RSOME案例：分布鲁棒投资组合模型

RSOME: Robust Stochastic Optimization Made Easy

---

Distributionally Robust Portfolio

Given $J$ investment options whose random returns are collectively denoted by a random vector $\widetilde{z\text{z}z}$, we choose a portfolio that maximizes the worst-case (i.e., least) expected utility:

$$\begin{array}{rl}\underset{\mathbf{x}}{\max }& \underset{\mathbb{P}\in \mathcal{F}}{\inf }{\mathbb{E}}_{\mathbb{P}}\left[U\left({\widetilde{d\text{d}d}}^{\top }x\text{x}x\right)\right]\\ \text{s.t.}& {e\text{e}e}^{\top }x\text{x}x=1\\ & x\text{x}x\in {\mathbb{R}}_{+}^J.\end{array}$$

Here we assume that a commonly used utility function is piecewise affine and concave such that U ( y ) = min ⁡ k ∈ [ K ] { α k y + β k } U(y) = \min_{k \in [K]} \left\{\alpha_k y + \beta_k\right\} U(y)=mink∈[K]​{ αk​y+βk​} with ${\alpha }_k\ge 0$. We also assume the mean returns, the variance of each option, as well as the variance of sum of these options are known, captured by the widely used partial cross-moment ambiguity set (Bertsimas et al. 2019, Delgage and Ye 2010, Wiesemann et al. 2014):

$$\mathcal{G}=\{\mathbb{P}\in {\mathcal{P}}_0({\mathbb{R}}^J)|\begin{array}{l}\widetilde{d\text{d}d}\sim \mathbb{R}\\ {\mathbb{E}}_{\mathbb{P}}[\widetilde{d\text{d}d}]=\mu \text{μ}\mu \\ {\mathbb{E}}_{\mathbb{P}}[({\tilde{d}}_j-{\mu }_j{)}^2]\le {\sigma }_j^2\\ {\mathbb{E}}_{\mathbb{P}}[(e\end{array}$$