Multi-objective investment portfolio optimization

Abstract

A well renowned problem in the world of finance is optimization of investment portfolios. An investor has a primary goal of maximising returns and minimizing risk at the same time for a portfolio. Constructing a well-diversified portfolio is not a straightforward task for investors or portfolio managers as optimization of two objectives simultaneously is tedious. A single-objective optimization as proposed by Henry Markowitz in his mean-variance theory for portfolio allocation does not meet expectations of a modern day investor. Moreover, the ability to use multi-objective functions not only cater to a wider range of possibilities but also provide a much higher accuracy for efficient asset allocation. Evolutionary algorithms using multiple objective functions prove to be much more efficient than traditional optimization techniques.This project focuses on the use of a few popular Multi-objective Evolutionary Algorithms to build well-diversified portfolios. Implementation on the S&P 500 dataset is carried out using four different algorithms and their Pareto fronts are compared. It includes the cardinality constraint and is also subject to further constraints that put a bound on the minimum allocation and add a feature of transaction costs. Furthermore, a few forecasting methods have been implemented in this research, these include Monte Carlo Simulation and NARX, to predict the future price of assets.