UC Merced Undergraduate Research Journal

In November 2022, the world of artificial intelligence, programming, and efficiency changed forever, as OpenAI created the first-ever publicly accessible large language generative chatbot: Generative Pre-trained Transformed (GPT)-3.5 (Open AI, 2022). The bot passed several Advanced Placement course exams, which are tests high school students can take to obtain college credit. It passed graduate-level exams such as the GRE, and even the BAR exam required to become a professional lawyer (Open AI, 2023). With all of GPT’s success, the specific issue with OpenAI’s model, GPT-3.5, is that it cannot access the internet or fetch real-time data (OpenAI, 2022). The challenge we undertook was to use the GPT chatbot to create a stock-prediction trading algorithm, guiding the model to provide a conclusive output, and limiting our influence on the model—outside of errors—as much as possible. From November 2 to November 17 of 2023, we manually compared GPT’s predictions to the actual results of ten stocks within a trading day (6:30 AM PST – 1:00 PM PST). It has been widely concluded in the past that GPT models are unable to make daily stock predictions accurately. Past researchers suggest, “It is too soon to claim AI can beat the stock markets” (Mokhtari, 2021) as they perform better in the long-term, which is why we are testing their short-term and long-term capabilities. From the tests we ran to evaluate GPT’s capabilities, we conclude there is great value in incorporating artificial intelligence into current trading models. However, generative AI models like GPT cannot be solely relied upon for accurate predictions. Models built with AI can help advise full-time stock traders, casual investors, and large trading firms about the effectiveness of AI models in their technical analysis before investing in a stock.