动态神经网络在量化投资预测中的应用

发布时间：2018-05-16 04:03

本文选题：量化投资 + 动态神经网络　；参考：《复旦大学》2014年硕士论文

【摘要】：随着金融学理论和计算机技术的发展,基于数据和规则的量化投资策略在中国已逐渐兴起,量化模型成为了预测市场和指导投资的有力工具。然而证券市场是一个复杂的非线性动力系统,利用传统的时间序列预测技术存在很大的局限性,而近十几年发展起来的神经网络预测理论在对非线性系统的预测和建模中展现出了独有的优势。本文引入了基于时间序列分析的NARX动态神经网络模型并研究将其应用于中国股票指数的时间序列预测,并将其同传统的时间序列预测模型作比较。通过对中国股市沪深300指数数据所做的实证研究表明,动态神经网络模型用于预测中国股市指数序列结果的准确性优于ARIMA-GARCH模型和静态BP神经网络模型。在引入的NARX动态神经网络预测模型的基础上,创造性地将其应用于股指序列的模式分类来判断股价指数是否处于阶段的“顶部”或“底部”,从而构建了基于动态神经网络的量化择时模型。实证结果表明,基于动态神经网络的模式分类准确性优于多元统计学的判别分析模型和基于普通静态神经网络的模式识别模型。在构建的该量化择时模型的基础上,进一步以此模拟了基于动态神经网络的积极型投资组合。通过计算和比较一系列组合投资业绩指标,结果表明各个业绩衡量指标结果均显示基于动态神经网络模型的投资组合优于市场指数组合,从而能成功将其应用于中国证券市场的量化投资。最后,本文通过研究行为金融的市场异常现象与动态神经网络模型有效性的相关关系,对其做统计检验,给出了动态神经网络模型有效预测能力来源和成因的解释。
[Abstract]:With the development of finance theory and computer technology, quantitative investment strategy based on data and rules has been rising gradually in China, and quantitative model has become a powerful tool to predict market and guide investment. However, the stock market is a complex nonlinear dynamic system, and the use of traditional time series prediction technology has great limitations. The neural network prediction theory developed in recent years has shown its unique advantages in the prediction and modeling of nonlinear systems. In this paper, the NARX dynamic neural network model based on time series analysis is introduced and applied to the time series prediction of Chinese stock index, and it is compared with the traditional time series prediction model. An empirical study on the CSI 300 index data of Chinese stock market shows that the dynamic neural network model is more accurate than the ARIMA-GARCH model and the static BP neural network model in predicting the results of the index series of the Chinese stock market. On the basis of the NARX dynamic neural network prediction model introduced, it is creatively applied to the model classification of stock index series to judge whether the stock price index is at the top or bottom of the stage. A quantitative timing model based on dynamic neural network is constructed. The empirical results show that the accuracy of pattern classification based on dynamic neural network is better than that of discriminant analysis model based on multivariate statistics and pattern recognition model based on general static neural network. On the basis of the quantitative timing model, the active portfolio based on dynamic neural network is simulated. By calculating and comparing a series of portfolio performance indicators, the results show that the portfolio based on dynamic neural network model is superior to the market index portfolio. Thus it can be successfully applied to the quantitative investment in China's securities market. Finally, by studying the correlation between the market anomaly of behavioral finance and the validity of dynamic neural network model, this paper makes a statistical test on it, and gives an explanation of the source and cause of effective prediction ability of dynamic neural network model.
【学位授予单位】：复旦大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP183;F832.51

【相似文献】