TA12 - Ingénierie financière / Financial Engineering

Risk budgeting is a widely used approach in portfolio management that allocates portfolio risk across assets according to prescribed budgets. Most existing risk budgeting models rely on smooth risk measures, such as variance, while only recent work has considered the non-smooth mean absolute deviation (MAD) risk measure, which measures risk through the average absolute deviation of portfolio returns from their mean. It is crucial to properly address the uncertainty in the return distribution in this problem, since portfolio allocations can be sensitive to distributional misspecification and sampling error in the available data, which may in turn weaken out-of-sample performance. In this paper, a distributionally robust risk budgeting framework under the MAD risk measure is proposed, where the return distribution is allowed to vary within a suitably defined Wasserstein ambiguity set. The resulting model seeks portfolio allocations whose risk contributions remain stable under adverse distributions while preserving the diversification principle of risk budgeting. Since this formulation leads to a challenging minimax optimization problem involving both non-smoothness and distributional uncertainty, a tractable reformulation is developed, and efficient numerical solution methods are designed for the resulting model. Numerical experiments are conducted to examine the effectiveness of the proposed framework under various performance measures.

We propose a deep hedging framework for index option portfolios, grounded in a realistic market simulator that captures the joint dynamics of S&P 500 returns and the full implied volatility surface. Our approach integrates surface-informed decisions with multiple hedging instruments and explicitly accounts for transaction costs. The hedging strategy also considers the variance risk premium embedded in the hedging instruments, enabling more informed and adaptive risk management. Tested on a historical out-of-sample set of straddles from 2020 to 2023, our method consistently outperforms traditional delta-gamma hedging strategies across a range of market conditions.

Commercial airlines use revenue management to maximize revenue through real-time pricing decisions across booking classes. Classical approaches heavily rely on models of passenger arrival, choice, and cancellation, making performance sensitive to model accuracy. To overcome these limitations, we apply deep reinforcement learning with leg-wise decomposition, employing Deep Controlled Learning and Proximal Policy Optimization. Results show consistent improvements over simulation-based optimization benchmarks.