Wouter Klijn, Director of Content at the Investment Innovation Institute [i3]

Wouter Klijn, Director of Content at the Investment Innovation Institute [i3]

Can ChatGPT Replace Your Investment Team?

New Study Shows Outperformance

Can ChatGPT replace human investors? A new study shows we are getting closer to that scenario.

Ever since the field of artificial intelligence (AI) emerged in the 1940s, investors have been trying to get computers to predict stock prices.

Until recently, these attempts failed to do so, at least consistently so.

But a paper released in September by academics from the University of Florida, titled “Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models”, claims that with the advent of ChatGPT 3.5, AI can now be applied to predict next-day stock price directions based on the news headlines of the day.

This sentiment-based technology is so successful, the authors of the paper claim, that when applying a long/short strategy – buying the stocks that rate positively in the analysis and selling the stocks that scored negatively – it can lead to hundreds of percentage points of outperformance.

“Without considering transaction costs, a self-financing strategy that buys the stocks with a positive ChatGPT score and sells stocks with a negative ChatGPT score after the news announcement earns a cumulative return of over 500 per cent from October 2021 to December 2022,” Alejandro Lopez-Lira and Yuehua Tang, Assistant Professors of Finance at the University of Florida, write.

Even when taking transaction costs as high as 25 basis points into account, the authors still found an outperformance of 50 per cent.

So what did they do?

Lopez-Lira and Tang asked ChatGPT to assess whether a headline was positive, negative or neutral for a particular company. For example, they used the following prompt to assess the impact of a news item about Oracle:

Forget all your previous instructions. Pretend you are a financial expert. You are
a financial expert with stock recommendation experience. Answer “YES” if good
news, “NO” if bad news, or “UNKNOWN” if uncertain in the first line. Then
elaborate with one short and concise sentence on the next line. Is this headline
good or bad for the stock price of Oracle in the short term?
Headline: Rimini Street Fined $630,000 in Case Against Oracle

They did this for all news headlines that day and assigned a 1 to positive, a -1 to negative and a 0 to neutral and added up the score.

“We document a significantly positive correlation between ChatGPT scores and subsequent daily stock returns,” the authors write.

The same is not true when testing earlier AI systems, including ChatGPT-1, ChatGPT-2 and BERT.

The question that arises is whether we are at the advent of robo-investing.

For an institutional investor, there are a number of hurdles to overcome before they could benefit from such a strategy. First of all, a strategy that trades many stocks on a daily basis will have very high turnover and, therefore, high transaction fees.

Although the authors address this question and calculate the impact of five, 10 and 25 basis points on returns, the issue is more complex for institutional investors. For example, large investors will have a greater market impact when trading than small to mid-sized funds, which makes their trades more expensive and it will take them longer to execute.

Small caps are also more expensive to trade than large caps, while the paper shows the correlation is more pronounced in small caps.

In addition, the experiment was conducted on US stocks, but it is unclear what the correlation would be with stocks in other markets.

[i3] Luncheon: A.I. & Investment Decision-making | Investment Innovation Institute

At [i3] Insights, we are also hearing anecdotal evidence that companies are starting to catch on to sentiment analysis and run their financial statements through optimisers to make sure they don’t score too negatively.

Yet, the study provides some food for thought. AI systems are becoming increasingly sophisticated and they will only become more so in the future. At some point, they might overcome the hurdles institutional investors face.

What then? Is this going to be a new form of investing? Will computers try to outcompete each other? Or will they always remain a tool that only assists the human investor in designing increasingly sophisticated investment strategies?

The research also highlights the role language and narratives play in stock price valuations, at least in the short term. In the long term, fundamental stock analysis is unlikely to be dethroned as an investment framework, but in the short term a story can spoil a good day.

Perhaps it is simply a trading strategy that over time will be arbitraged away as more market participants catch on, but it is certainly a space to watch.

Wouter Klijn


[i3] Insights is the official educational bulletin of the Investment Innovation Institute [i3]. It covers major trends and innovations in institutional investing, providing independent and thought-provoking content about pension funds, insurance companies and sovereign wealth funds across the globe.