Building a stock portfolio

• Main Author: Nim Jin Xiang
• Other Authors: Rajapakse Jagath Chandana
• Format: Final Year Project
• Language: English
• Published: 2014
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; DRNTU::Business::Finance::Portfolio management
• Online Access: http://hdl.handle.net/10356/61934
• Institution: Nanyang Technological University

Portfolio construction and optimization is one of the most popular topics in the finance industry. Investors and researchers spend most of their time finding the best techniques to predict the market movements, identify the profitable stocks, diversify their investments as well as optimize their investment portfolio. With the advancement in technology and computing power, investors and researchers are able to crunch huge volume of data to analyze the market and improve their investment performances. This project is an exploration of applying computational intelligence and techniques on financial market. This report details several techniques that are studied to replicate and optimize a stock portfolio. Artificial neural network and linear regression are presented in this report for portfolio replication, whereas quadratic programming and genetic algorithm are used for portfolio optimization. The objective functions used in portfolio optimization are mainly based on the Harry Markowitz’s Modern Portfolio Theory (MPT). He won the 1990 Nobel Memorial Prize in Economics Sciences with his work on MPT. For portfolio replication, both artificial neural network and linear regression approaches are shown to be able to replicate the Straits Times Index (STI) portfolio. The neural network and linear regression calculated portfolios are closely correlated to STI portfolio and has a positive correlation 71% and 78% respectively. Several types of portfolios configuration are demonstrated in this project for portfolio optimization. They are minimum risk portfolio, return-risk balanced portfolio, portfolio with no short position and no heavy concentration constraints, etc. The quadratic programming and genetic algorithm are shown to be able to search for an optimize portfolio according to the investors’ investment criterions and risk appetite. Genetic algorithm is better in avoiding the local minima in the optimization problem as compared to quadratic programming. R software programming language is used for the implementations and experiments for the entire project. The experiments are conducted with the historical price data extracted from Yahoo! Finance.