technologies in forming their portfolios and improving their performance for competitive advantage." Hu and coauthors reveal that utilizing AI and machine learning in the industry requires both sophistication and trading skill—that of human and physical capital. She also parses AI-marketed funds and trends about AI talent. 

The extent to which the industry integrates machine learning into trading strategies remains mostly unknown. One significant contribution of the study was developing an empirical method to detect which funds are genuinely using AI in their investment processes, beyond just marketing claims. "Our measure is novel in that we systematically study how asset management firms adopt AI technologies and skills; how that happens over time; and how it affects performance," Hu explains. It is a more rigorous comprehensive study than just handpicking cases from prospectuses or fund names.

The paper looks specifically at the AI subset of machine learning, synonymous with algorithmic trading. Hu explains, “When people talk about machine learning, the stereotype is, ‘Oh, that's a big black box. I don't understand what's happening in the box.’” In simple terms, machine learning equates to computers learning to process information like the human brain. 

In the 2010s, a paradigm shift in AI adoption within the industry was driven by increases in big data and advancements in deep learning. Rather than focusing on the most cutting-edge or complex AI (or machine learning) models, the research examined realistic, practical applications that would be feasible for funds to implement. Hu notes, “We developed an average model that realistically could have been used in the past two decades.” Neural network technology [of machine learning] dates back to the 1960s to 1980s period. Hedge funds were early adopters in the ‘80s to ‘90s, with incentives to promote their use of machine learning and algorithms. 

The mutual fund industry, the study’s focus, differs from hedge funds because they are more regulated with less incentives to disclose their trading methods, a competitive advantage. The researchers feasible and realistic machine learning model analyzed how mutual fund firm-level holdings aligned with trading signals that show AI use.

Power of Knowledge

Machine learning is particularly valuable for asset management because of its ability to process vast amounts of diverse data that traditional models simply cannot handle. "With natural language processing, machines can synthesize huge text files like financial filings (eg. 10Ks, quarterly reports, 8Ks), earnings call transcripts, and firm-level news," explains Hu. "All this added information wasn't feasible to analyze in earlier days."

The research demonstrates that this information advantage is a key driver of performance: "We show that it's the rich information input that really helps drive performance," Hu surmises. Interestingly, findings show that successful AI implementation in fund management isn't about replacing human judgment but enhancing it. 

Both the information and the ability of machine learning algorithms to handle complex patterns and interactions together drive superior performance, according to Hu. "It really requires manager skills to decide what to trade based on machine output,” she relays. “If you only follow machine learning signals, you indeed fall into trading with high turnover and high costs that chew up performance." The study analyzed the universe of publicly-listed firms on NYSE, AMEX and NASDAQ from 1987 to 2022, and underlies their machine learning-based trading signals of detection. This meant a universe covering 18,803 stocks, in other words, crunching massive amounts of data.

Really AI?

People might question how the research determined whether self-designated "AI funds" were actually using AI. “We looked at the titles of funds that claim they are AI funds and scrutinized them to see if our measure aligns with them,” Hu relayed. The study's methodology focused on actual trading patterns and portfolio holdings to identify genuine AI implementation.

While the growth of funds hiring AI talent is increasing, raw hiring numbers don't necessarily translate into effective implementation. "Many funds hire AI-skilled people for marketing or to improve their customer platform rather than for trading," Hu points out. 

Solely considering AI hiring is too noisy and cannot really predict performance. In our paper, if firms are hiring AI talent, it has to be relevant to trading and performance, says Hu. “We're not seeing as many people with those AI skills,” she adds. Funds that effectively combine machine learning capabilities with human expertise are positioned to outperform. Our measure is empirically trade-based and holding-based, thus differentiating performance, Hu clarifies. 

The paper notes that AI talent with specific expertise of asset management was less than 1% in the early 2000s and now it's just under 2%. “You don't need a lot of employees working on AI,” Hu says. “Most employees don't need to deal with it—you just need a few experts if you're really doing it.” She says the trend of doubling AI talent is more informative than the absolute numbers.

Effectively using AI

Smaller funds can more easily adopt AI and have more incentives to do so, the authors say. Smaller funds can focus their AI applications in areas where they have specific domain (or investment) expertise. However, this doesn't mean large funds aren't benefiting from AI. "Our paper is not saying that large funds aren't using AI, Hu says. “We're showing that on the aggregate level, we see a stronger effect with relatively smaller funds."

The researchers suggest the trend of AI adoption is more informative than the current scale of implementation. The research demonstrates that simply following machine learning signals without human intervention can lead to high-turnover strategies and prohibitive trading costs. The most successful approaches combine the integration of AI capabilities and human expertise.

“Our measure is novel in that we can systematically study how asset management firms adopt AI skills, how that happens over time, and how it affects performance,” Hu concludes.

“Active Machine Learning Based Trading and Mutual Fund Performance” is a working paper by Xiaowen Hu of Cox School of Business, ����ר��; Maximilian Rohrer of Norwegian School of Economics; and Hanjiang Zhang, Washington State University. 

Written by Jennifer Warren.