基于支持向量机的量化择时策略及实证研究

发布时间：2018-04-28 00:27

本文选题：择时策略 + 支持向量机　；参考：《西安工业大学》2017年硕士论文

【摘要】：谷歌Alpha Go带来的人工智能的风暴,正在横扫各个行业,同样也会对金融投资行业产生深远的影响。而现实中量化投资和程序化交易,已经成为很多金融市场中机构投资者的常规操作模式。量化投资以其理性客观、决策效率高、信息处理能力强等特点越来越受到学术界与投资实务界的重视。而量化择时策略是量化投资策略的一个重要分支。支持向量机(SVM)是一种机器学习算法,弥补了传统神经网络学习算法的多项不足,在解决模式识别和回归问题时性能优越。对于SVM,国内在金融领域的研究主要用于金融时间序列预测,还没有与量化择时策略相结合的研究,而且在研究的过程中主要是侧重对SVM方法和应用的研究,往往忽视了策略本身。针对于以上问题,本文通过研究现有的量化择时策略和SVM算法,结合两者的优势,构建基于SVM的量化择时策略。首先,本文介绍量化投资的相关概念,简要梳理量化投资在国内外的发展状况;给出量化择时策略的定义、分析其特点并对现有的量化择时策略进行了分类。其次,从机器学习、统计学习理论等六个方面对SVM的相关理论进行较为全面深入的研究。接下来,系统的构建基于SVM的量化择时策略,主要有两大部分,一是基于SVM择时策略的构建,二是策略模型算法的设置。最后,运用中国石油、浦发银行、沪深300指数、中证500指数和创业板指指数的各600组、时间跨度约两年半的数据进行训练与测试,分析验证策略的有效性。本文研究的创新性工作主要有两方面:一是对于量化择时策略进行了系统的梳理,并建立了自己的量化择时策略。本文量化择时策略的思路是:策略选择在我国股票市场运行,SVM预测模型每日收盘后运行一次,对下一日收盘价进行预测,如果预测出上涨,在当下一日的价格低于前一日收盘价时,全仓买入;如果预测出下跌,当下一日的价格高于前一日收盘价时,清仓卖出;如果预测出没有变化,就不进行操作,同时加入了止损判断,也就是说,每日只进行一次交易或不进行交易,整个过程由交易系统自动进行。二是引入支持向量机优化算法,系统地构建和检验了量化择时策略。采用SVM算法,可以将量化择时策略进行优化,取得更好的投资效果。在基于SVM择时策略的构建部分,本文从择时模型设计的总体思路、预测期限、预测目标、投资范围、特征指标、买卖时点、模型设置这七个方面构建了择时策略。在策略模型算法的设置部分,本文对SVM算法以及整个模型算法的各个方面进行具体的设置,主要包括SVM的多分类算法选择、SVM核函数选取、参数寻优、不平衡数据的处理、滚动预测这五个方面的内容。通过研究,本文构建了基于支持向量机的量化择时策略;使用真实数据进行实证检验。通过对模型预测能力的分析、与买入持有策略的对比,以及从不同市场行情下的表现和策略的各项评价指标来看,本论文的量化择时策略表现优异,所构建的基于SVM量化择时策略是有效的。本论文的研究对于将支持向量机方法应用于量化投资策略的构建,对于完善和优化量化择时策略,对于量化投资实践具有一定的指导和参考意义。
[Abstract]:The storm of artificial intelligence brought by Google Alpha Go is sweeping across all industries, and it will also have a far-reaching impact on the financial investment industry. In reality, quantitative investment and procedural transactions have become the conventional mode of operation for institutional investors in many financial markets. Quantitative investment is rational and objective, efficient in decision-making, and information processing. The characteristics of ability and ability are becoming more and more important in academic circles and investment practice circles. Quantitative timing strategy is an important branch of quantitative investment strategy. Support vector machine (SVM) is a kind of machine learning algorithm, which makes up many shortcomings of traditional neural network learning algorithm, and has superior performance in solving pattern recognition and regression problems. For SVM, The domestic research in the financial field is mainly used in the financial time series prediction, and there is no research on the combination of the quantitative timing strategy, and in the process of the study, the main focus is on the study of the SVM method and application, often ignoring the strategy itself. In view of the above problems, this paper studies the existing quantitative timing strategy and SVM algorithm. Combining the advantages of the two, this paper constructs a quantitative timing strategy based on SVM. Firstly, this paper introduces the related concepts of quantitative investment, briefly combs the development of quantitative investment at home and abroad, gives the definition of quantitative timing strategy, analyzes its characteristics and classifies the existing quantitative timing strategies. Secondly, from machine learning and statistical learning theory. In the following six aspects, a more comprehensive and in-depth study of the related theories of SVM is carried out. Next, the system builds a quantitative timing strategy based on SVM, including two major parts. One is based on the construction of the SVM timing strategy and the two is the setting of the strategy model algorithm. Finally, it uses CNPC, Pufa Bank, Shanghai and Shenzhen 300 index, CSI 500 index and entrepreneurship. The 600 groups of the index index, the time span of about two and a half years of data training and testing, analysis and validation of the effectiveness of the strategy. The innovative work of this study mainly has two aspects: first, the quantitative timing strategy is systematically combed, and the establishment of their own quantitative timing strategy. This paper quantifies the strategy of timing strategy is: Strategy Choose to run in our stock market. The SVM forecast model runs once a day to predict the closing price of the next day. If the price is predicted to rise and the price of the next day is lower than the closing price of the previous day, the whole warehouse is bought. If the forecast is down and the price of the next day is higher than the closing price of the previous day, the warehouse will be sold out; if predicted, if it is predicted Without change, the operation is not carried out, and the stop loss judgment is added, that is to say, only one transaction or no transaction is carried out every day. The whole process is automatically carried out by the transaction system. Two, the support vector machine optimization algorithm is introduced, and the quantitative timing strategy is constructed and tested systematically. The SVM algorithm can be used to optimize the timing strategy. In the construction part of the SVM timing strategy, this paper constructs the timing strategy from seven aspects: the overall idea of the timing model design, the prediction period, the forecast target, the investment scope, the characteristic index, the time point of the sale and the model setting. In the setting part of the strategy model algorithm, this paper is on the SVM algorithm and the whole model. All aspects of the algorithm are set up, mainly including the selection of SVM multi classification algorithm, SVM kernel function selection, parameter optimization, unbalance data processing, rolling prediction. This paper constructs a quantitative timing strategy based on support vector machine, and uses real data to verify the five aspects. The analysis of model forecasting ability, compared with the buying and holding strategy, and the evaluation indexes of performance and strategy in different market quotations, the quantitative timing strategy of this paper is excellent. The construction of the SVM quantization timing strategy is effective. The research of this paper applies the support vector machine method to quantitative investment. The construction of strategy is of guiding and referential significance for improving and optimizing quantitative timing strategies and quantifying investment practices.

【学位授予单位】：西安工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.51

【参考文献】