Predictability of intraday stock index

Open, high, low, and close are four common features describing the intraday stock index. The predictability of three of them, namely high, low, and close, is studied based on the available prior information of open. Using linear regression and nonlinear back-propagation neural networks, the prediction error variance of high, low, and close are shown to be substantially lower by the effective modeling of open. Empirical evidences are given for the NASDAQ composite index and Hong Kong\´s Hang Seng Index, indicating that the observed facts should remain valid in other similar domains as well. The proposed linear and nonlinear models can effectively be used to give better prediction of high, low, and close by taking advantage of the causal "news" effect and strong correlation of open