量化交易中股票择时的策略研究

发布时间：2018-01-19 13:20

本文关键词： 主成分分析 ARIMA-GARCH 神经网络遗传算法　出处：《天津商业大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来,中国的经济快速增长,证券市场不断扩大,股票市场迎来空前的繁荣期,投资者面临着前所未有的机遇,同时也面临着高风险和高挑战。庞大的股民数量和便捷的交易方式,使得股市市场产生了大量有价值的信息,每个交易日都有着海量的数据流通,这些数据构成了投资者进行投资分析的主体之一。在中国股市市场,基于计算机和大数据发展起来的量化投资策略逐渐兴起,量化模型日益成为指导市场投资的主流工具。与此同时,股票市场本身具有的非线性和复杂性等特点,使得传统的投资策略很难达到人们的期望,而人工神经网络具有的较强的非线性逼近能力和其自学习、自适应等特性,使得其在股市预测方面的优越性日益凸显。本文选取了自2012年11月21日至2016年12月30日共1000个交易日、15个指标的上证综合指数数据为研究对象,首先通过主成分分析法对输入指标进行降维处理,最终得到5个主成分变量,然后建立了针对上证综合指数的时间序列ARIMA-GARCH模型和BP神经网络预测模型。然而,BP神经网络固有的一些缺陷,如学习收敛速度太慢、网络结构不易确定等,制约了其预测的精度和速度,因而本文又利用遗传算法来对神经网络进行优化,建立了遗传算法优化的神经网络模型。最后,论文选取相对误差为衡量指标,对模型结果进行对比分析。论文研究结果表明,神经网络在股价预测精度方面显著优于ARIMA-GACH模型,且经遗传算法优化的神经网络较之一般的神经网络在预测速度和精度方面也有所提高。
[Abstract]:In recent years, with the rapid growth of Chinese economy and the expansion of the stock market, the stock market is facing an unprecedented boom period, and investors are facing unprecedented opportunities. At the same time, it also faces high risk and high challenge. The huge number of shareholders and convenient trading methods make the stock market produce a lot of valuable information, every trading day has a huge amount of data circulation. These data constitute one of the main bodies of investors' investment analysis. In the Chinese stock market, the quantitative investment strategy based on computer and big data is emerging. The quantitative model is becoming the mainstream tool to guide the market investment. At the same time, the nonlinear and complexity of the stock market makes it difficult for the traditional investment strategy to meet the expectations of people. The artificial neural network has strong nonlinear approximation ability and its self-learning, adaptive and other characteristics. This paper selects 1000 trading days from November 21st 2012 to December 2016. Shanghai Composite Index data of 15 indexes as the research object, first through the principal component analysis to reduce the dimension of the input indicators, finally get 5 principal component variables. Then the time series ARIMA-GARCH model and BP neural network prediction model for Shanghai Composite Index are established. However, there are some inherent defects of BP neural network, such as slow learning convergence rate. The network structure is not easy to determine, which restricts the precision and speed of prediction. Therefore, the genetic algorithm is used to optimize the neural network, and the neural network model of genetic algorithm optimization is established. Finally. The relative error is chosen as the index to compare and analyze the model results. The results show that the neural network is significantly better than the ARIMA-GACH model in the accuracy of stock price prediction. The prediction speed and precision of the neural network optimized by genetic algorithm are also improved compared with that of the normal neural network.
【学位授予单位】：天津商业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.51

【参考文献】