Can AI Be Used to Manage Money? A Conversation between WisdomTree and Voya

We were fortunate to have the opportunity to speak with Chrissy Bargeron, Vice President and Client Portfolio Manager on the Equity team at Voya Investment Management. The conversation was structured around one simple question:

Can AI be used to manage money?

We recognize that at this moment in 2023, there is a good deal of hype surrounding ChatGPT, GPT-4 and generative AI. What’s clear is that the world is seeking to sort out the ways in which economic value will be extracted from these tools. It is certainly positive to see the different prognostications as to how much these tools may impact productivity, but we all have work to do in order to translate those into reality.

In the meantime, have these systems changed how asset managers are running portfolios?

The short answer: not too much, yet.

Following the release of ChatGPT and the uptake of large language models, it was more or less business as usual at Voya, where the team uses machine learning for stock selection.

Chrissy and I discussed how the process at Voya, of constantly seeking to improve the model, has been ongoing for years, well before ChatGPT’s release. And after those events, the team continues to use machine learning models for stock selection and consistently improving the model.

It is exciting to see the attention that ChatGPT has brought to AI generally, and it is true that many more people want to talk about it now. On the other hand, Voya’s approach incorporates much more than just a large language model on the back-end to manage the process, select the stocks and be responsible for the returns.

Introducing the Virtual Analysts

Chrissy discussed the more than 26 distinct ‘virtual analysts’ inherent to the process and helped to explain their role by saying they compete against each other to get their ideas represented in a portfolio, just like a team of human analysts might. Success for these multiple models—a virtual analyst is really an instance of a particular model—has been defined as showcasing a robustness for being able to select stocks that ‘outperform,’ which can be defined in absolute terms or relative to a benchmark.

Also like a team of human analysts, the system is constantly seeking to evaluate new data sources. Chrissy noted that sometimes data sources are found to not add significant value to the process, in which case they are removed. What’s important to have in mind is that the machine learning system is doing the vast majority of the evaluation of ‘what is useful’ with regard to data—it is not a committee of people mixing and matching. People are in the loop and people are monitoring, but it is not for people to say if a data source is or is not adding value.

The Warren Buffett Archetype

As we write these words, it is the week following the annual Berkshire Hathaway meeting of shareholders in Omaha, Nebraska. Warren Buffett’s track record has been so phenomenal that many of us look to his approach to see what we can learn. One of the things we see is the incredible cognitive feat of information processing. An enormous amount of reading, learning and remembering has allowed Buffett to make certain decisions and avoid other mistakes and, over time, create the track record that we all admire.

Admittedly, even if machines process information in a completely different way, the ‘Buffett Archetype’ gives us something tangible to look toward—maybe the way to think of AI is to assume that it no longer matters to read the ‘right’ stuff or to ‘remember’—now the cost of ingesting information and the cost of remembering information is being brought closer and closer to zero. Are these systems just ‘Buffett times 1 million,’ to put it simply?

What Voya is running today is not seeking to ingest all possible text—Chrissy noted that the system is attempting to only add and use things that are ‘additive’ and not over-fit the models. It is likely the case that the biggest asset in the system regards the data, as with these models it’s really ‘garbage-in-garbage-out,’ so if there is going to be value to be extracted, that input data is really the guide for this.

Know Your Biases

When seeking to understand how a portfolio manager is running a given investment process, it is important to note any perceived tendencies or biases that have accumulated over time. Do they tend to hold winners for an extended period? Do they tend to avoid certain sectors? It is interesting to analyze a machine learning model in similar fashion.

We were able to speak with Chrissy a bit about 2022 and the first quarter of 2023. From our discussion about a strategy focused on value:

• We know that energy was very strong in terms of performance through much of 2022. Value-oriented strategies that went more heavily in this direction would have done better, all other things equal.

• We know that growth stocks, particularly unprofitable growth stocks, were running into the wind significantly during 2022, but then these same stocks started 2023 with a strong rally. Certain value approaches, like those oriented toward the S&P 500 Value Index, did see the negative momentum in these growth stocks and add some of them to their approaches. Doing this has worked well, at least so far, in 2023.

Voya’s model tends to be a bit ‘contrarian’ in how it positions itself, which is neither good nor bad, but can be responsible for not positioning too heavily in energy when it was working strongly and also not positioning too heavily in growth-oriented stocks when they were working. Rather than chase the latest momentum trade, the system seeks out companies that exhibit fundamentally driven patterns that have historically outperformed. Over the short term, in a market being driven by particular macroeconomic announcements punctuated by large language models and generative AI, it may not be the ideal environment for this model, but these things can always evolve and change.

We note that looking at any strategy over short periods can be hit or miss, and that 1-3 months can be a tough lens through which to view this kind of approach. We always offer to help provide context when investors come to us wanting to do deeper performance comparisons, even if we recognize that no strategy always outperforms.

Bottom Line: WisdomTree Has a Lot to Add in the Current AI Discussion

If we are certain of one thing, it’s this, and Chrissy mentioned it repeatedly as the third wave in asset management: it is likely that what worked in the past is not what is likely to work in the future. Chrissy even mentioned the portfolio managers from the days of old with the physical financial statement books—not the most efficient way to gather information if one has machine learning systems sitting at the ready. When people think of artificial intelligence at WisdomTree, there are two primary paths:

• People may consider how the systems are used to extract value, such as by contributing directly to an investment process, in which case the WisdomTree U.S. AI Enhanced Value Fund (AIVL) is worth closer study.

• People may want to invest more directly in AI tools and software, in which case the WisdomTree Artificial Intelligence & Innovation Fund (WTAI) may be the closer fit to the goal.

In this rapidly evolving landscape, we’ll be coming back often with updates as to how these different paths are unfolding.