GPT-4 Outperforms Human Analysts in Predicting Company Earnings

Researchers recently examined if OpenAI’s GPT-4 could analyze financial statements as effectively as human analysts. Surprisingly, GPT-4 outperformed human analysts in predicting changes in company earnings and matched the performance of advanced machine learning models, even when given only raw financial data without context.

The study used standardized financial statements without company names or dates, guiding GPT-4 through analysis with special prompts to simulate human reasoning. Comparing GPT-4’s predictions with those of human analysts from the IBES database, GPT-4 achieved a 60.35% prediction accuracy, surpassing the human analysts’ 52.71%. GPT-4’s F1-score also exceeded that of human analysts.

Without any textual information, GPT-4 could still accurately analyze and synthesize numerical data, thanks to step-by-step prompts that helped it identify trends and compute financial ratios. GPT-4 excelled particularly with small companies and those with volatile earnings, areas where human analysts often struggle.

Moreover, GPT-4’s performance was comparable to, and sometimes better than, advanced machine learning models like artificial neural networks (ANNs). The study suggests that combining insights from GPT-4 and human analysts could enhance prediction accuracy.

The findings imply that GPT-4 could assist analysts by making quicker, more accurate predictions, handling large data sets efficiently, and potentially streamlining the financial analysis process.