We study a specific solution concept of the canonical consumption-savings problem à la Allen & Carroll (2001). We use the machine learning toolbox to approximate the optimal rule with the help of tensors, through the pytorch library. We aim at maximizing the lifetime-reward, by the grandient descent technique, repeatedly simulating lives and computing the gradient of the lifetime utilities.
The code is replicated at the end of the pdf.
September 2019, Paris-Dauphine University
Supervision: Marie Bessec and Pablo Winant
Grade: 19/20