Abstract:

The mean variance （MV） model is the basis of modern asset allocation theory. MV disperses risk through minimizing the correlation of asset returns within the portfolio. Since the variance is not applicable to the non-Gaussian distribution of asset returns and the covariance can only describe the linear correlation between assets， MV is interfered by modeling random errors and the internal or external risks in the market， resulting in poor generalization performance. The existing portfolio researches mainly improve the MV modeling error by modifying the risk measurement and optimizing the risk structure， but those models still have strong data dependence which makes the risk offset unstable. Besides， those models are limited to risk dispersion which cannot resist exogenous shocks. Risk conduction exists partially deterministic causal relationship. However， financial risk forms a complex association through multiple paths of conduction， which makes the overall behavior show high order nonlinear characteristics. In this paper， an improved idea of risk auto disturbance rejection is proposed. Using the sequential risk actions， MV is improved to construct Portfolio Risk Active Disturbance Rejection Model （PRADR）. Empirical research is conducted using A-share stock market data from July 10， 2017 to December 30， 2022， and the results show that the stock market risk is formed by the high-order of independent risk causes interacted and sequentially conducted along the supply chain. The high-order risk conduction network only needs \mathrm{1}/\mathrm{7} risk causes of the first-order linear network， which achieves the same risk interpretation degree； Through two-dimensional cross repeated test， PRADR Sharpe ratio is larger and less volatile， which illustrates its portfolio performance is better than MV； Suffering from the exogenous stock， PRADR has stronger risk auto disturbance rejection ability and higher generalization performance. The key to portfolio risk autoimmunity is to extract the deterministic relationships contained in uncertain risks and resist uncertainty interference. The deterministic risk transmission laws contained in various enterprise relationships need to be explored and applied.

Key words: mean-variance model, risk active disturbance rejection, risk conduction network, independent component correlation algorithm