量化投资在国外已有40多年的历史，但我国的量化投资起步晚、发展十分缓慢。随着计算机技术的发展，计算机的计算能力大幅提高，为量化投资奠定了计算基础；数据存储能力增强、数据存储量增大，为量化投资提供了数据基础；全球信息化高度发达，使得量化投资的信息更全面，决策更准确；金融市场逐步全球化，利用量化投资可以实现无缝化投资。我国现在主要运用传统的计量方法进行量化投资，但传统计量方法不能有效捕捉证券市场的非线性特征，而机器学习算法能够捕捉非线性特征，从而适用于证券市场的预测研究。通过研究机器学习在量化择时方法的应用效果，有助于推广机器学习算法在金融领域的应用，有利于完善金融体系的理论。 本文将机器学习算法应用于量化择时策略。构建量化择时模型的思路是：首先利用机器学习算法(SVR和神经网络模型)预测投资对象下一日的收盘价，预测模型在每日收盘之后运行一次；然后利用量化择时策略(趋势量化择时策略和Hurst指数量化择时策略)基于预测的收盘价确定买卖信号，如果出现买入信号，则在下一日开盘时全仓买入(开盘未涨停)，如果出现卖出信号，则在下一日开盘时全仓卖出，如果没有买卖信号，则下一日不进行任何操作。 本文首先对量化择时策略进行了系统梳理，然后介绍了常见机器学习算法，最后选取上证综合指数进行实证分析。实证数据的时间区间为2013年1月4日到2018年9月13日，利用ADF检验验证了上证综合指数序列非平稳，说明上证综合指数代表的我国证券市场未达到弱有效。 对不同的择时策略进行对比分析得出：当所选取的量化择时策略相同时，基于SVR的量化择时策略优于基于神经网络的量化择时策略；当选取的机器学习算法相同时，基于机器学习的趋势量化择时策略优于基于机器学习的Hurst指数量化择时策略。

Quantitative investment has been in foreign countries for more than 40 years. However, China's quantitative investment develop slowly. With the development of computer technology, the computing power has been greatly improved, providing a computing foundation for quantitative investment. The enhanced data storage capacity and increased data storage provide a data foundation for quantitative investment. Global information technology is highly developed, making quantitative investment information more comprehensive. With the globalization of financial markets, you can use quantitative investment to achieve seamless investment. China's existing quantitative investment methods are mainly traditional measurement methods, which can’t effectively capture the nonlinear characteristics of the securities market. Machine learning algorithms can capture nonlinear features, thus it’s suitable for applying to the forecasting research of the securities market. Based on machine learning algorithms, we constructed a quantitative timing strategy. Firstly, we use the machine learning algorithm (SVR and neural network model) to predict the closing price of the next day of the investment object. Secondly, we use the quantitative timing strategy (The trend quantization timing strategy and the Hurst exponential timing strategy) to get the buying and selling signal based on the predicted closing price. If there is a buying signal, the whole position will be bought at the opening of the next day. If the selling signal appears, the whole position will be sold at the beginning of the next day. If there is no trading signal, do not do anything on the next day. Firstly, this paper introduces the theory of quantitative timing strategies. Secondly, we introduce the theory of machine learning algorithms. Finally, we selects the Shanghai Composite Index for empirical analysis. The time interval of empirical analysis data is from January 4, 2013 to September 13, 2018. The ADF test is used to verify that the Shanghai Composite Index sequence is not a random walk sequence, indicating that the China’s securities market represented by the Shanghai Composite Index may not reach the weak position effectively. By comparing different timing strategies, it is concluded that when the selected quantitative timing strategy is the same, the quantitative timing method based on SVR model is better than the quantitative timing method based on neural network model. When the selected machine learning algorithm is the same, the quantitative timing method based on trend quantization timing is superior to the Hurst exponential timing strategy.