Machine-learning a family of solutions to an optimal pension investment problem
We use a neural network to identify the optimal solutions to a family of pension investment problems, where the parameters determining an investor's risk and consumption preferences are given as inputs to the neural network in addition to economic variables. Training a single network across such a family fails without modification. Our main contribution is a scaling of the loss function that resolves this, together with a proof that the resulting algorithm converges. We use this to develop a practical tool for exploring how pension outcomes vary with preference parameters. We use a Black-Scholes economic model so that we may validate the accuracy of the network using a classical and provably convergent numerical method developed using the duality approach.