BANQUE INTERNATIONALE DE SUJETS DE THESE

Ecole Doctorale : EDMITT - Ecole Doctorale Mathématiques, Informatique et Télécommunications de Toulouse

Machine Learning for Stochastic Control and Structural Econometrics
n°195

Encadrant de thèse :
Eric GAUTIER
Adresse mail :
eric.gautier@tse-fr.eu
Laboratoire de rattachement :
TSE-R - Toulouse School of Economics - Recherche
Adresse du site internet du laboratoire :
http://www.tse-fr.eu/fr/umr-tse-r
Directeur de laboratoire :
 Arnaud REYNAUD
Adresse mail :
 arnaud.reynaud@inra.fr
École Doctorale :
 EDMITT
 EDTSE
PAYS :

Chine
Colombie
Mexique - discipline :
Langue de la thèse :
francais ,anglais ,

Directeur de thèse :
Eric GAUTIER
Descriptif : PhD co-supervised by Stéphane Villeneuve

A large class of models in Economics rely on conditional expectations involving multidimensional processes. Dynamic problems for decision making under uncertainty can be modelled as a stochastic optimal control problem in discrete time and finite horizon. Dynamic programming allows, when the state process is Markovian, to characterise the optimal controls using a recursive algorithm which requires to compute a conditional expectation at each step. Inference on structural parameters based on non-experimental data usually relies on nonparametric restrictions involving conditional expectations. The goal of this thesis is to develop and analyse algorithms based on machine learning, in particular deep neural networks, for Monte-Carlo approximation/estimation of high-dimensional conditional expectation functions, and dynamic programming/inverse problems, and understand how they adapt to unknown structure.


Publications relatives au sujet :
[1] S Becker, P Cheridito, and A Jentzen, Deep optimal stopping, https://arxiv.org/pdf/1804.05394.pdf; [2] M Germain and X Warin, Neural networks-based algorithms for stochastic control and PDEs in finance, to appear in Machine Learning for Financial Markets: a guide to contemporary practices, Cambridge University Press, Editors: Agostino Capponi and Charles-Albert Lehalle; [3] A Bachouch, C Huré, and N Langrené, Deep neural networks algorithms for stochastic control problems on finite horizon: numerical applications, to appear in Methodology and Computing in Applied Probability
[4] M Carrasco, J-P Florens, and E Renault Linear Inverse Problems and Structural Econometrics Estimation Based on Spectral Decomposition and Regularization, The Handbook of Econometrics, vol 6B (2007); [5] M Farrell, T Liang, S Misra, Deep Neural Networks for Estimation and Inference, Econometrica 89 (2021)

Mots clés : Deep learning ;Optimal Stopping ;Stochastic Control ;Structural Econometrics ;

Collaboration avec un laboratoire : TSE-R - Toulouse School of Economics - Recherche