【作者】 王志刚；

【导师】 曾勇；

【作者基本信息】 电子科技大学 ， 管理科学与工程， 2009， 博士

【摘要】 现代资本市场理论与金融投资实践之间的重大分歧之一是，有效市场假说与技术分析之间的矛盾。本文在对技术分析有效性研究的早期和近期文献进行综述的基础上，发现尽管早期研究对技术分析持否定态度，但自上个世纪90年来以来的近期研究却得到大量支持技术分析的证据。这一研究结论的差异主要来源于两个方面，一是早期研究通常只考虑价格变化的线性相关模式，而近期研究则对真实的收益率动态过程进行更为深入的探讨；二是早期研究通常只检验几种常见的简单技术交易规则的获利性，而近期研究则对技术交易规则和交易策略进行更为审慎的考察。目前，对中国股票市场技术分析的现有研究仍然以早期研究的内容和方法为主，以此得到的研究结果并不足以评价技术分析的有效性和市场的有效程度。因此，本文综合运用近期研究所使用的各种统计工具、计量方法和建模理论，在拓展和改进国内外现有研究的基础上，对中国股票市场技术分析的预测能力与获利性进行系统和深入的研究。首先，结合Bootstrap方法与人工神经网络建模方法，考察真实收益率动态过程对评价技术分析有效性的影响。通过假设收益率过程服从各种常见的线性原假设模型和以人工神经网络方法建立的非线性原假设模型，对技术交易规则的预测能力进行Bootstrap检验，发现尽管各种线性过程均不能复制移动平均规则在真实价格变化过程中的收益特征，但非线性过程却可以很好地解释其预测能力和获利水平。因此，线性相关性并不足以评价技术分析的有效性。此外，基于异质市场假说的研究表明，收益率过程中的非线性相关性是不同类型投资者在不同时间水平上进行交易并相互作用的结果。其次，基于移动平均规则和交易量的历史信息建立神经网络模型并与各种线性模型的预测能力进行比较，同时应用thick modeling方法改进神经网络建模中不可避免的模型不确定性问题。研究发现，基于不同参数设定下的单个神经网络模型的thick models不但可降低单个模型的参数不确定性偏差，提高其统计预测精度，而且具有好于各种线性模型的预测绩效。并且，尽管各种移动平均规则本身均不能获利，但基于这些规则的thick models预测结果构造的交易策略却可获得显著高于买入持有策略的超额收益。最后，与中国股票市场现有研究主要针对简单技术交易规则不同，本文运用非参数核回归方法，对一些在实务界广受关注的复杂技术图形进行量化定义和识别检验。通过对技术图形的条件收益率与非条件收益率经验分布的差异性检验，发现这些技术图形具有可用于预测的额外信息含量；进一步考虑风险补偿后，发现基于一些复杂技术图形所预示的价格变化趋势构造的交易策略可获得显著的超额收益。更多还原

【Abstract】 One of the greatest gulfs between modem capital theory and financial practice isthe separation that exists between Efficient Market Hypothesis (EMH) and TechnicalAnalysis (TA).Based on a survey of the early and recent empirical literatures on testingthe efficiency of TA,the author finds that although most early studies denied theusefulness of TA in stock markets,a number of recent studies since 1990s indicated thatTA can generate profits.This reversal of conclusion can be explained in terms of twomain differences in the testing procedures.First,early studies tend to consider only thelinear correlation between price changes,while recent studies place greater emphasis onthe actual dynamic process of returns.Second,early studies usually test several simpletechnical trading rules,while recent studies investigate TA through a morecomprehensive scrutiny of various technical trading rules and strategies.However,most of the existing researches on testing the efficiency of TA in China’sstock market are still focused on the early studies and their methods.As a consequence,their empirical results do not provide an accurate assessment of the efficiency of TA andthe market itself.Accordingly,this thesis investigates the predictability and profitabilityof TA in China’s stock market by utilizing and extending some advanced statistic tools,econometric methods and modeling theories,and by dealing with some limitations thatexisted in recent studies.Firstly,we study how the actual return dynamics affect the assessment of theefficiency of TA by integrating the Bootstrap method and Artificial Neural Network(ANN).Assuming that the actual return process follows various popular linear nullmodels and a nonlinear model constructed by ANN,we test the predictability oftechnical rules through Bootstrapping.The empirical results indicate that,although noneof the linear processes can replicate the stochastic properties of returns obtained fromtechnical trading rules in the actual return dynamic process,the nonlinear process canexplain the predictions and profits of technical rules better.Thus,linear correlation isnot sufficient to evaluate the efficiency of TA.Besides,evidences based onHeterogeneous Market Hypothesis indicate that trading behaviors of different types of investors at different time horizons and their interreactions may cause return process topresent nonlinear correlation.Secondly,we construct a nonlinear model based on the past information containedin moving average rules and volumes,then compare its predictive power with somepopular linear models.Specifically,we apply the thick modeling method to try toresolve the inevitable problem of model uncertainty for ANN modeling.We find thatthick models based on individual ANN models with different parameters can not onlydecrease model uncertainty bias and improve the statistic prediction accuracy,but alsooutperform linear models according to the out-of-sample forecasting criteria.Moreover,despite the negative performance of moving average rules themselves comparing withthe buy-and-hold strategy,the technical trading strategies directed by the predictions ofthick models based on these rules can generate significant excess profits.Thirdly,in contrast with existing studies on China’s stock market which merelyfocus on testing some simple technical trading rules,we investigate some popularcomplex technical patterns among technical practitioners,which are quantitativelydefined and automatically recognized using nonparametric kernel regression method.By comparing the unconditional empirical distribution of daily returns to the conditionaldistribution,we find that most of the complex patterns can provide incrementalinformation that may be used to forecast further prices changes.Furthermore,theinformation contained in some patterns can generate significant excess trading profitseven after risk adjustment.更多还原