基于深度学习的股票价格预测研究- I - 摘 要随着经济的发展，中国股票市场的规模持续扩大，早已成为金融投资的重要部分，掌握股票市场的变化规律无论是对监管者还是投资者都具有极其重要的意义。正因如此，人们不断探索着股票市场的变化规律，其中使用深度学习预测股价是当前国内国际研究与应用的热点。本文首先从有效市场假说和分形市场假说两个角度讨论了中国股票市场的有效性，说明股票市场具有复杂的非线性特征。其次，结合股票市场特征对比了当前的预测方法，认为深度学习在股价预测中更具优势。接着，基于深度学习中的长短期记忆网络进行股价预测实验。通过对比试验，本文得出了长短期记忆网络在预测股价方面比三层全连接网络更有实际意义的结论，同时发现了多日数据作为输入变量较单日数据更加准确，增加训练数据在一定程度上能提高准确率，且模型的预测准确率能达到 68%。最后，本文重新选取了 10 支股票进行预测，以此进一步验证模型的效果。预测结果的平均准确率为 62%，且能为绝大多数股票带来了正向效益，说明了模型具有适应性，进一步证明了深度学习在股价预测方面的意义。关键词：股价预测；人工神经网络；深度学习；长短期记忆网络- II -AbstractWith the development of economy, the scale of China's stock market continues to expand, which has already become an important part of financial investment. It is of great significance for both regulators and investors to master the changing rules of the stock market.Firstly, this paper discusses the efficiency of China's stock market on the basis of efficient market hypothesis and fractal market hypothesis, which shows that the stock market has complex nonlinear characteristics. Secondly, combining with the characteristics of the stock market, this paper compares the current forecasting methods. Obviously deep learning has more advantages in stock price forecasting. Then, the stock price prediction experiment is carried out based on Long Short-term Memory Network (LSTM) in deep learning. The prediction results show that LSTM is more meaningful than the 3-laye...
