Investor Overconfidence in the AI Era: Human vs. Algorithmic Decision-Making

Abstract

We investigate investor overconfidence in the age of artificially intelligent coincident with human–algorithmic and hybrid decision‐making models on the Pakistan Stock Exchange (PSX). Leveraging behavioral finance literature and advances in AI-based investment tools, the paper explores whether algorithmic behaviors alleviate or change overconfidence when human judgments persist. Based on panel data for PSX investors, 2020:2025, overconfidence is proxied by turnover, holding bias and relative deviation from trading algorithms. The empirical evidence shows that human-only investors have a higher turnover, more portfolio concentration and earn less risk-adjusted returns than algorithm-based portfolios. AI-driven portfolios exhibit better diversification and lower downside risk compared to traditional portfolios, but hybrid investors tend to ignore machine suggestions after an initial period of profit, which is consistent with learning-based overconfidence and illusion of control. Regressions suggest that overconfidence undermines the efficiency gains of AI via discretionary intervention, resulting in higher volatilities and more pronounced draw-downs when under financial stress. In general, the results imply that AI doesn’t remove behavioral biases but rather re-sculpts their manifestation in hybrid decision worlds. Our paper extends overconfidence theory into AI-mediated markets and has significant implications for investors, financial institutions and regulators in emerging markets.