A globally convergent method for solving a quartic generalized Markowitz portfolio problem (Q2143112)

The authors present an alternating minimization method to solve the variance-kurtosis generalized Markowitz problem \[ \begin{aligned} \min \: \: \: &f(\mathbf{y})+\delta_{\Delta_n}(\mathbf{y})+\delta_K(\mathbf{z})\\ \text{s.t.} \: \: \: &\mathbf{y}-\mathbf{z}=0, \end{aligned} \] where \[ \begin{aligned} f(\mathbf{y}) =& A\mathbf{y}^4+\nu \mathbf{y}^T\sum \mathbf{y}\quad (\mathbf{y}\in \mathbb{R}^n),\\ \Delta_n :=& \left \{\mathbf{y}\in \mathbb{R}^n \mid \sum_{i=1}^n y_i=1,\: \mathbf{y}\ge 0 \right \},\\ K :=& \left \{\mathbf{y} \mid \overline {\mathbf{x}}^T\mathbf{y}\ge r_{\min},\: \mathbf{y}^T\sum \mathbf{ y} \le k_{\max} \right \},\\ \delta_{S}(\mathbf{x}) :=& \begin{cases} 0 & \text{if }\mathbf{x}\in S,\\ +\infty &\text{otherwise}. \end{cases} \end{aligned} \] An inexact ADMM (iADMM) algorithm for solving the proposed optimization model is also given and the global convergence of this algorithm is established.