Improving Financial Returns using Neural Networks and Adaptive Particle Swarm Optimization

For financial investment, the problem that we often encounter is how to extract information hidden in the volatile and noise data and forecast it into future. This study proposes a novel three-stage neural-network-based nonlinear weighted ensemble model. In proposed model, three different types neural-network base models, i.e., Elman network, generalized regression neural network (GRNN) and wavelet neural network (WNN) are generated by three non-overlapping training sets, further, they are optimized by improved particle swarm optimization (IPSO) with adaptive nonlinear inertia weight and dynamic arccosine function acceleration parameters. Finally, a neural-network-based nonlinear weighted meta-model be produced by learning three neural-network base models through support vector machines (SVM) neural network. The superiority of the proposed approach is due to its flexibility to account for potentially complex nonlinear relationships that are not easily captured by single or linear models. Empirical results suggest that the novel ensemble model generally produces forecasts which, on the basis of out-of-sample forecast encompassing tests and comparisons of four different statistic measures, routinely dominate the forecasts from single modeling and linear modeling approach with two daily stock indices time series processes.