Benchmarking Portfolio Models by Back-testing

Abstract

Investing in financial markets is a complex and challenging task that requires a deep understanding of financial theory, statistical analysis, and computational methods. One of the most critical decisions that investors face is the selection of an optimal portfolio that maximizes returns while minimizing risk. This decision is even more challenging in today's fast-paced and highly interconnected financial markets, where the number of available assets and investment strategies has increased significantly. To tackle this challenge, many investors turn to optimization techniques to find the most efficient portfolio allocation. This research project aims to develop and compare several portfolio optimization models based on modern financial theory and computational methods. Our goal is to provide investors with a practical and effective tool for selecting optimal portfolios that meet their investment objectives and risk tolerance. The project focuses on seven different portfolio optimization models, each based on a unique set of assumptions and dataset. The different optimal portfolios show varied allocation patterns across assets, influenced by risk preferences and market conditions. Performance analysis reveals diverse outcomes, with some models prioritizing returns such as the SSD, ISD First Order and CVaR while others focus on stable performance with lower volatility such as the Mean Variance and Black-Litterman. Incorporating a risk-free asset shifts optimal allocations, underscoring trade-offs in portfolio optimization. Equal Weights portfolios provide consistent diversified allocations, suitable for uninformed investors, while SSD, ISD First Order, and CVaR offer higher risk options. Mean-Variance and Black-Litterman models with a risk-free asset provide stable performance yet results in highly concentrated optimal portfolios. In general, asset selection and markets conditions coupled with the models’ objective function formulation influences optimal allocations. Employing a fixed single-period investment horizon ensure stable results, enhancing real-world applicability, and enabling direct comparison of model performance under varying market conditions. Overall, the thesis provides valuable insights for constructing portfolios aligned with investors' preferences and objectives.

Date of Award	8 May 2024
Original language	American English
Supervisor	Giorgio Consigli (Supervisor)