This event is reserved for GERAD members and their industrial partners. Registration is required.

14 h 00  Welcome address of Olivier Bahn

14:05   Geneviève Gauthier, HEC Montréal
Enhancing Deep Hedging of Options with Implied Volatility Surface Feedback Information

We present a dynamic hedging scheme for S&P 500 options, where rebalancing decisions are enhanced by integrating information about the implied volatility surface dynamics. The optimal hedging strategy is obtained through a deep policy gradient-type reinforcement learning algorithm, with a novel hybrid neural network architecture improving the training performance. The favorable inclusion of forward-looking information embedded in the volatility surface allows our procedure to outperform several conventional benchmarks such as practitioner and smiled-implied delta hedging procedures, both in simulation and backtesting experiments. (Joint work with Pascal François, Frédéric Godin and Carlos Octavio Pérez Mendoza)

14:35   Jonathan Jalbert, Polytechnique Montréal
Modélisation des valeurs extrêmes des précipitations sur plusieurs durées d'accumulation

La théorie des valeurs extrêmes recommande l'utilisation de la loi généralisée des valeurs extrêmes (GEV) pour modéliser les extrêmes des précipitations et extrapoler les valeurs correspondant à des quantiles d’ordre élevé. Les valeurs extrêmes étant rares par définition, les estimations des paramètres de la loi GEV sont souvent caractérisées par une grande variance échantillonnale, impactant ainsi l'estimation des quantiles d’ordre élevé. Il est donc crucial d'utiliser au mieux les informations disponibles pour réduire l'incertitude liée à l'estimation des valeurs extrêmes. Une façon de faire consiste à exploiter la dépendance des précipitations accumulées sur plusieurs périodes afin de partager les informations contenues dans les différentes durées. Le but de cette présentation consiste à développer un modèle statistique intégrant la dépendance des précipitations accumulées sur différentes périodes pour améliorer l’estimation des quantiles. Le modèle développé est particulièrement adapté à l’estimation de courbes intensité-durée-fréquence des précipitations, un outil réglementaire essentiel pour le dimensionnement des infrastructures par les ingénieurs.

15:05  Okan Arslan, HEC Montréal
Survivability and Quality of Service in Network Design

In network design, survivability is broadly defined as the network's ability to guarantee the connectivity of origin-destination (O-D) pairs when certain edges fail. This is achieved by ensuring the existence of backup paths. Quality of service (QoS), on the other hand, is expressed as a function of the length of a path connecting the O-D pair, with shorter paths corresponding to higher QoS. We review fundamental problems arising in survivable network design and discuss some recent advances. The emphasis in this talk is on novel modeling techniques based on length-bounded cuts and the associated combinatorial complexities.

15:35  Coffee break

16:00   Issmail El Hallaoui, Polytechnique Montréal
OCP Optimizes its Supply Chain for Africa and The World

Operations research specialists at OCP Group, Mohammed VI Polytechnic University, and GERAD/Polytechnique Montreal operationalized a system that optimizes the OCP downstream supply chain operations. The system simultaneously schedules production, inventory, and vessels while ensuring the highest demand fulfillment level.

To operationalize the system, the team equipped it with various heuristic and exact operations research tools. Furthermore, inspired by the practice, the team implemented novel variants of Benders decomposition, particularly a hybrid one which consists of fixing some complicating variables related to confirmed orders and freeing others related to unconfirmed orders in the Benders subproblem.

Initially, the system was a bottleneck, curbing the use of other supply chain management tools. OCP management now attributes very significant operational advantages to the operationalization of the system.

16:30  James R. Forbes, McGill University
Applications of Linear Systems Theory to Koopman Operator Approximation

This talk will discuss the use of linear systems theory for Koopman operator approximation. The Koopman operator allows nonlinear systems to be represented as infinite-dimensional linear systems by viewing their evolution in terms of an infinite set of lifting functions. Data-driven methods are used to identify a finite-dimensional approximation of the Koopman operator using a finite selection of lifting functions. Thanks to its linearity, the approximate Koopman model can be used for analysis, design, and optimal controller or observer synthesis. Depending on the choice of lifting functions, Koopman operator approximation methods can incorrectly identify unstable models for asymptotically stable systems. Two methods are proposed to guarantee asymptotically stable Koopman models: the first constrains the spectral radius of the Koopman matrix, while the second regularizes the H∞ norm of the Koopman system. Inherently unstable systems often require a stabilizing controller to operate practically. To account for the structure of a closed-loop system, a closed-loop Koopman operator identification method is proposed, in which Koopman models of the closed-loop system and plant are simultaneously identified given prior knowledge of the controller. This work was done with Steven Dahdah and can be found in https://arxiv.org/abs/2110.09658 and https://arxiv.org/abs/2303.15318.

17:00 - 19:00  Cocktail