A Hybrid Method for Tuning Neural Network for Financial Time Series Forecasting

A.R.L. Junior and T.A.E. Ferreira (Brazil)


Neural Networks, Evolutionary Algorithms, Time Series Analysis, Forecasting, Financial Models


This paper presents an new hybrid method for financial time series prediction called GRASPES. It is based on the Greedy Randomized Adaptive Search Procedure(GRASP), which is a multi-start metaheuristic for combinatorial pro blems, and Evolutionary Strategies (ES) concepts for tun ing of the structure and parameters of an Artificial Neural Network (ANN). The work proposed here consists of an ANN trained and adjusted by GRASPES, which is capa ble to evolve the parameters configuration and the weights of the ANN, searches for the minimum number of relevant time lags for a correct time series representation and found an optimal or sub-optimal forecasting model. An experi mental investigation is conducted with the GRASPES with four real world financial time series and the results achieved are discussed and compared, according to five well-known performance measures, to other works reported in the liter ature, demonstrating the good performance of GRASPES for financial time series forecasting.

Important Links:

Go Back