Short-term financial market prediction using an information-theoretic approach

James Kar
Portland State University

Martin Zwick
Portland State University

Beverly Fuller
Portland State University

     Full text: Not available
     Last modified: April 25, 2006

Abstract
Financial market prediction has been researched extensively, most commonly with linear methods. In this paper, we present a technique based on information theory for short-term prediction using discrete time series data binned into increase, decrease, or stay-the-same categories. Calculations were done with a Reconstructability Analysis software package. The technique orders the predicting power of the input variables and produces a predictive model (a set of conditional probabilities). Models are selected by statistical significance and the Akaike and Bayesian Information Criteria and are assessed by %uncertainty reduction of the dependent variable and the %correct of the predictions. The technique is successful in forecasting the change in the Standard and Poor’s 500 index for at least one to two days in the future using several other financial indicators.