Contrôle Stochastique et Apprentissage Statistique : exemple de la gestion middle-out d'un portefeuille

schedule le lundi 19 mars 2018 de 17h00 à 18h00

Organisé par : C. Cosco, S. Coste, L. Marêché, P. Melotti, N. Meyer

Intervenant : Alexis Bismuth (CMLA (ENS Paris-Saclay), LPSM (Paris VI - Paris VII), LGLS (CEA Saclay) and CES (Sorbonne))
Lieu : Jussieu, salle Paul Lévy, couloir 16-26 salle 113

Sujet : Contrôle Stochastique et Apprentissage Statistique

Résumé :

In stochastic optimization two approaches seem antagonic. Dynamic programming or stochastic control is used to find optimal strategies when the underlying model is known, but at the expense of heavy computations or complex PDEs (named HJB).  To the contrary, online machine learning requires easier computation and is applied when the underlying model is unknown, but there is generally no guaranty of optimality.  We present a way to reconcile these backward and forward reasoning approaches, in the frequentist and Bayesian frameworks, without approximations, without increasing the dimensionality, and apply it to a problem of portfolio choice.