Dynamic adjustment of stock position based on hybrid deep neural network

At the stock market, any investors will have to predict the overall trend of the market before making investment strategy so that their profits will be maximized. Such predictions reflect the “timing ability” of investors. Investors will only maximize economic benefits when they choose the best time to invest. As time series of stock price are related to each other, historical price trends can indicate the future direction of a stock, so they can be analyzed to forecast the closing price of the stocks. This paper regarded the stock historical data as a one-dimensional grid, in which samples were taken at fixed time intervals. By extracting spatial features with convolutional neural network, obtaining temporal features with long short-term memory network, and using the attention mechanism in natural language processing, the paper outlines a hybrid deep neural network-based model designed to predict the position ratio and solve the problem of market timing. Next, the proposed method is compared with other prediction models, in which prediction indicators and trade simulation indicators are used to measure their performances. Results show that the model proposed by this paper is more accurate in predicting position ratio, and it has more significant effect as a timing method.