Financial Transaction Forecasting using Neural Network and Bayesian Optimization

The stock price forecasting has been a well-known topic attracting many researchers and investors. How to buy in a low price and sell it in a high price is an eternal focus. Until now, there are two schools about the stock market forecasting: fundamental analysis and technical analysis. This paper is mainly about the technical analysis which utilized the historical time-series data of the market, such as transaction price and trading volume to predict whether we do financial transaction or not. In this paper, we propose a stock trading action selection prediction model which is based on neural networks and hyperparameter optimization methods. Our method makes predictions better by effectively analyzing attributes of different dimensions. Our trading behavior is evaluated on the Jane Street dataset provided by the kaggle competition. Experimental results show that our method achieves good performance compared with other machine learning methods. On the dimension of utility score we proposed to evaluate the performance of model, the unified score of our model is 2111 and 3179 higher than the results of ResNet and XGBoost models. In addition, we also studied to get the better hyperparameters with the Bayesian optimization algorithm.