Optimal Feature Selection for Market Risk Assessment: A Dimensional Reduction Approach in Quantitative Finance

Abstract

This paper addresses the dimensional challenges in market risk assessment through a comprehensive investigation of feature selection methodologies in quantitative finance. We propose a hierarchical feature selection framework that integrates statistical and machine learning approaches to identify optimal feature subsets for market risk modeling. Experimental validation using multiple financial datasets, including 8-year historical data from the Chinese A-share market encompassing 3,000 listed companies, demonstrates the efficacy of the proposed approach. The Random Forest-based feature selection methodology achieves superior performance with 76.2% dimensional reduction while improving predictive accuracy by 5.1% compared to traditional approaches. Performance evaluation across various market scenarios reveals significant enhancements in Value-at-Risk estimation accuracy during high volatility periods, with error reduction of 12.5% in crisis scenarios. The hybrid RF-RF approach demonstrates robust performance with a Sharpe ratio of 1.57 in portfolio backtesting, substantially outperforming models utilizing full feature sets. The proposed framework offers practical implications for financial institutions by enhancing computational efficiency and regulatory compliance while maintaining model interpretability. This study contributes to the advancement of market risk assessment methodologies by establishing a systematic approach to dimensional reduction in complex financial data environments.