AI Challenges

Stockholm (HedgeNordic) – Artificial Intelligence may have found its way in almost all industries, asset management being no exception. While many are talking about the possible applications of AI in asset management and the advantages these applications bring, the challenges faced by those working with AI-assisted investment processes and strategies are a less chartered terrain. HedgeNordic sheds some light on this topic by talking with Nordic managers about some of the challenges they faced when designing, running and, equally important, marketing their AI-assisted investment strategies.

Patrik Säfvenblad, CIO at Volt Capital Management

“For Volt, machine learning and artificial intelligence is a tool. It does not eliminate any of the normal challenges of investing, but it makes our process more precise and more robust,” Patrik Säfvenblad, chief investment officer at Volt Capital Management, tells HedgeNordic. “Once you have understood the characteristics of your machine learning algorithm, it is not inherently more complex than any other investment process,” says Säfvenblad, who oversees a systematic macro fund that uses machine learning and fundamental data to capture price moves across markets. However, the process of designing and understanding a machine learning algorithm may face its own unique set of challenges.

“Once you have understood the characteristics of your machine learning algorithm, it is not inherently more complex than any other investment process.”

Ville Rantanen, Portfolio Manager at Mandatum Life

“The biggest challenge is to find talent,” reckons Ville Rantanen of Mandatum Life, who oversees artificial intelligence-assisted systematic Mandatum Managed Futures Fund. “Value is created where experience and expertise are combined. One truly does need the exceptional talent that understands investing and AI to make this work,” argues Rantanen, a portfolio manager at Finnish insurer Mandatum Life. “We would expect that large asset managers and niche players dedicated to AI will have AI capabilities. However, for the vast majority of asset managers, the key roadblock is the lack of talent,” continues Rantanen. To tackle this challenge, Mandatum Life has partnered with Gradient Systems, the “AI for Finance” spin-off of Cambridge-based PROWLER.io to launch its managed futures fund.

“We would expect that large asset managers and niche players dedicated to AI will have AI capabilities. However, for the vast majority of asset managers, the key roadblock is the lack of talent.”

Martin Estlander, Founder of Estlander & Partners

Martin Estlander, the founder of a pioneering firm in the systematic investment, asserts that the low signal-to-noise ratio in financial markets is a key challenge in running an AI-assisted quantitative strategy. “We have worked a lot with different AI-related methodologies over the years,” Estlander tells HedgeNordic. “The biggest challenge is the fact that the signal-to-noise ratio in financial markets is low, which makes the problem complex and increases the risk of finding relationships which are irrelevant in the future,” he points out. “The other fundamental issue is that once you trade on an observed relationship, you actually affect it with your own presence in the market, making it even more unstable.”

The challenges associated with implementing and running an AI-assisted process, however, do not undermine its benefits. “Although we find the use of AI to come up with trading signals challenging, we have found value in using AI to identify environments when a strategy will deliver and when it will not,” emphasizes Estlander. “AI has also been useful in providing input for our research process, in coming up with alternative ways of looking at problems.”

Opening the Black Box

The ability to interpret and explain artificial intelligence-assisted models is turning out to be a defining factor for the acceptance of these models, both among fund managers and their investors. The key barrier preventing AI from going mainstream in the asset management industry is the “black box” problem of artificial intelligence. Even large, educated institutional investors have singled out the “black box” worry as a key hurdle to accepting AI for investment solutions.

“Investors are often concerned that AI means investing in a “black box” and worry that it might “go rogue” or its starts trading alone insanely.”

“Investors are often concerned that AI means investing in a “black box” and worry that it might “go rogue” or its starts trading alone insanely,” points out Ville Rantanen. He sees three main types of investor perceptions of artificial intelligence. “First, there are investors who see that AI is a way forward for the industry; then there are investors who are skeptical towards the value-add of AI in an investment fund; and investors who consider AI only as a marketing buzzword and, hence, ignore it.” According to Rantanen, the second type of investors “can be convinced by explaining design principles, architecture, and risk limits in-built into the AI systems.”

“AI is typically deemed a black box, which is hard to explain. Some investors are keen on the topic being quite hot, others are very skeptical.”

Martin Estlander of Estlander & Partners shares Rantenen’s views. “AI is typically deemed a black box, which is hard to explain. Some investors are keen on the topic being quite hot, others are very skeptical,” Estlander tells HedgeNordic. “The lack of understanding due to the opaque nature of the strategies, and the lack of trust in machines over humans” can make investors less comfortable investing in AI-assisted investment funds. The ‘explainability’ problem of AI does not only represent a challenge for prospective investors but also for the architects of the AI-assisted strategy, reckons Estlander. “If an AI-based strategy is complex to understand, will the manager have the guts to stick with it when he should or understand to cut it if it stops working?” asks Estlander.

“We need to convince investors that AI is doing optimal things and win investor trust.”

One approach to stave off investor worries about AI’s “black box” problem and sometimes lack of interpretability is increased transparency about the broader strategy and risk management tools. “We need to convince investors that AI is doing optimal things and win investor trust,” argues Rantanen. Patrik Säfvenblad of Volt Capital Management points out that “the only way we have found to convince investors is to be transparent.” Säfvenblad goes on to say that “we protect our intellectual property, but try to be as transparent as possible.”

Effective risk management also plays a pivotal role in AI acceptance by the investor community. “We employ a ‘Glass Box’ approach that allows us to trace back the factors behind every trade and every position,” explains Säfvenblad. “This makes it easier to explain why our strategy should work, and it also makes it straightforward to show how risk management kicks in and reduces losing positions,” he continues. “Looking back at our own experience in sales, our clear risk management philosophy has been key in convincing investors to commit,” Säfvenblad tells HedgeNordic.

“A systematic strategy or any quantitative modelling should be pictured more like an engine with individual components that are aligned with each other to maximize the overall performance.”

To dissuade worries about AI’s ‘explainability’ problem, Rantanen reckons that “a systematic strategy or any quantitative modelling should be pictured more like an engine with individual components that are aligned with each other to maximize the overall performance.” Rantanen goes on to say that “each AI component has its inputs and outputs as well as a “reward” function measuring its independent performance.” The role of a portfolio manager overseeing an AI-assisted vehicle is “than analogous to a portfolio manager with a team of human analysts: define their individual roles, define what good and bad looks like and which rewards can be obtained, as well as give them constraints and limits to what they do,” explains Rantanen.

Performance is Key

“The vast majority of client feedback we received was positive or neutral towards the idea of using AI in an investment fund,” says Rantanen about how Mandatum Life clients perceived the firm’s AI-assisted systematic strategy. “In the end, performance in terms of returns and achievement of the targeted risk/reward profile is what matters most,” he emphasizes. “But as good performance on AI-assisted fund continues, more money starts flowing in, and this ultimately brings more market share for these types of strategies.”

“The proof is in the pudding. Once there are enough examples of AI strategies consistently outperforming their non-AI counterparts, I think people will slowly come to accept them. Until then, there is bound to be a certain amount of skepticism.”

Estlander agrees, saying that “lack of proven success is probably the biggest issue” that explains why some investors are not embracing AI-assisted strategies. “Lot of talk over the years, but little proof,” he asserts. “The proof is in the pudding. Once there are enough examples of AI strategies consistently outperforming their non-AI counterparts, I think people will slowly come to accept them. Until then, there is bound to be a certain amount of skepticism.”

This article featured in HedgeNordic’s report “Technology and Hedge Funds.”

Photo by Rod Long on Unsplash

Share.

About Author

Eugeniu Guzun serves as a data analyst responsible for maintaining and gatekeeping the Nordic Hedge Index (NHX), as well as being a novice columnist covering the Nordic hedge fund industry for HedgeNordic. Prior to joining HedgeNordic, Eugeniu had served as a columnist for a U.S. journal covering insider trading activity, activist campaigns and hedge fund moves. Eugeniu completed his Master’s degree at the Stockholm School of Economics in 2018.

Leave A Reply

Time limit is exhausted. Please reload CAPTCHA.