Steuer-Klientel-Effekte: Realität oder Illusion ?

In this article we examine how model selection in neural networks can be guided by statistical procedures such as hypotheses tests,information criteria and cross validation. The application of these methods in neural network models is discussed, paying attentionespecially to the identification problems encountered. We then propose five specification strategies based on different statisticalprocedures and compare them in a simulation study. As the results of the study are promising it is suggested that a statistical analysisshould become an integral part of neural network modeling.

Anders, U. and O. Korn (1996), Steuer-Klientel-Effekte: Realität oder Illusion ?, ZEW Discussion Paper No. 96-21, Mannheim.