Predicting reactions to anomalies in stock movements using a feed-forward deep learning network

Stock markets may be exposed to large movements for various reasons, which lead to negative or positive returns. This research aims to find if we can predict future prices after a shock? The problem is not concerned with the time series of stock prices but rather with stock price anomalies. Thus, a feed-forward deep learning network is more suitable for this problem than a recurrent neural network. We deploy deep learning mechanisms to predict stock prices at a given time following a significant movement given a few inputs. An experimental case study is implemented on stocks in the S&P 500 for particular variations of large movements. The results show that a feed-forward DNN can predict price anomaly reactions with a reasonable error margin. However, the mean absolute error is reduced when we aim to predict shorter periods. In addition, the prediction accuracy increases as the threshold are tightened.