Reducing emotional bias in investment decisions: the role of GPT-4 in financial analysis

Purpose This study examines the investment recommendations made by AI models (GPT-4) in two versions (V1 and V2) compared to human analysts. The research investigates whether AI provides a more objective evaluation with reduced emotional and cognitive biases, addressing concerns about human analysts’ susceptibility to market sentiment. Design/methodology/approach Using Probit regression analysis, this study compares the investment recommendations of V1, which relies on financial statements, and V2, which integrates both financial statements and news sentiment. Additionally, the study assesses the performance of stocks following AI and human recommendations using buy-and-hold and abnormal return measures. Findings Findings reveal that both AI models issue fewer “Strong Buy” recommendations and more “Sell” recommendations than human analysts, particularly for underperforming stocks. V2 outperforms human analysts in predicting returns for “Strong Buy” recommendations, suggesting that AI’s integration of financial and news sentiment data improves predictive accuracy. AI models offer more balanced and consistent investment recommendations, reducing emotional biases. Research limitations/implications A limitation of this study is its reliance on historical data, which may not fully reflect real-time market conditions. Additionally, AI’s conservative nature may limit its ability to identify sentiment-driven opportunities. Future research should explore hybrid frameworks that integrate AI with human expertise to enhance decision-making. Originality/value This study contributes to the literature on AI in financial decision-making by demonstrating its ability to mitigate cognitive biases. The findings highlight AI’s potential to complement human analysts, improving objectivity and consistency in investment recommendations. These insights are valuable for investors, financial institutions and policymakers.