On the forecastability of share prices on the Viennese stock exchange

Abstract

In this paper the Viennese stock exchange data are analysed by using ARMA and GARCH technology. After using AIC and BIC for estimating the linear structure of the time series, to the resulting innovations a GARCH(1,1) model is fit. The resulting residuals are then tested for serial independence and constancy of its distribution to check whether the models are reasonable. Main result is that the residuals of this ARMA-GARCH(1,1)-model are reasonably iid (which is checked by BDS and classical independence tests) for index data and significantly less well-behaved for stock data. Second, there is considerable autocorrelation in the data (especially in the Viennese indices WBK and ATX) which can be exploited even with 1.25% transaction costs (which is checked by a posteriori analysis of a strategy which exploits an underlying time-varying AR(1) model), however, much higher profit can be made with 0.5% transaction costs. Furthermore, the same techniques are applied to US Standard & Poor 500 index and the results for both data sets are compared giving the result that the US-market looks much more “mature” than the Viennese one.