Portfolio Optimization and Financial Risk Analysis using Quantitative and Computational Methods: Evidence from selected S&P 500 Firms
Oyebamiji Sodiq Akorede, Adeyemi Paul Oluwatimilehin, Adedeji Teslim Olaide, Akinyemi Akeem Alabi, Ruth Sinmilejesu Alesinloye
This study examines portfolio optimization and financial risk analysis for ten large-cap U.S. equities drawn from seven GICS sectors of the S&P 500, covering the period January 2021 to December 2025. Motivated by the well-documented limitations of classical mean-variance (MV) optimization (including sensitivity to return estimation error and the systematic underestimation of tail risk in non-normal return environments), the study adopts an integrated quantitative and computational framework comprising LASSO regression-based return prediction, Markowitz MV optimization, historical Value-at-Risk (VaR), Conditional Value-at-Risk (CVaR), and Monte Carlo simulation. Three portfolio strategies are constructed and evaluated which included a Classical MVO portfolio using historical mean returns, a Machine Learning-Enhanced portfolio substituting LASSO-predicted returns as expected return inputs, and an Equal-Weight benchmark. Results indicate that both optimized portfolios substantially outperform the Equal-Weight benchmark on a risk-adjusted basis, achieving Sharpe ratios of 1.374 and 1.317 respectively against the benchmark's 0.826. The ML-Enhanced portfolio generates the highest mean simulated five-year terminal wealth of 4.037 times the initial investment, with a near-zero probability of capital loss (0.05%) compared to 0.75% for the Equal-Weight strategy. Individual firm risk analysis reveals pronounced tail-risk heterogeneity, with Amazon recording the most extreme 99% CVaR of −25.210% and Coca-Cola the most resilient at −5.964%. The systematic divergence between VaR and CVaR estimates across all entities provides direct empirical support for CVaR's theoretical superiority as a risk measure in leptokurtic return environments which affirm the value of integrating machine learning with classical optimization theory and tail-sensitive risk measurement for evidence-based portfolio construction.
