Index forecasting and model selection

In this paper we derive a trading strategy that exploits the informational difference implied by different stock market index construction principles. In order to gain a competitive advantage over other market participants we forecast the indexes one day ahead and subsequently generate buy and sell signals through the trading rule. To illustrate how the system works we apply it to select from those stocks that are included in the Austrian Traded Index (ATX). The forecasting of the indexes is performed on the basis of standard financial econometric techniques and feedforward neural networks. We discuss the importance of parsimonious modeling and the applicability of information criteria for architecture selection in artificial neural networks. [PERIODICAL ABSTRACT]