1 摘要 量化投资起源于20世纪初，从开始的资产交易行为与预期股价有关，到股价和 市场收益能反映有效市场，再到市场波动风险与预期收益呈正比关系，这一系列的 研究结论，为量化投资策略之后的研究工作提供了更多的可能。近年来，国内外的 证券市场发展地如火如荼，如果想要更好地实现量化投资策略，使用传统的研究方 法，在面对市场频繁波动时，则容易错失投资机会。而RNN-LSTM模型可以利用已有 的历史数据以及自我学习，可以加快投资策略的计算速度以及研究的机率提升。因 此对证券市场的投资管理领域的研究，具有一定的现实意义。 目前的量化投资策略研究，较少采用基本类因素和技术类因素结合的方式进行 研究。并且现有研究在投资策略构建成功后，极少对投资策略的不足进行优化以及 对优化后的模型再进行对比分析。即目前的研究文献表明，较少有学者对完整的量 化投资策略的研究工作。本文提出了一种基于多因素模型检验和RNN-LSTM模型预测 的完整量化投资策略,首先对120个基本类因素进行多次检验，对检验结果进行综合 分析后，得到5个有效的影响因素，然后使用这5个影响因素在A股市场中进行计 算筛选，将筛选得到的股票建立股票池。在待选股票的基础上，通过运用其历史交 易数据对RNN-LSTM模型进行训练，得到损失函数曲线值稳定在0.69左右，以及预 测准确率为61.33%，使用该模型对股票池中的股票进行涨跌预测以及回测交易，将 该投资策略的交易结果与沪深300的指标进行对比，对比发现，该投资策略的表现 优于沪深300的表现。 鉴于投资策略结果中的风险评价不够理想，考虑在投资策略中增加AH股溢价率 和AH股的方法，来降低策略组合风险。其中AH股溢价率是作为影响因素来筛选股 票，由于交易规则以及市场定位的不同，同一家公司在AH股两个市场上的股价存在 价差，利用股价价差进行策略优化。实证结果表明，加入AH股优化后的策略风险明 显比未加入的策略风险降低了，最大回撤率从21.5%降低至8.5%，收益波动率从30.8% 降低至18%，总收益率提升了10.3%，夏普比率提升近2倍，该实证结果说明基于多 因素选股模型和RNN-LSTM模型预测的量化投资策略是有效的，该结果旨在为量化基 金管理人提供投资策略思路，在面对投资市场中降低风险的同时，获取有效收益。 关键词：量化投资；机器学习；风险管理 基于RNN-LSTM模型的A股市场量化投资策略研究 I Abstract The quantitative investment began in the early 20th century, first of all, asset trading behavioris related to the expected share price, andthen, through the stock market gains and can reflect the effective market, the risk of market fluctuation is in direct proportion to the expected return, this series of research conclusion, for quantitative investment strategy after research work offers more possibilities .In recent years, the stock market at home and abroad has been developing like a raging fire,in order to better realize the quantitative investment strategy, it is easy to miss the investment opportunity in the face of frequent market fluctuations by using traditional research methods.The RNN-LSTM model can make use of the existing historical data and self-learning, which can speed up the calculation of investment strategy and improve the probability of research,it is of practical significance to study the investment management of securities market. At present, the quantitative investment strategy research mainly adopts the method of single test of fundamental factors or technical factors, but seldom adopts the method of combination of basic factors and technical factors.Moreover, after the successful construction of investment strategy, existing researches rarely optimize the deficiencies of investment strategy and make comparative analysis on the optimized model.In general, according to the current research literature, few scholars have studied the complete quantitative investment strategy.Therefore, this paper proposes a complete quantitative investment strategy based on Multi-Factor model test analysis and RNN-LSTM model prediction.First, the 120 basic class factors were tested for several times. After A comprehensive analysis of the test results, the 5 effective influencing factors were obtained. The 5 influencing factors were used to calculate and analyse the A-share market, and A stock pool was established for the screened stocks.On to pick stocks, on the basis of using the historical transaction data for training RNN-LSTM model, get the loss function curve value stable at around 0.69, and the forecast accuracy of 61.33%, use the model or projections for shares in the pool and back-test the transaction, the investment strategy of trading results compared with the CSI 300 index, the contrast found strategy better than the performance of the CSI 300 index. Since the risk evaluation results in the investment strategy are not good, this paper want to increase the premium rate of AH Shares and AH Shares in the investment strategy to reduce the risk.Among them, the premium rate of AH Shares is used as the influencing factor to screen the shares. Due to the different trading rules and market positioning, the price difference of the same company in the two markets of AH Shares exists, so the price difference can be used to optimize the investment strategy.The empirical results show that after joining AH optimize the risk of the investment strategy is clearly not to join the investment strategy of risk reduction,maximum loss rate reduced from 21.5% to 8.5%,earnings volatility is reduced from 30.8% to 18%,total yield of 10.3%,sharpe ratio increase nearly 2 times, the result shows that model based on Multi-Factors to pick stocks and RNN-LSTM models predict quantitative investment strategies are effective.The empirical results is designed to provide investment strategy thinking for quantitative fund managers, at the same time, to reduce risk in the investment market can also take the effective yield. Key words: Quantitative Investment;Machine Learning;Risk Management 基于RNN-LSTM模型的A股市场量化投资策略研究 1 目录 导 论 ································································································ 1 一、研究背景和意义 ·········································································· 1 二、国内外相关文献综述 ···································································· 3 三、本文的研究思路 ·········································································· 8 第一章 相关概念及理论 ···································································11 第一节投资理论基础 ············································································11 一、价值投资理论 ············································································11 二、现代资产组合理论 ····································································· 12 三、资产定价模型 ··········································································· 12 四、资产组合的风险评价 ·································································· 14 五、程序化交易 ·············································································· 15 第二节机器学习理论 ··········································································· 15 一、RNN模型 ················································································ 16 二、LSTM模型 ··············································································· 18 第二章 多因