Multi-factor Quantitative Portfolio Risk-return Collaborative Optimization Model

Abstract

In the traditional quantitative investment model, the separation of risk and return will lead to tail risk exposure and parameter sensitivity. This paper proposes a multi-factor quantitative portfolio risk-return collaborative optimization model. Construct a risk-benefit cooperative objective function, introduce a gain term based on risk contribution entropy, and upgrade risk diversification from passive constraint to active incentive; The dynamic IC_IR time-varying weighting mechanism is designed, and the factor information coefficient (IC) and information ratio (IR) are calculated by rolling window, and the failure factors are eliminated in real time and the weights are adjusted adaptively. Ledoit-Wolf covariance matrix compression estimation is used to reduce the influence of high-dimensional data noise on the optimization results. Hard constraints such as industry deviation, upper limit of stock weight and maximum retracement warning are embedded, and endogenous optimization of transaction costs is combined to ensure the enforceability of the strategy. The empirical results show that the model has an annualized return of 18.75%, a Sharpe ratio of 1.02, and a maximum drawdown of -22.48% in the Chinese A-share market, which is superior to mean variance optimization (MVO) and IC-IR equal weight combinations. The inter-partition test shows that the model has a robust income and risk control ability in bull-bear conversion, unilateral downside and structural bull market, and the sortino ratio-Kumar ratio advantage highlights its precise management of downside risks.