Global stock index forecasting using multiple generalized regression neural networks with a gating network

This research area is the study of the effectiveness and utility of neural networks in accurately forecasting the nonlinear behavior of financial problem data and proposes a new neural network model to forecast the global stock indices.; This study introduces a neural network model, which comprises multiple Generalized Regression Neural Networks (GRNNs) and a gating network for financial forecasting. The GRNN has long been proven to be an effective tool for function approximation problems. This work adds the capability of adjusting the weights of all output from the multiple GRNNs. The modified architecture is suitable for the financial forecasting problem because the importance of each input can change as time changes. The model trains all input individually using GRNNs and uses a weight-updated gating network to improve the forecast accuracy. The results of the proposed model are compared with the single GRNN.; The model is implemented with 14 data attributes of stock market indices from 10 countries. The predictive abilities of the model are evaluated on the basis of mean squared percentage error, mean absolute percentage error, and direction accuracy; The proposed model shows a promising result of the forecastings in all ten countries. Further considerations of a greater variety of input and different network structures should be encouraged.