Article Details
  • Published Online:
    January  2026
  • Product Name:
    The IUP Journal of Applied Economics
  • Product Type:
    Article
  • Product Code:
    IJAE010126
  • DOI:
    10.71329/IUPJAE/2026.25.1.5-21
  • Author Name:
    Alaknanda Lonare and Nityoday Tekchandani
  • Availability:
    YES
  • Subject/Domain:
    Economics
  • Download Format:
    PDF
  • Pages:
    5-21
Volume 25, Issue 1,January-March 2026
Evaluating Classical and Machine Learning Portfolio Optimization Strategies: A Walk-Forward Analysis of BSE Industrials Index
Abstract

This study examines the performance of classical and machine learning (ML)-based portfolio optimization strategies within the Indian equity market, focusing on the BSE Industrials index from March 2022 to March 2025. Classical models, including Markowitz Mean-Variance Optimization (MVO) and the Single Index Model (SIM), are compared against ML-driven portfolios using predictions from five models— Linear Regression, Ridge, Random Forest, XGBoost, and LightGBM—to forecast next-day returns. A rigorous walk-forward backtesting methodology is employed to evaluate all strategies under realistic, out-of-sample conditions. Performance metrics include annualized return, volatility, Sharpe ratio, and paired t-tests on daily returns for statistical significance. The results show that classical models optimized with full-period data yield inflated performance due to hindsight bias, while walkforward versions underperform passive benchmarks. Among ML models, Random Forest achieved the lowest mean squared error (MSE) and produced the highest Sharpe ratio (1.27) in realistic backtests—slightly outperforming the index (Sharpe 1.185) but without statistical significance in mean returns. Directional accuracy of all models hovered around 50%, indicating limited ability to predict market direction. The study concludes that consistent outperformance over passive benchmarks using either classical rolling data or ML return predictions is highly challenging. The results underscore the importance of robust out-of-sample testing and highlight the difficulties of converting ML predictive power into practical portfolio alpha.

Introduction

Financial planning and portfolio management are cornerstone disciplines in modern finance, aiming to maximize investor returns for a given level of risk, or minimize risk for a target