基于非线性方法和VaR的均线交易系统研究

发布时间：2018-04-13 05:35

本文选题：支持向量机 + 非支配解　；参考：《哈尔滨工业大学》2014年博士论文

【摘要】：世界范围内的金融市场正处于一个迅速发展的历史时期，近年来，交易系统的研究与应用在国外得到了快速的发展，在交易中占有越来越大的比重。根据纽约证券交易所公布的数据，截至2011年5月20日当周该交易所的日均程序化交易占比为28.6%。成功的交易系统能够产生稳定和超额的回报，根据美国权威交易系统评选杂志2011年发布的交易系统排名，前三名模型年收益率均在200%以上。交易系统通常包含有追求最大收益率的阿尔法模型和以控制风险敞口规模为主的风险控制模型。但当前国内外对交易系统的研究主要倾向于交易信号的设计和挖掘，试图以一致的方法在任何的趋势中获利,没有注意到交易系统本身所存在的缺陷。在对交易系统的风险控制模型的研究较少，将阿尔法模型和风险控制模型两者结合在一起的研究尚不多见。为解决上述研究存在的不足，本文以传统技术分析中的均线交易系统为基础，使用支持向量机（SVM）、多目标优化算法中的非支配解和风险管理的VaR方法，构建了交易系统中重要的两个模型：阿尔法模型和风险控制模型，形成了基于非线性方法和VaR的均线交易系统。传统的均线交易系统在趋势市场中具有明显的赢利效应，但在横盘市场中却反复亏损。针对交易系统的这个缺陷，本文首先利用SVM分类器对市场进行趋势识别，使用RAVI等5种趋向技术指标将股票价格时间序列映射到高维特征空间，构建了支持向量机分类器对趋势进行分类和过滤，对不利于均线系统交易的横盘趋势进行过滤（空仓），以上证指数为研究对象，将5-60日均线作为基本参数，改进基于趋势跟随的均线交易系统，建立了基于SVM分类器的均线交易系统。在这个基础上，进一步优化参数。在参数优化过程中，为防止出现参数的过度拟合，将交易系统中常用且重要的两个评价指标，最大收益与连续最大回撤作为目标，使用了多目标优化算法中的非支配解的方法。经过优化，完成了对交易模型中的一个重要的组成部分-阿尔法模型的构建。为建立风险控制模型，本文以5-60日均线交易系统为研究对象，建立了非特定时间动态VaR模型。用蒙特卡罗方法产生了近3000个交易收益率数据、分析了非特定时间动态VaR收益率分布特征，并进行了模型准确性检验；在使用非特定时间动态VaR模型进行风险管理后，，研究结果表明可以优化交易策略。因此研究完成了对非特定时间动态VaR模型-风险控制模型的构建。最后将阿尔法模型与风险控制模型组合起来，构建了基于非线性方法和VaR的均线交易系统。为了将非特定时间动态VaR模型引入，首先使用威尔科克森秩和检验的方法验证了使用SVM前后，交易系统所生成的收益率序列的VaR值在置信条件下是没有统计差别的。然后通过对参数优化后的均线交易系统进行动态VaR建模求解。结果表明，基于非线性方法和VaR的均线交易系统可以有效地提高收益和降低风险。将非线性方法和VaR方法与投资交易相结合，有利于推动非线性科学在投资领域的应用，同时基于非线性方法和VaR的均线交易系统的构建也为投资者提供了一整套科学的投资方法，丰富了投资的研究方法，为程序化交易在中国股市的应用提供经验证据。
[Abstract]:The worldwide financial market is in a rapid development period in recent years, the research and application of the trading system has been rapid development in foreign countries, and play more and more important role in the transaction. According to the data released by the New York stock exchange, the daily program trading as of May 20, 2011 week the exchange ratio 28.6%.
The successful trading system can generate stable and excess returns. According to the authoritative trading system of the United States, the annual ranking of the top three models is over 200%, which is selected by the magazine in 2011.
The trading system usually contains a risk control model with Alfa model in pursuit of the maximum rate of return and to control the risk exposure of the size of the main design and mining. But the current research on the trading system at home and abroad mainly tend to trading signals, trying to consistent in any trend of profit, note that no defects of transaction the system itself. In the study of risk control model for the trading system will be less, the Alfa model and risk control model of them in combination with the research is still rare.
In order to solve the deficiency of the existing research, this paper is based on the average transaction system of traditional analysis, using support vector machine (SVM), a multi-objective optimization method of VaR non dominated solutions and risk management algorithm, constructs two important models of trading system: Alfa model and risk control model, formation the average transaction system based on VaR and nonlinear methods.
The average transaction system with traditional profit effect obvious trend in the market, but the market has repeatedly sideways loss. In order to overcome the defect of the trading system, this paper use the SVM classifier for recognition of the market trend, the use of RAVI and other 5 kinds of trends in technical indicators of stock price time series will be mapped into high dimensional feature space, construction the support vector machine classifier to classify and filter the trend, to filter the sideways trend is not conducive to the average system transactions (short), with the Shanghai Composite Index as the research object, the 5-60 day moving average as the basic parameters, the improved moving average trading system based on trend following, a moving average trading system based on SVM classifier.
On this basis, further parameter optimization. In the optimizing process, to prevent over fitting parameters, the trading system in common and important two evaluation index, the maximum income and continuous maximum retracement as the target, using the multi-objective optimization method of non dominated solution algorithm. After optimization, complete the construction of an important part of the transaction model of the Alfa model.
In order to establish the risk control model, based on the 5-60 day average trading system as the research object, to establish the non specific time dynamic VaR model. Using Monte Carlo method produced nearly 3000 trading return data, the analysis of non specific time dynamic VaR return distribution, and the accuracy of the model test in the use of non specific dynamic time; the VaR model of risk management, the results of the study show that can optimize the trading strategy. So the research done on the construction of risk control model of non specific time dynamic VaR model.
The Alfa model and risk control model together, constructs a nonlinear method and moving average trading system based on VaR. In order to introduce non specific time dynamic model of VaR, Kekesen will first use rank sum test method is verified using SVM before and after the trading system generated returns VaR values in confidence conditions there is no statistical difference. Then through the dynamic VaR modeling and solving of moving average trading system after optimization. The results show that the average transaction system nonlinear method and VaR can effectively improve the yield and reduce risk based.
The nonlinear method and the VaR method and the combination of investment transactions, to promote the application of Nonlinear Science in the field of investment, at the same time, based on average trading system nonlinear method and VaR also provides a set of scientific methods of investment for investors, enrich the research methods of investment, for program trading to provide empirical evidence in the application Chinese the stock market.

【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：F830.91;F224

【参考文献】