Stock Prediction Based On Neural Network

Algorithmic trading can greatly reduce the impact of mood swings of decision makers on the decision making, so it is favored by foreign investors. With the increasing maturity of the securities market in China, algorithmic trading must be able to flex its muscles. Deep learning has achieved impressive results in image recognition and natural language processing field. The castle named finance will be captured by deep learning.Based on deep learning, this paper hopes to make a clear judgment on the future trend of the stock price. The specific research contents are as follow:Firstly, From the structure of the neural network, this paper expands deep learning how to solve many defects of BP neural network, then thinks about how to build a deep neural network by using the latest research results of deep learning like batch normalization, PRe LU and dropout.Secondly, recurrent neural network and its improved model named long-short time memory, as special model for processing sequence data. This paper builds a deep long-short memory model by using deep neural network and Long-short time memory.Thirdly, in empirical research, historical transaction data of Amazn in past ten days is used to predict stock trend of next day. Compared with deep neural network and long-short time memory with many different depth, this paper finds that deep neural network in the forecast stock price change can give some help, which achieves 54% of the forecast accuracy for three strategies which includes rising, concussion and dropping. It improves 21 percentage points compared to random guessing while 7 percentage points if guess a concussion. In contrast, long-short time memory in high hopes is inferior to deep neural network a lot.