Optimizing Real Estate Portfolio Using Genetic Algorithm and Machine Learning Approaches

This study presents an integrated approach to Real Estate Investment Trust (REIT) portfolio optimization using Genetic Algorithms (GA) and Machine Learning (ML), emphasizing reproducibility, risk analysis, and theoretical grounding. The proposed framework combines a GA-driven alpha search mechanism with advanced ML models−Multiple Linear Regression (MLR), Neural Networks (NN), and Long Short-Term Memory (LSTM)−trained on 23 technical and fundamental features. Improvements over traditional models (ARIMA, GARCH) are demonstrated through robust benchmarking, including transaction cost analysis and COVID-19 drawdown performance. Notably, GA yields a Sharpe ratio of 9.72, while NN achieves a Sharpe of 3.50. This study aligns with modern portfolio theory and addresses behavioral market responses, offering a data-driven yet interpretable foundation for REIT portfolio strategies.