Forecasting Austrian IPOs: An application of linear and neural network error-correction models

This paper applied cointegration and Granger-causality analyses to construct linear and neural network error-correction models for an Austrian Initial Public Operings IndeX (IPOXATX). It uses the significant relationship between the IPOXATX and the Austrian Stock Market Index ATX to forecast the IPOXATX. For prediction purposes it applies augmented feedforward neural networks whose architecture is determined by Sequential Network Construction with the Schwartz Information Criterion as an estimator for the prediction risk. Trading based on the forecasts yields results superior to Buy and Hold or Moving Average trading strategies in terms of mean-variance considerations.