Forecasting S&P and gold futures prices: An application of neural networks

✍ Scribed by Gary Grudnitski; Larry Osburn

• Publisher: John Wiley and Sons
• Year: 1993
• Tongue: English
• Weight: 732 KB
• Volume: 13
• Category: Article
• ISSN: 0270-7314
• DOI: 10.1002/fut.3990130605

✦ Synopsis

We would like to acknowledge N. Marovac and R. Swiniarski of the Department of Mathematical Sciences at San Diego State University for introducing us to neural networks. Also, the helpful comments of E. Ornberg and N. Varaiya of the Department of Finance, San Diego State University, and two anonymous referees are gratefully acknowledged. Last, we would like to thank the School of Accountancy, San Diego State University for providing financial support for this project. 'White (1988) used neural networks in an attempt to predict daily IBM stock prices. Harston (1990, p. 396) made the following comments on White's efforts:

Unfonunately, the results were disappointing. In some ways, this result could have been expected. After all, neural networks can only process information, make data transformations and detect patterns. They cannot make up something from nothing. Where no information exists, neural networks cannot magically find meaning. 'Presumed here is that traders' beliefs lead to actions and market prices reflect those actions. Therefore, by analyzing patterns of traders' beliefs, discernible trends in prices of specific futures can be forecast. Inclusion of variables that reflect beliefs or expectations in a neural network forecasting model of trading behavior is warranted by the recent works of Zaremba (1990) andCollard (1991).