Market insights and analysis

By Will Hubbard

ChatGPT is a new artificial intelligence (AI) tool that has the potential to disrupt the investment community. It can help brainstorm new ideas, summarize research, generate code, and many other use cases that we’re only just beginning to learn about or imagine.

In light of all of this potential, I was excited to test it out to see how helpful it could be with some of my tasks as a financial professional. Can it speed up data testing—or create a thesis for a new investment indicator or process?

I started by asking it to perform some simple math.

It’s pretty good if you ask it to calculate something like 2+2. It’s less good at solving word problems and multistep arithmetic. For example, here’s how it responded to a word problem I posed:

Not great.

Then I gave it a pure math problem and the answer.

As you can see, the first method results in 76, not 74. And ChatGPT changes the problem entirely to fit the narrative in the second method.

Again, not great.

Can AI design your investment portfolio?

Given what I just witnessed, I wasn’t sure how ChatGPT would handle something more complicated—like designing an investment portfolio.

ChatGPT was trained on billions of parameters using deep-learning AI techniques to generate human-like text. So I thought if I trained it a bit—fed it some relevant financial research and data—then maybe I could get it to parse fact sheets and descriptions of various mutual funds, ETFs, and stocks to understand how to make a risk-based investment portfolio. No math involved—just pure reading comprehension.

I started with a high-level explanation of who I am as an investor:

Nice work, ChatGPT! This investment portfolio meets all of my requirements; is globally focused; and is diversified across stocks, bonds, commodities, and real estate.

This is a nice place to start, but it’s a bit generic. So I asked ChatGPT to be a little more specific and create the portfolio again, but this time with funds that are not market-cap weighted. This isn’t a particularly challenging task for a trained professional, but I wanted to see if ChatGPT could distinguish between investment methodologies at a very basic level: market-cap weighted versus non-market-cap weighted.

Not a great result.

I gave ChatGPT specific information about how I felt about risk, and while its suggestion is technically accurate, it does not accurately represent the targeted portfolio.

If I was new to investing, this might look good at the outset. But someone with more knowledge would quickly see that the result is completely undesirable.

ChatGPT’s recommended funds:

• 40% global equity ETFs (exchange-traded funds):

◦ First Trust Global Wind Energy ETF (FAN)

◦ WisdomTree U.S. LargeCap Dividend Fund (DLN)

• 30% bond ETFs:

◦ Franklin Core Balanced Active ETF (FLBA)

◦ iShares Core Aggressive Allocation (AOA)

• 20% factor-based ETFs:

◦ Hartford Multifactor REIT ETF (RORE)—Closed in August 2020

◦ Fidelity MSCI Real Estate Index ETF (FREL)

• 10% commodity ETFs:

◦ United States Commodity Index (USCI)

◦ ProShares UltraShort Gold (GLL)

Skimming the results, a few things are clear: The global equity ETFs are not exactly “global,” and the bond ETFs are more like asset-allocation ETFs that ChatGPT picked simply because they contain bonds. The final “whoa” moment is the recommendation for GLL, the ProShares UltraShort Gold position. GLL’s main goal is to be 200% short the daily return of gold—not exactly a great recommendation for a request that amounts to a strategic asset allocation.

Future iterations of ChatGPT might sort these issues out, especially as the training data grows and becomes more current, allowing for better conclusions to be drawn. This is just an example of how AI can really fail to capture the heart of a simple investment question using only text, no performance relationships.

Humans: The link between today and tomorrow

AI is and will continue to be incorporated across all industries in a variety of ways, advancing on methods already in use. The main considerations are: How well do the people using these tools know and understand the inputs relative to their outputs? And how well do people know and understand the limitations of the AI (for example, in the case of ChatGPT, that it can be terrible at basic math)?

For now, using the outputs effectively in the investment world still requires humans to see how everything works together for the benefit of the investor.

Humans are the link between the iterative “spaghetti” that AI throws at the wall and managing the real risk that exists in investor portfolios. This is where quantitative investment professionals will continue to add value for investors. We develop the quantitative models that seek to address the rhyming of the past with the uncertainties of the future—and all while incorporating new technologies like ChatGPT into the process.

In conclusion, we should all be mindful of what AI spits out. It could be the next best thing, completely inaccurate, or accurate but irrelevant.

If you’re looking for help with your investments, seek out knowledgeable professionals that are mindful of history and the role it plays while looking for ways to adopt new and emerging technologies to produce more robust portfolios. Humans continue to be the bedrock investing technology and the foreseeable future of finance.