Large-Scale Portfolio Construction with Regularised Regression-Based Methods

• Main Author: Tay, Jeremiah Wei Jie
• Other Authors: PUN Chi Seng
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Business::Finance::Portfolio management; Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/146123
• Institution: Nanyang Technological University

Optimal portfolio asset allocation has played an increasingly important role in finance ever since Markowitz laid down a mathematical approach to portfolio optimisation in the 1950s. This article extends the current body of literature by examining the portfolio optimisation approach in a new light, introducing a methodological way to construct large-scale portfolios using regularised regression methods. It demonstrates that with an appropriate choice of the regularisation parameter, the regularised regression portfolios are able to achieve a level of risk that is comparable to the oracle risk. In addition, it shows that the portfolios formed by the different regularisation techniques exhibited different characteristics. This allows an investor to select an optimal portfolio based on the desired portfolio characteristics and specifications. We illustrated this by using simulation studies on the 100 Fama-French industrial portfolios and 500 randomly selected stocks from the Russell 3000 index.