Transforming portfolio optimisation: a hybrid machine learning and Monte Carlo approach for superior asset allocation

This study aims at combining machine learning (ML) methods for smart asset choices with modern portfolio theory (MPT) and Monte Carlo simulations. Hybrid strategy was applied utilising Python for combining supervised ML models (XGBoost, random forest) and unsupervised learning (K-means clustering) for selecting stocks based on engineered features like rolling mean, log returns, and volatility. The chosen assets were then optimised by MPT and Monte Carlo strategies to generate risk-aware portfolios. Data were drawn from 18 diversified stocks from developed and developing economies from 2013-2023. Random forest classifier performed above 70% accuracy in selecting leading-performing stocks. Monte Carlo simulations provided the best Sharpe ratio (~0.78), which surpassed MPT's optimal value (~0.74), establishing better risk-adjusted returns. ML-based filtering also proved that the study further validated for more stable and diversified portfolios. The hybrid approach provides improved accuracy, diversification, and offers investors a more practical tool for making balanced investment decisions in volatile markets.