Abstract: Stock price prediction is essentially a problem of data mining. The market trend is a good signal for stock trading timing. In quantitative analysis, deep learning technology is often used to fit and extract the characteristics of the market historical data, so as to provide decision-making references for stock investment. In this paper, a classical deep neural network is presented and trained to implement supervised learning based on the daily candlestick chart's volume and price data of CSI 300; a binary predictor is realized to output the ''rise'' or ''fall'' labels and simulate the transaction with the strategy of the predictor; and then the test data set is utilized to calculate the cumulative rate of return, so as to estimate the advantages and disadvantages of the investment strategy. In addition, another hybrid quantum-classical neural network model is constructed, it makes full use of the characteristics of the circuit model of quantum computing to form the parameterized variational quantum circuit and realize the quantum implementation of feedforward neural network. In the frame of quantum circuit learning, the stock indicators are encoded into the amplitude of quantum states and the parameters θ of the quantum neural network U are trained, and finally an optimal classifier is obtained after iterations. The results show the characteristics of strong expression and high robustness: the run-time of the quantum algorithm is 7.7% shorter than that of the classical algorithm and the prediction accuracy is higher, resulting in a return rate advantage of 3%.