Reducing Portfolio Volatility with Artificial Neural Networks

C.A. Casas (USA)


: artificial intelligence, decision supportsystems, expert systems, neural networks.


Artificial neural networks were trained to forecast monthly returns of three investment asset classes: large cap growth, large cap value, and small cap growth. Based on these forecasts, three investment portfolios were rebalanced monthly in order to reduce their exposure to the worst performing asset class. The portfolios under consideration included income, capital growth and aggressive growth. The neural networks were trained using the back propagation algorithm. Overfitting is a potential problem in artificial neural networks. Data selection techniques and weight decay were used to prevent overfitting. Five financial indicators were used as inputs to the neural network model: S&P500 index, 10 year Treasury bill, University of Michigan Consumer Sentiment index, CPI and the Chicago Board of Trade volatility index (CBOE). The artificial neural network based rebalancing strategies were tested over a 24-month period. The results demonstrate that portfolio volatility could be reduced by 32% during the test period.

Important Links:

Go Back