Abstract

Memory patterns do exist in timeseries data. Long-term or short-term predictionis possible by analysing memory patterns. The Hurst coefficient (H) is a statistical measure for predictability of time series. In this paper, memory patterns of financial data are analysedusing Hurst statistics. Experiments with radialbasis function (RBF) networks and multilayerperceptron (MLP) networks show that predictions in series with large H values aremore accurate than those with H close to 0.5.