Multi-objective Loan Portfolio Optimization in Peer-to-Peer Lending Markets using Machine-Learning Techniques

Abstract

Portfolio optimization problems in the Peer-to-Peer lending Platforms involve selecting good loan applications (less risky) from various potential borrowers. Such loans have lower level of risk in terms of funding and earning higher returns. The aim of this study is to find ways to maximize returns and minimize the risks associated with the investment. It becomes more complicated to optimally allocate weights to the loan application when there is an increased number of applications for funding. This study focused on devising techniques which can be used to optimally select portfolios of loan applications for funding with desired returns on the investment. Harry Markowitz pioneered the Modern Portfolio theory also known as Meanvariance theory to construct a portfolio but the theory failed since it was built on unrealistic assumptions in terms of real life situations. This study explored and compared the meanvariance theory and other machine learning methods to construct a portfolio of loans from peer-to-peer lending market in order to be able to recommend the best approach to achieving high returns with minimum risk. The study employed the evolutionary algorithms (Particle Swarm Optimization and Genetic Algorithm) and the Reinforcement learning algorithm