Chaire du Prof. ordinaire Morellec

Chaire

À propos de ce résultat

Cette page est générée automatiquement et peut contenir des informations qui ne sont pas correctes, complètes, à jour ou pertinentes par rapport à votre recherche. Il en va de même pour toutes les autres pages de ce site. Veillez à vérifier les informations auprès des sources officielles de l'EPFL.

Personnes associées (13)

Personnes associées

Chargement

Unités effectuant des recherches similaires (118)

Afficher plus

Unités effectuant des recherches similaires

Chargement

Domaines de recheche associés (34)

Domaines de recheche associés

Chargement

Publications associées (58)

Publications associées

Chargement

Identifiability and Generalizability from Multiple Experts in Inverse Reinforcement Learning

Volkan Cevher, Boris Nikolov, Paul Thierry Yves Rolland, Luca Viano

While Reinforcement Learning (RL) aims to train an agent from a reward function in a given environment, Inverse Reinforcement Learning (IRL) seeks to recover the reward function from observing an expert’s behavior. It is well known that, in general, various reward functions can lead to the same optimal policy, and hence, IRL is ill-defined. However, [1] showed that, if we observe two or more experts with different discount factors or acting in different environments, the reward function can under certain conditions be identified up to a constant. This work starts by showing an equivalent identifiability statement from multiple experts in tabular MDPs based on a rank condition, which is easily verifiable and is shown to be also necessary. We then extend our result to various different scenarios, i.e., we characterize reward identifiability in the case where the reward function can be represented as a linear combination of given features, making it more interpretable, or when we have access to approximate transition matrices. Even when the reward is not identifiable, we provide conditions characterizing when data on multiple experts in a given environment allows to generalize and train an optimal agent in a new environment. Our theoretical results on reward identifiability and generalizability are validated in various numerical experiments.

2022

Chargement

We challenge the view that short-term debt curbs moral hazard and demonstrate that, in a world with financing frictions and fair debt pricing, short-term debt generates incentives for risk-taking. To do so, we develop a model in which firms are financed with equity and short-term debt and cannot freely optimize their default decision because of financing frictions. We show that when firms are close to distress, the dynamic interaction of operating and rollover losses fuels default risk. In such instances, shareholders find it optimal to increase asset risk to improve interim debt repricing and prevent inefficient liquidation. These risk-taking incentives do not arise when debt maturity is sufficiently long. (C) 2020 Elsevier B.V. All rights reserved.

ELSEVIER SCIENCE SA2020

Chargement

We build a dynamic agency model in which the agent controls both current earnings via short-term investment and firm growth via long-term investment. Under the optimal contract, agency conflicts can induce short- and long-term investment levels beyond first best, leading to short- or long-termism in corporate policies. The paper analytically shows how firm characteristics shape the optimal contract and the horizon of corporate policies, thereby generating a number of novel empirical predictions on the optimality of short-versus long-termism. It also demonstrates that combining short- and long-term agency conflicts naturally leads to asymmetric pay-for-performance in managerial contracts. (C) 2019 Elsevier B.V. All rights reserved.

ELSEVIER SCIENCE SA2020

Chargement