| The artificial neural networks represent an exciting new technology with wide scope for potential financial applications ranging from routine credit assessment operations to the driving of large scale portfolio management strategies. This study is an attempt to identify stock winners and losers using this technology. Two sets of attributes are used for the network training. The first set is modelled along the lines of attributes that are used by Value Line Investment Survey, whereas the second set of attributes is a combination of the first set and some attributes taken from the anomaly literature.;The evidence presented in this study suggests that a blind, black box approach to artificial neural networks may not be very useful. Some pragmatism and careful knowledge of data is required for good results. A design that takes into consideration issues such as, sample size requirements, choice of attributes, presence of influential data points in the input data and use of classical techniques, such as multiple discriminant analysis, to enhance the network performance, may result in a potentially good system; it can even compare favourably with the classical methods, but with some caveats. |
