Abstract:
The growing use of artificial intelligence (AI) in finance has led to the development of robo-advisory tools capable of generating investment portfolios with limited human input. However, empirical evidence on their effectiveness remains limited, particularly in developing markets. This study evaluates the performance of an AI-generated equity portfolio, referred to as DeepSeek V1, and compares it with traditional benchmark portfolios. A diversified portfolio of 15 U.S. equities was generated using a prompt-based large language model and backtested using Portfolio Visualizer over the period January 2015 to October 2025. The portfolio’s performance was compared with the S&P 500 Index, a traditional 60/40 portfolio, and the All-Weather portfolio using key performance measures including CAGR, volatility, Sharpe ratio, Sortino ratio, alpha, and maximum drawdown. The findings indicate that the DeepSeek V1 portfolio achieved superior risk-adjusted performance and lower drawdowns relative to traditional portfolios, although it did not consistently outperform the S&P 500 in absolute returns. The study highlights the potential of AI-assisted portfolio construction while emphasizing the need for cautious interpretation of results.
