Title
Enhancing Portfolio Allocation with Machine Learning-Based Return Predictions: Does Frequency Matter
Authors
Abstract
Research aims: This study explores the integration of machine learning and optimization techniques (both classical & heuristic) in portfolio allocation, focusing on how data frequency influences predictive accuracy and investment performance.
Design/Methodology/Approach: Using five major technology stocks (AAPL, MSFT, GOOG, AMZN, META), six regression models—Ridge, Linear, Random Forest, XGBoost, Multilayer Perceptron (MLP) and Support Vector Regression (SVR)—were employed to predict asset returns based on historical data at daily, weekly, and monthly frequencies. Predicted returns were then optimized using three methods: Sequential Least Squares Programming (SLSQP), Particle Swarm Optimization (PSO), and Differential Evolution (DE), under realistic weight constraints (0.05 to 0.8 per asset, no short-selling).
Research findings: While machine learning (ML) models achieve higher predictive accuracy with higher-frequency (daily) data, this does not necessarily lead to better portfolio performance. PSO consistently outperformed other optimizers across all frequencies. Notably, portfolios optimized using PSO with weekly and monthly predictions delivered higher Sharpe ratios than the equal-weighted benchmark, while daily-frequency portfolios underperformed.
Theoretical contribution/Originality: This study offers valuable insights into the integration of machine learning in portfolio optimization, highlighting the effect of data frequency on the performance of the ML-embedded optimized portfolio.
Practitioner/Policy implication: Moderate to lower data frequencies may provide more robust signals for ML-embedded portfolio optimization, offering a better balance between prediction quality and investment returns.
Keywords
Portfolio Optimization, Machine Learning, Data Frequency, Particle Swarm Optimization, Predictive Accuracy, Investment Performance
JEL Codes
C45, C61, G11, G17
How to Cite
Wong, Z. Z., & Hooy, C. W. (2025). Enhancing Portfolio Allocation with Machine Learning-Based Return Predictions: Does Frequency Matter. The International Journal of Finance, 37(1), 20–38.