Fueled by breakthroughs such as ChatGPT, AI is more ubiquitous than ever — and making substantial inroads into financial services.
As adoption grows in the world of financial services, including exchange traded funds (ETFs), more and more advisors may need to explain not just how (or if) they use AI-based products in client portfolios, but how the underlying technology works. As is often true with novel technologies, misunderstandings abound. This has not been helped by numerous popular culture comparisons of ChatGPT (Chat Generative Pre-Trained Transformer) to HAL, the human-like computer made famous in Stanley Kubrick’s 1970 sci-fi epic, 2001, A Space Odyssey. (A search for ChatGPT + HAL yielded 19.7 million Google results and growing by mid-February.)
Arthur C. Clarke, the author of the book on which that movie was based, is famously quoted as saying, “Any sufficiently advanced technology is indistinguishable from magic.” But ChatGPT is neither magic nor HAL. At its heart, AI is basically a set of rules designed to solve a problem. Those rules are expressed through a mathematical formula known as an algorithm, a process that is the opposite of magic. The algorithm is designed to execute a series of steps defined by its creators to reach a conclusion based on an analysis of a dataset constructed by its human operators. It is in this way that AI can identify potential investment opportunities.
Despite its many strengths, however, AI still struggles with some of the more complex aspects of financial decision-making, such as long-term forecasting or handling unusual market conditions. It’s also no match for human financial analyst’s creativity and intuition when it comes to identifying new opportunities or anticipating shifts in the market. Lastly, AI still lacks the nuanced understanding of human behavior necessary to make more subjective financial decisions, such as assessing the creditworthiness of individual borrowers.
“AI is not good at context and nuance,” said Ryan Pannell, CEO of Kaiju ETF Advisors, during a recent VettaFi webinar. “For instance, there is no way for AI to effectively determine what the economic response to a geopolitical decision is going to be. That’s not in its bailiwick.”
Despite those limitations, AI has proven to be a valuable tool in many areas of the financial industry. A growing body of research (eg, "Artificial Intelligence Machine Learning and Big Data in Finance" by the OECD) suggests that AI performs best when the job is highly targeted and the time frame is short. By focusing on specific tasks and time periods, AI algorithms can be optimized for accuracy and speed, with the potential to lead to significant gains in productivity and efficiency.
“Machine learning,” another term commonly heard in these discussions, is effectively a subset of AI. Machine learning allows the system to improve its decision-making ability based on observed patterns and relationships in the data. Performance is continually optimized with new incoming data. When multiple machine learning models are combined with expert systems that classify or limit outputs to predetermined boundaries, the resulting outcomes can simulate certain aspects of human behavior, leading to the term "artificial intelligence."
In fact, in cases such as we are discussing here, the AI’s performance is not unlike what a portfolio manager might do, just greatly accelerated, and based on a much larger set of data than an investment management team could reasonably manage in that time frame.
A big misconception: AI is not a technology that is turned loose on the market and told to “go find winners.” The likelihood of uncovering misleading “signals” is massive. There are so-called “black box” applications where it’s unclear how the AI determines an investment decision, but we don’t believe this represents the best approach for an ETF. We think advisors and their clients want transparency and performance attribution. In our case, because we are primarily looking for patterns and relationships within a discrete data set, we believe it is possible to identify the factors that dictate decisions to buy and sell.
Doing this means having known rules and applying those rules to a well-defined dataset. Both these pre-conditions can be addressed by the algorithm’s creators. The AI performance will depend in part on how well these parameters are defined. One final point to keep in mind – if the AI is operating within a mutual fund or ETF structure, it is required to adhere to the same SEC regulations as any other registered fund, including those involving transparency and portfolio concentration, among others.
AI algorithms can use historical market data to build predictive models that can identify patterns and trends in stock prices. By analyzing data, such as stock prices, trading volume, and other indicators, the algorithms can help identify stocks that are potentially oversold and make predictions about future performance.
- AI (as we employ it) is a mathematical process designed to improve and accelerate investment decision making in an ETF.
- We do this primarily through data analysis and pattern recognition – like humans, only faster and with access to a much bigger dataset.
- As we apply it, AI is not a “black box.” It operates according to rules defined by its creators (us). As it is put to work in an ETF or mutual fund, it is governed by regulations established by the SEC and FINRA.
AI is a big idea but it’s more straightforward than it might first appear (at least, as it’s used in our corner of the asset management world). It may be best explained as a way of making more informed investment decisions, faster, in a way that can potentially improve investment performance. We think that makes it an application worth exploring.