Investment portfolio selection using a combination of improved BCO BP neural network prediction and genetic algorithms

• Main Author: Ong, Desmond Kok Chuan.
• Other Authors: Yow Kin Choong
• Format: Final Year Project
• Language: English
• Published: 2009
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: http://hdl.handle.net/10356/19291
• Institution: Nanyang Technological University

Portfolio selection has always been a very popular topic in the financial community. Investors and fund managers are constantly on the search for better techniques to accurately identify potentially profitable stocks. In recent years, most of them have learned to leverage on the technological advancement in computers to crunch huge volumes of data and try to predict the market as well as improve their portfolio management strategies. This report presents a portfolio selection system that selects stocks from an index and predicts their stock price in 7 days time. The stocks’ returns are reduced by 0.5% which is the estimated transaction cost. After which, the stocks are passed to a genetic algorithm (GA) to construct their suggested capital allocation. There are two types of investors considered, the risk-seeking investor and the risk-averse investor. The genetic algorithm is set to maximise the Sharpe ratio of the stocks with respect to the USD London Interbank Offered Rate (LIBOR) as the risk free asset for the risk-seeking investor. As for the risk-averse investor, the GA is set to minimise the variance also known as the risk of the portfolio. The prediction module is a Feedforward Backpropagation Multi-layer Perceptron neural network powered by an improved bacteria chemotaxis optimisation (IBCO) method. The module uses 10 technical indicators to predict the price difference of the stock price curve. The experiment was conducted with historical stock data from the Dow Jones Industrial Average (DJIA) index.