Democratizing Alpha: LLM-Driven Portfolio Construction for Retail Investors Using Public Financial Media

Recent technological advancements and macroeconomic changes have led to a surge in individual investors' participation in capital markets. However, these investors face challenges in making investment decisions due to bounded rationality, characterized by constraints on their time, cognitive capacity, and ability to process vast information. This study empirically examine whether publicly accessible large language models (LLMs) and financial media information to enhance the investment performance of retail investors. We employ daily market commentary video transcripts from publicly available YouTube channels, including Bloomberg Television and Yahoo Finance, to prompt four LLMs (LLaMA 3, Qwen2, Gemma, GPT 4o-mini) to construct investment portfolios. These portfolios are then backtested against the S\&P 500 and NASDAQ from June 2024 to July 2025. The analysis demonstrates that LLM-based portfolios exhibited consistently outperform market benchmarks across critical performance metrics, including CAGR, Sharpe ratio, and Calmar ratio. Qualitative analysis further confirms that LLMs successfully extract coherent and economically meaningful investment rationales from unstructured video content. Our findings provide a practical methodology for retail investors to leverage accessible AI, democratizing advanced analytical techniques once exclusive to institutional investors and demonstrating that AI-based tools can effectively support rational decision-making.