基于深度学习算法的股价趋势预测毕业论文
2022-01-16 21:02:34
论文总字数:34402字
摘 要
本文主要为了研究深度学习尤其是LSTM算法在股票价格预测问题上的能力,先是对深度学习和各种算法作出简单介绍,然后着重介绍了本文所运用的LSTM算法,如核心思想,推导过程及应用,最后借助bigquant平台中相应模块通过组合,优化参数,添加算法实现股价趋势预测策略模型,使用此策略模型针对沪深300成分股进行训练与回测。训练与回测具体的是针对2018年1年的沪深300成分股数据进行训练,得出预测最好的前20个股票;其后采用LSTM算法构建了择时交易模型,并在接下来的1个月内,只对这20个股票进行投资交易,以000300.SHA为基准代码,进行收益率对比,以此来判断模型优劣并做了模型改进与验证;之后做了泛化分析,主要看在相同训练及预测时间段,不同股票代码的情况下,此模型是否适用。
此策略月收益率3.67%(年化收益率61.34%)比较可观,回撤率为3.5%,收益波动率21.69%,说明该策略收益较平稳。此模型有一定的泛化能力,但也有局限性,对创业板泛化能力不理想。
关键字:深度学习 BigQuant平台 股票价格预测 LSTM算法 择时交易模型
Stock price trend prediction based on deep learning algorithm
ABSTRACT
In order to study the ability of in-depth learning, especially LSTM algorithm in stock price prediction, this paper first gives a brief introduction to in-depth learning and various algorithms, then focuses on the LSTM algorithm used in this paper, such as the core idea, derivation process and application. Finally, it realizes stock price trend by combining, optimizing parameters and adding algorithms with the corresponding modules in bigquant platform. The forecasting strategy model is used to train and retest 300 stocks in Shanghai and Shenzhen. The training and retest are aimed at the data of Shanghai and Shenzhen 300 component stocks in 2018, and the best 20 stocks are obtained. Then the LSTM algorithm is used to construct the timing trading model. In the next month, only 20 stocks are invested and traded, and the return is compared with 000300.SHA as the benchmark code, in order to judge the model's quality and make model modification. After that, the generalization analysis is made, mainly depending on whether the model is applicable in the same training and prediction period and different stock codes.
The monthly rate of return of this strategy is 3.67% (annual rate of return 61.34%) and the withdrawal rate is 3.5%, and the volatility rate of return is 21.69%. This shows that the return of this strategy is relatively stable. This model has some generalization ability, but it also has limitations. It is not ideal for GEM generalization ability.
Key words: in-depth learning; BigQuant platform; stock price prediction; LSTM algorithm; timing trading model
目 录
摘要 I
ABSTRACT II
第 1 章 引言 1
1.1 股票价格预测意义和目的 1
1.2 来源和研究背景 1
(1)第一次热潮(20世纪50,60年代) 1
(2)第二次热潮(20世纪80,90年代) 1
(3)第三次热潮(2006年至今) 2
1.3 国内外研究现状 2
第 2 章 关于深度学习和LSTM算法 3
2.1 深度学习 3
2.1.1 神经网络 4
2.1.2 卷积神经网络 (CNN) 5
2.1.3 递归神经网络(RNN) 5
2.1.4 其它 6
2.2 长短期记忆网络算法(LSTM) 7
2.2.1 LSTM概述 7
2.2.2 LSTM背后的核心思想 9
2.2.3 LSTM的推导 9
2.2.4 LSTM的应用 11
第 3 章 股票价格趋势预测模型构建 11
3.1 数据介绍 11
3.2 预测模型算法设计 12
3.2.1 LSTM的模型参数 12
3.2.2 神经网络结构 12
3.2.3 模型 13
3.3 建立股票预测和择时交易模型 13
3.3.1 选股 14
3.3.2 回测 15
3.3.3 其它主要参数配置 16
第 4 章 实验分析 17
4.1 模型回测检验 17
4.1.1 回测检验 17
4.1.2 收益概况 18
4.1.3 根据预测收益率的前20名排序 20
4.1.4 模型改进 21
4.1.5 模型验证 21
4.2 泛化分析 22
4.2.1 上证50 23
4.2.2 创业板 24
4.2.3 小结 25
第 5 章 总结 26
参考文献 27
致谢 30
附录 31
第 1 章 引言
1.1 股票价格预测意义和目的
十九世纪末二十世纪初,股票市场渐渐走近人们的视野。可是股市将如何变化包含着太多因素了,如经济的增长与衰退,国内外政治局势等等,还有很多人对股市的认知粗浅,以及经验不足,使得大多股民投资得不到回报,可热度从未有所减退。
请支付后下载全文,论文总字数:34402字