Abstract:In this article we present a general framework for non-concave distributionally robust stochastic control problems in a discrete time finite horizon setting. Our framework allows to consider a variety of different path-dependent ambiguity sets of probability measures comprising, as a natural example, the ambiguity set defined via Wasserstein-balls around path-dependent reference measures, as well as parametric classes of probability distributions. We establish a dynamic programming principle which allows to derive both optimal control and worst-case measure by solving recursively a sequence of one-step optimization problems. As a concrete application, we study the robust hedging problem of a financial derivative under an asymmetric (and non-convex) loss function accounting for different preferences of sell- and buy side when it comes to the hedging of financial derivatives. As our entirely data-driven ambiguity set of probability measures, we consider Wasserstein-balls around the empirical measure derived from real financial data. We demonstrate that during adverse scenarios such as a financial crisis, our robust approach outperforms typical model-based hedging strategies such as the classical Delta-hedging strategy as well as the hedging strategy obtained in the non-robust setting with respect to the empirical measure and therefore overcomes the problem of model misspecification in such critical periods.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the non - concave distributionally robust stochastic control problem in the discrete - time finite - horizon setting. Specifically, the author proposes a general framework to deal with this type of control problem, in which the set of probability measures involved may be path - dependent and allows for the consideration of various different probability measure ambiguity sets, such as the ambiguity set defined based on the Wasserstein ball and the parameterized probability distribution classes.
### Background and Problem Description of the Paper
In many practical applications, such as in the fields of finance, economy, and physics, decision - makers need to take a series of actions in an uncertain environment to maximize their expected returns. However, it is very difficult to model the underlying probability distribution of the environment because the choice of the model may be highly subject to the risk of model mis - specification (Knightian uncertainty or model risk). To deal with this uncertainty, one method is to use distributionally robust optimization (DRO), that is, to consider an ambiguity set that contains multiple possible true probability measures instead of a single deterministic model. The goal of the decision - maker is to optimize their expected returns under the worst - case probability measure.
### Main Contributions of the Paper
1. **General Framework**: The paper proposes a general framework for dealing with non - concave distributionally robust stochastic control problems, which is applicable to the discrete - time finite - horizon setting.
2. **Dynamic Programming Principle**: The author proves a dynamic programming principle, which finds the optimal control and the worst - case probability measure by recursively solving a series of one - step optimization problems.
3. **Non - concave Objective Function**: Different from most of the existing literature, this paper does not require the objective function to be concave (or convex), so that some important problems that are usually difficult to solve due to non - concavity can be analyzed.
4. **Specific Application**: As a specific application, the paper studies the robust hedging problem of financial derivatives under an asymmetric (and non - convex) loss function. The author uses the Wasserstein ball around the empirical measure obtained from the real financial data as a data - driven ambiguity set, and shows that in adverse scenarios (such as the financial crisis), their robust method is superior to traditional model - based hedging strategies, such as the classic Delta hedging strategy.
### Technical Details
- **Definition of Ambiguity Set**: The ambiguity set in the paper can be defined by referring to the Wasserstein ball around the probability measure or other parameterized probability distribution classes.
- **Dynamic Programming Principle**: By introducing a new stability condition (see Assumption 2.1 (iii)) and applying Berge's maximum theorem, the author proves the main results.
- **Numerical Method**: The paper also provides a numerical method based on deep neural networks to approximately solve the optimal control problem.
### Experimental Verification
The author uses the historical daily return data of Apple stocks to construct a data - driven ambiguity set, and trains non - robust (ε = 0) and robust (ε = 0.001) hedging strategies. The experimental results show that during the test period, the robust hedging strategy performs better in adverse situations, especially during the COVID - 19 crisis in 2020.
### Conclusion
This paper proposes a general framework for dealing with non - concave distributionally robust stochastic control problems, and verifies its effectiveness and superiority through specific applications and experimental verification. This method has important theoretical and practical significance in dealing with decision - making problems in highly uncertain environments.