Boosting AI with HI (Human Intelligence): An Application to Strategic Asset Allocation

Can Large Language Models (LLMs) such as ChatGPT serve as effective tools for strategic asset allocation? This project investigates whether LLM-based portfolio construction can match or outperform traditional quantitative approaches such as mean-variance optimization, minimum-variance portfolios, and the Black-Litterman model. We propose a hybrid framework that combines the qualitative screening capabilities of LLMs with rigorous quantitative optimization techniques. Using an extensive out-of-sample analysis across eight major global equity indices (DJIA, S&P 500, CAC 40, DAX 40, FTSE 100, FTSE MIB, IBEX 35, and TSX), we examine whether LLM-generated portfolios add economic value relative to traditional benchmarks. Preliminary results from over 4,000 API simulations per index suggest that while pure LLM portfolios generally underperform classical optimization methods, hybrid approaches that leverage LLMs for stock screening and human-guided quantitative models for weight determination show promise. The research will deliver practical guidelines for pension funds and asset managers on the effective integration of AI tools in strategic asset allocation decisions.