Designing Translation Invariant Operators for Financial Time Series Forecasting

This work presents an adaptive evolutionary method for designing translation invariant operators, via Matheron decomposition by dilations or erosions and via Banon and Barrera decomposition by sup-generators or infgenerators, for financial time series forecasting. It consists of an intelligent adaptive evolutionary model composed of a modular morphological neural network (MMNN) and an adaptive genetic algorithm (AGA), which searches for the minimum number of time lags (and their corresponding specific positions) to represent the time series and the weights, architecture and number of modules of the MMNN. An experimental analysis is conducted with the proposed method by using two real world financial time series, and the experimental results are discussed according to five performance measures.