Multi-resolution time-series prediction using fuzzy inductive reasoning

This paper describes a new approach to multi-resolution prediction of time series using fuzzy inductive reasoning (FIR). The time series is decomposed into a trend series and another series describing the deviation from the trend. The two time series are then predicted independently of each other, and the two predictions are superposed in the end. The trend series is obtained by means of a moving average, whereas the deviation series is obtained by a process of de-trending using "daily return" calculations. The paper deals both with interpolation and with extrapolation problems.