improving the performance of financial forecasting using different combination architectures of arima and ann models

Despite several individual forecasting models that have been proposed in the literature, accurate forecasting is yet one of the major challenging problems facing decision makers in various fields, especially financial markets. this is the main reason that numerous researchers have been devoted to develop strategies to improve forecasting accuracy. one of the most well established and widely used solutions is hybrid methodologies that combine linear statistical and nonlinear intelligent models. the main idea of these methods is based on this fact that real time series often contain complex patterns. so single models are inadequate to model and process all kinds of existing relationships in the data, comprehensively. in this paper, the auto regressive integrated moving average (arima) and artificial neural networks (anns), which respectively are the most important linear statistical and nonlinear intelligent models, are selected to construct a set of hybrid models. in this way, three combination architectures of the arima and ann models are presented in order to lift their limitations and improve forecasting accuracy in financial markets. empirical results of forecasting the benchmark data sets including the opening of the dow jones industrial average index (djiai), closing of the shenzhen integrated index (szii) and closing of standard and poor’s (s&p 500) indicates that hybrid models can generate superior results in comparison with both arima and ann models in forecasting stock prices.