Deep learning for portfolio optimization with transaction costs

Abstract

In this paper, we review the work of [5], who addressed the portfolio optimization problem with transac- tion costs under the Constant Relative Risk Aversion (CRRA) utility function using deep learning, and we apply the parameters based on real data from bonds, real estate, and the KOSPI index. We begin by examining the characteristics of the portfolio optimization problem with transaction costs through a re- view of existing literature. Additionally, we provide an explanation of the principles of deep learning as a means to solve this problem. The objective of [5] is to overcome the curse of dimensionality inherent in numerical Partial Differential Equation (PDE) methods by directly computing the optimal strategy using a deep learning algorithm. We train deep learning using parameters obtained from real data to derive the No Trading region (NT) and the value function. Additionally, we compare these results with those obtained through the numerical penalty PDE method.