Stock index returns forecasting using a hybrid neural network model

Stock returns are likely to consist of linear parts as well as nonlinear parts over time. Autoregressive or moving average models may be suitable for capturing linearly behaving parts, whereas nonlinear smooth transitioning or artificial neural network models can be used to describe the nonlinear habits. In general, none of the aforementioned individual models is able to capture both linearity and nonlinearity of time series completely. To overcome this, my study combines two models from the linear and nonlinear field into a hybrid model to describe and forecast the returns on the Spanish Ibex-35 stock index. Recursive and nonrecursive one-step-ahead and multi-step-ahead forecasting methods are utilised over three forecast horizons to measure and compare the performance of the hybrid model relative to the individual models. I find that the hybrid model is able to surpass the individual models for short horizon one-step-ahead forecasts. The hybrid model seems particularly useful for predicting the correct sign of the returns, which indicates that the combined model can be of interest to incorporate in trading strategies.