Modeling Movement Behavior of Stock Price Using Neural Hawkes Model

Traditional price movement is modeled by the machine learning methods and neural network methods. However, the prediction is often concerned with the correlations rather than the causality. In this paper, we do not only consider the correlation but also borrow the idea of the Neural Hawkes model to help build the decaying effects between the stock price dynamics. In the work, we evaluate the prediction quality of the results using the log likelihood. Results show that our methods are competitive, the Neural Hawkes model achieved log likelihood value of seq (combining the time and type) to -0.6358 and -2.3878 in five days prediction and ten days prediction respectively, better than -4.3243 and -4.5841 by Hawkes model and -11.353 and -24.8147 by Inhibition Hawkes model.