On Cost-Sensitive Distributionally Robust Log-Optimal Portfolio

Chung-Han Hsieh,Xiao-Rou Yu

2024-10-31

Abstract:This paper addresses a novel \emph{cost-sensitive} distributionally robust log-optimal portfolio problem, where the investor faces \emph{ambiguous} return distributions, and a general convex transaction cost model is incorporated. The uncertainty in the return distribution is quantified using the \emph{Wasserstein} metric, which captures distributional ambiguity. We establish conditions that ensure robustly survivable trades for all distributions in the Wasserstein ball under convex transaction costs. By leveraging duality theory, we approximate the infinite-dimensional distributionally robust optimization problem with a finite convex program, enabling computational tractability for mid-sized portfolios. Empirical studies using S\&P 500 data validate our theoretical framework: without transaction costs, the optimal portfolio converges to an equal-weighted allocation, while with transaction costs, the portfolio shifts slightly towards the risk-free asset, reflecting the trade-off between cost considerations and optimal allocation.

Optimization and Control,Systems and Control,Computational Finance,Portfolio Management

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to construct an investment strategy that is both robust and can maximize long - term returns in the face of uncertain return distributions and trading costs in financial portfolio optimization. Specifically, the paper proposes a new cost - sensitive distributionally robust log - optimal portfolio problem, in which investors face ambiguous return distributions and incorporate a general convex trading cost model. By using the Wasserstein metric to quantify the uncertainty of return distributions, the paper aims to establish conditions to ensure robust trading under all distributions within the Wasserstein ball and under convex trading costs. Using duality theory, the paper approximates the infinite - dimensional distributionally robust optimization problem as a finite - dimensional convex programming problem, making the computation of medium - sized portfolios possible.

### Main Problem Description
1. **Uncertain Return Distributions**: The true investment return distribution is often unknown or partially known, which poses a challenge to investment decision - making.
2. **Trading Costs**: The costs involved in the trading process (such as commission fees, bid - ask spreads, latency costs, and trading taxes, etc.) have an important impact on the optimal decision - making process.
3. **Robustness**: In an uncertain market environment, how to ensure the robustness of the portfolio, that is, to maintain positive returns under all possible market conditions.

### Main Contributions of the Paper
1. **Problem Modeling**: Proposed a cost - sensitive distributionally robust log - optimal portfolio problem that combines the Wasserstein metric and the convex trading cost model.
2. **Theoretical Analysis**: Derived conditions to ensure robust trading under all distributions within the Wasserstein ball.
3. **Computational Method**: Used duality theory to transform the infinite - dimensional distributionally robust optimization problem into a finite - dimensional convex programming problem, improving the computational feasibility.
4. **Empirical Research**: Conducted empirical research using S&P 500 data to verify the effectiveness of the theoretical framework. The study found that in the absence of trading costs, the optimal portfolio tends to be equally weighted; while in the case of considering trading costs, the portfolio will be slightly biased towards risk - free assets, reflecting the trade - off between cost considerations and optimal allocation.

### Formula Examples
- **Wasserstein Metric**:
\[
d(F_1, F_2):=\inf_{\Pi}\left\{\mathbb{E}_{(X_1, X_2)\sim\Pi}[\|X_1 - X_2\|]\right\}
\]
where $\Pi$ is the joint distribution of $X_1$ and $X_2$, and the marginal distributions are $F_1$ and $F_2$ respectively.

- **Cost - Sensitive Account Dynamics**:
\[
V(n)=V(0)\left(1+\sum_{i = 1}^m w_i(0)\frac{S_i(n)-S_i(0)}{S_i(0)}-\frac{TC(u(0))}{V(0)}\right)
\]
where $V(n)$ is the account value in the $n$ - th period, $w_i(0)$ is the initial portfolio weight, $S_i(n)$ is the price of the $i$ - th asset in the $n$ - th period, and $TC(u(0))$ is the trading cost.

- **Distributionally Robust Optimization Problem**:
\[
\max_{w\in W_s}\inf_{F\in B_\epsilon(\hat{F})}\frac{1}{n}\mathbb{E}_F\left[\log\left(\frac{V_w(n)}{V(0)}\right)\right]
\]
where $W_s$ is the set of feasible weights, and $B_\epsilon(\hat{F})$ is centered at the empirical distribution $\hat{F}$ with a radius of $\epsilon$.