Title: Comparing and combining neural networks for stock market direction prediction
Authors: Daniel Grabowski - Universitaet Giessen (Germany) [presenting]
Abstract: Artificial neural networks (ANN) have recently proven extremely successful in a variety of tasks, including forecasting. Their appearance in the econometrics and financial literature is, however, still relatively sparse. To fill this gap, different types of neural networks are applied to stock market direction prediction. The performance of these neural networks is compared to standard econometric as well as machine learning methods. Different neural network architectures and different activation functions are considered. This includes simple single-layer ANN as well as recurrent neural networks, which are better able to capture the time series structure of financial data. The performance of nonlinear models has been shown to benefit from forecast combination. Consequently, the combined forecasts of the different neural networks are also evaluated.