Optimizing Real Estate Portfolios: The Role of Generative AI in Geographic Diversification

In their novel study, "Optimizing Real Estate Portfolios: The Role of Generative AI in Geographic Diversification," Dombrowski and Seagraves explore an untested application of generative AI in real estate investment. Their research demonstrates how GPT-4o can analyze real estate data, develop predictive models, and construct portfolios that compare favorably with established industry benchmarks.

The researchers designed an experiment where GPT-4o analyzed a comprehensive dataset of 433 U.S. cities spanning nearly two decades. The AI processed Zillow Home Value Indices alongside multiple predictive factors including population data, unemployment rates, mortgage rates, and Google Trends search patterns for real estate terms. This represents one of the first academic attempts to use generative AI for constructing real estate investment portfolios optimized for risk-adjusted returns.

A distinctive aspect of their methodology was the comparison between obfuscated and unobfuscated datasets. By creating versions with and without city names, states, and dates, they discovered that when GPT-4o had access to complete geographic information, it created more diversified portfolios spanning eight different states, compared to just four states in the obfuscated selections. More importantly, these more geographically diversified portfolios achieved better risk-adjusted returns.

The results demonstrate AI's potential in real estate finance. The AI-generated portfolios claimed three of the four top-performing spots in out-of-sample backtests, with the 40-city portfolio achieving a 1.89 Sharpe ratio that outperformed all eight benchmarks. This represents a different approach to geographic diversification strategy, prioritizing data-driven market selection over traditional risk-spreading methods.

The researchers also developed technical innovations including a method for extracting city-specific real estate search interest from Google Trends and an approach for programmatically identifying Google Trends city topic codes. Their work is among the first academic explorations of using generative AI for real estate portfolio construction and among the earliest investigations into how data quality impacts AI's investment recommendations.

As demographic shifts and environmental risks continue reshaping the real estate landscape, this research offers insights into how AI-augmented portfolio optimization might enhance investment strategies, suggesting that sophisticated AI models could become valuable tools for data-driven financial decision-making in real estate markets.