基于主成分—神经网络的农产品期货预测研究及模型实现
摘要
农产品期货交易成功的关键,在于对市场的正确分析,尤其是对市场价格走势的正确预测。价格是期货交易的焦点,是对未来供需关系变化预期的一种提前反映,交易者在期货市场中的盈亏状况由交易者对价格走势的判断来决定。作为解决我国“耕者保其利”社会难题的重要载体,农产品期货市场对农业生产经营者提供了较准确的远期价格指导信息,能减少生产的盲目性。因此,对农产品期货价格的分析和预测是很有意义的。
期货市场是一个复杂的非线性动态系统。当前,国内外期货价格预测的研究方法多为主成分分析法(PCA)、移动平均法、指数平滑法等传统的经济预测方法,以及误差反向传播神经网络(BP)、模糊神经网络、概率神经网络等人工神经网络方法。实际运用中,传统单一预测方法操作简便,但预测的准确率并不太高。利用传统的时间序列预测方法很难揭示其内在的规律。
本文在分析考察传统预测分析方法的基础上,提出一个面向农产品期货市场分析、预测,由神经网络和主成分分析法组合的新的预测方法,并针对该新方法性能的改善和提高进行了深入研究。将主成分分析法引入期货市场预测,对神经网络的原始输入变量进行预处理,选择输入变量的主成分作为网络输入,一方面减少了输入维度,消除了各输入变量的相关性;另一方面提高了神经网络的收敛性和稳定性,也简化了神经网络的结构。
由于很多用户缺乏必要的神经网络、甚至计算机编程的知识,因此我们使用Matlab语言的神经网络工具箱和VC++6.0,设计了一个主成分-神经网络模型系统。该模型系统可以给用户提供强大的计算能力,并能让用户按照他们的需求方便地定制自己的神经网络预测模型。同时,文章对神经网络的核心算法BP算法进行了改进,提出了三种改进算法,分别是动量和学习速率自调节的梯度下降算法、自适应学习率梯度下降算法和带动量的梯度下降算法,并给出三种算法的具体实现,通过数据证实三种优化算法在预测性能上表现更佳。
基于设计的模型,我们提供了农产品期货的两个典型实例作为具体设计应用的参考,同时对农产品的其他主要产品分别进行主成分神经网络预测,进一步说明主成分神经网络的可扩展性。通过实例证实基于主成分的神经网络应用于农产品期货市场预测的可行性和精确性,也告诉用户使用本模型的基本思路,为期货投资者提高资金收益提供了有利的理论依据。将人工神经网络中的BP算法应用于期货预测是对于BP算法应用领域的扩展,使其在金融领域有更好的使用价值。
Key to the success of agricultural futures, the market is the correct analysis, in particular on the market price of the correct forecast. Futures price is the focus of future trace, and it is advanced reflect on the future relationship between supply and demand, the profit and loss situation in the futures market is on the price of the judge to decide by traders. As a solution to China's "Gengzhe-benefit" social problems the major carriers, agricultural futures markets provide a more accurate long-term price guide information to agricultural production operators, which can reduce the production of blindness. Therefore, analysis and prediction of agricultural futures prices is of great significance.
Futures market is a complex nonlinear dynamic system. At present, domestic and foreign research methods for futures price forecasts are much more based component analysis (PCA), the moving average, exponential smoothing, and other traditional economic forecasting methods, and error back-propagation neural network (BP), Fuzzy neural networks, probability Neural networks, and other artificial neural network method. In practical using, the traditional single prediction method is simple, but the accuracy of prediction is not too high. The use of traditional time-series forecasting methods can not reveal its internal law.
Based on the analysis of the traditional inspection methods of predicting and analyzing, proposed on the basis of an analysis of the futures market for agricultural products, forecast by the neural network and principal component analysis of the composition of the new forecasting method, and new methods for the performance improvements and enhancements to the in-depth study. This article will introduce the principal component analysis method to futures market forecast, the original input variables pretreatment, select input variables as the principal component network input, on the one hand to reduce the importation of dimensions, eliminating all of the relevant input variables; on the other hand increase the neural network convergence and stability, but also to simplify the structure of the neural network.
Since many users lack the necessary neural network, or even computer programming knowledge, we use Matlab language neural network toolbox, a principal component design -neural network model system. The model system can provide a powerful computing capability to the user, and allows users customize their neural network model according to their needs easily. At the same time, the core of the neural network algorithm BP algorithm is proposed to improve the three improved algorithm, namely the momentum and learning rate adjustment since the gradient descent algorithm, adaptive learning rate gradient descent algorithm and lead in the gradient descent algorithm, and Three algorithms are given a concrete realization of the data confirmed that three optimization algorithms in predicting performance better performance.
Based on the design of the model, we provide the agricultural futures of the two typical examples of specific design applications as a reference, while the other major agricultural products are a principal component neural network, and further that the main components of the neural network scalability. Confirmed through examples based on principal component of neural network used in agricultural futures markets predict the feasibility and accuracy, also told users of the model's basic ideas and raising capital gains of investors in futures provide a favorable basis for the theory. Artificial neural network algorithm is applied to the BP futures forecast is for BP to expand the field of application of algorithms, in the financial sector to have a better value.
期货市场是一个复杂的非线性动态系统。当前,国内外期货价格预测的研究方法多为主成分分析法(PCA)、移动平均法、指数平滑法等传统的经济预测方法,以及误差反向传播神经网络(BP)、模糊神经网络、概率神经网络等人工神经网络方法。实际运用中,传统单一预测方法操作简便,但预测的准确率并不太高。利用传统的时间序列预测方法很难揭示其内在的规律。
本文在分析考察传统预测分析方法的基础上,提出一个面向农产品期货市场分析、预测,由神经网络和主成分分析法组合的新的预测方法,并针对该新方法性能的改善和提高进行了深入研究。将主成分分析法引入期货市场预测,对神经网络的原始输入变量进行预处理,选择输入变量的主成分作为网络输入,一方面减少了输入维度,消除了各输入变量的相关性;另一方面提高了神经网络的收敛性和稳定性,也简化了神经网络的结构。
由于很多用户缺乏必要的神经网络、甚至计算机编程的知识,因此我们使用Matlab语言的神经网络工具箱和VC++6.0,设计了一个主成分-神经网络模型系统。该模型系统可以给用户提供强大的计算能力,并能让用户按照他们的需求方便地定制自己的神经网络预测模型。同时,文章对神经网络的核心算法BP算法进行了改进,提出了三种改进算法,分别是动量和学习速率自调节的梯度下降算法、自适应学习率梯度下降算法和带动量的梯度下降算法,并给出三种算法的具体实现,通过数据证实三种优化算法在预测性能上表现更佳。
基于设计的模型,我们提供了农产品期货的两个典型实例作为具体设计应用的参考,同时对农产品的其他主要产品分别进行主成分神经网络预测,进一步说明主成分神经网络的可扩展性。通过实例证实基于主成分的神经网络应用于农产品期货市场预测的可行性和精确性,也告诉用户使用本模型的基本思路,为期货投资者提高资金收益提供了有利的理论依据。将人工神经网络中的BP算法应用于期货预测是对于BP算法应用领域的扩展,使其在金融领域有更好的使用价值。
Key to the success of agricultural futures, the market is the correct analysis, in particular on the market price of the correct forecast. Futures price is the focus of future trace, and it is advanced reflect on the future relationship between supply and demand, the profit and loss situation in the futures market is on the price of the judge to decide by traders. As a solution to China's "Gengzhe-benefit" social problems the major carriers, agricultural futures markets provide a more accurate long-term price guide information to agricultural production operators, which can reduce the production of blindness. Therefore, analysis and prediction of agricultural futures prices is of great significance.
Futures market is a complex nonlinear dynamic system. At present, domestic and foreign research methods for futures price forecasts are much more based component analysis (PCA), the moving average, exponential smoothing, and other traditional economic forecasting methods, and error back-propagation neural network (BP), Fuzzy neural networks, probability Neural networks, and other artificial neural network method. In practical using, the traditional single prediction method is simple, but the accuracy of prediction is not too high. The use of traditional time-series forecasting methods can not reveal its internal law.
Based on the analysis of the traditional inspection methods of predicting and analyzing, proposed on the basis of an analysis of the futures market for agricultural products, forecast by the neural network and principal component analysis of the composition of the new forecasting method, and new methods for the performance improvements and enhancements to the in-depth study. This article will introduce the principal component analysis method to futures market forecast, the original input variables pretreatment, select input variables as the principal component network input, on the one hand to reduce the importation of dimensions, eliminating all of the relevant input variables; on the other hand increase the neural network convergence and stability, but also to simplify the structure of the neural network.
Since many users lack the necessary neural network, or even computer programming knowledge, we use Matlab language neural network toolbox, a principal component design -neural network model system. The model system can provide a powerful computing capability to the user, and allows users customize their neural network model according to their needs easily. At the same time, the core of the neural network algorithm BP algorithm is proposed to improve the three improved algorithm, namely the momentum and learning rate adjustment since the gradient descent algorithm, adaptive learning rate gradient descent algorithm and lead in the gradient descent algorithm, and Three algorithms are given a concrete realization of the data confirmed that three optimization algorithms in predicting performance better performance.
Based on the design of the model, we provide the agricultural futures of the two typical examples of specific design applications as a reference, while the other major agricultural products are a principal component neural network, and further that the main components of the neural network scalability. Confirmed through examples based on principal component of neural network used in agricultural futures markets predict the feasibility and accuracy, also told users of the model's basic ideas and raising capital gains of investors in futures provide a favorable basis for the theory. Artificial neural network algorithm is applied to the BP futures forecast is for BP to expand the field of application of algorithms, in the financial sector to have a better value.
引文
[1]甘正平,农产品期货市场经济功能分析[D].博士学位,中国社会科学院研究生院,2003.
[2]什建玲、贺仲雄。期货价格预测的新方法--反馈灰色Markov模型[A]。1996中国控制与决策学术年会论文集;1996.
[3]张棉,严质彬.属性聚类网络和属性识别准则在期货价格预测中的应用.哈尔并商业大学学报(自然科学版).2003.6.
[4]陈志航程乾生.属性识别方法及其在期货价格预测中的应用.系统工程理论与实践.1999,19(6).
[5]胡斌,朱莉等。遗传算法的期货交易决策辅助系统应用方法[J].电脑开发与应用.2005,18(2):44-45.
[6]张方杰,胡燕京。ARMA模型在期货价格预测中的应用[J]。陕西统计与社会。2005,3:39-41。
[7]王习涛.ARIMA模型在期货交易预测中的应用研究[J]。微计算机信息.2006,15:139-140。
[8]刘轶芳,迟国泰。基于GARCH-EWMA的期货价格预测模型[J].哈尔滨工业大学学报。第9期:1572-1575.
[9]童宛生、胡俞越、冯中越.中国商品期货价格形成理论与实证分析[M].第一版。北京:中国财政经济出版社.1997.12.
[10]贺涛。上海粮食期货市场价格发现功能分析。华东经济管理[J].1999,3(5).
[11]华仁海,仲伟俊.中国期货市场期货价格波动与成交量和空盘量动态关系的实证分析[J].数量经济技术经济研究,2004(7).
[12]曹庆.基于神经网络和模糊理论的股市分析、预测与决策.[D].硕士学位,北京工业大学,2004.
[13]王科明.基于网格神经计算平台的预测模型的研究设计与实现[D].硕士,浙江大学,2004.
[14]王引斌.测定核心期刊的新方法--主成分分析法[J].情报学报.1998,17(5).
[15]陈晓红,朱霞。基于神经网络的期货套期保值决策支持系统[J].管理科学学报。2001,4(8):18-23.
[16]Martin T.Hagan,Howard B.Demuth,Mark H.Beale.神经网络设计[M].机械工业出版社,2002。9.
[17]伍华健.MATLAB/VC++组合编程方法及其应用.软件时空[J].2007.
[18]祖彦柱.时间序列ARCH模型在期货市场中的应用研究[D].硕士,河北工业大学,2005。
[19]杨行峻,郑君里.人工神经网络.高等教育出版社.1992
[20]袁曾任.人工神经网络及其应用.清华大学出版社.1992.
[21]闻新,周露等.Matlab神经网络应用技术.科学出版社.2000.
[22]黄文梅,杨勇等.系统仿真分析与设计。国防科技大学出版社.2001.
[23]韩力群.人工神经网络理论、设计及应用[M].化学工业出版社,2002.1.
[24]王科明,杨建刚,周炜彤.利用人工神经网络技术预测期货行情[J]。计算机工程与设计,2004,25(7):1164.
[25]李焕荣、王树明.一种改进的BP神经网络预测方法及其应用[J].系统工程,2000.9,18(5):86.
[25]禹建丽,孙增圻,Valgi.Kroumov,成久洋之,刘治军.基于BP神经网络的股市建模与决策[J]。系统工程理论与实践,2003.5,5;15.
[27]Karel Janecek.Futures Trading Model with Transaction Costs[D].Doctor,Carnegie Mellon University,2004.
[28]馀秉淨,张百灵,韦岗.神经网络理论及应用[D].华南理工大学出版社1994.
[29]王引斌.测定核心期刊的新方法--主成分分析法[J].情报学报1998,17(5)。
[30]徐自祥,周德云,罗奕然.基于主成分的模糊神经网络[J].计算机工程与应用2006.5.
[31]刘家学,陈世国.基于主成分分析的投资决策[J].运筹与管理2006,15(2)。
[32]魏广芬,唐祯安、余隽.基于主成分和BP神经网络的气体识别研究[J]2001.12 4.
[33]姬春煦,张骏.基于主成分分析的股票指数预测研究[J]。2006,28(8).
[34]王文剑。BP神经网络模型的优化.计算机工程与设计。2000,21(6):8-10.
[35]王晖,何新贵。BP网络泛化能力改进研究.系统工程与电子技术.2001,23(3):85-88。
[36]王继成.一种新的神经网络学习算法.计算机工程与科学.2000,22(3):6-11。
[37]江虹,曾立波,胡继明.优化的BP神经网络分类器的设计与实现.计算机工程与应用.2001(5).
[38]何晓群.回归分析与经济数据建模.中国人民大学出版社。1997.
[39]范剑波,杨建刚.宏观经济智能化预测性的研究,2001.
[40]丁雪梅。基于BP算法的证券预测技术研究[D].硕士,哈尔滨工程大学,2002.
[41]刘豹、胡代平.神经网络在预测中的一些应用研究[J].系统工程学报,1999.
[42]黄小原.神经网络预测系统[J].预测,1995.
[43]陈小前、罗世斌、王振国.BP神经网络应用中的前后处理过程[J]。系统工程理论与实践,2002.
[44]吕瑜、马良华.中国农产品期货价格与期货市场功能研究[D].硕士学位,浙江大学,2003.
[45]刘志新,贾福清.证券市场有效性理论与实证,航空C业出版社,2001.
[46][英」特伦斯C.米尔斯.金融时间序列的经济计量学模型(2).北京:经济科学出版社,2002.
[47]王耀东,张德远,张海雄.经济时间序列分析.上海:上海财经大学出版社,1996.
[48]张永东,毕秋香.上海股市波动性预测模型的实证比较.管理工程学报,2003(2),16-9.
[49]张定胜.计量经济学武汉:武汉大学出版社,2002.
[50]戴国强,吴林祥.金融市场微观结构理论.上海:上海财经大学出版社,1999.
[51]刘波.中国证券市场实证分析.北京:学林出版社,1997.
[52]中国期货业协会.期货市场教程(2).北京:中国财政经济出版社,2003.
[53]倪加勋,袁卫.应用统计学.北京:中国人民大学出版社,1992.
[54][美]纳坦恩伯格.期权价格波动率与定价理论.北京:经济科学出版社,2000.
[55]姬广坡.期货市场风险控制论.北京:中国财政经济出版社,2000.
[56」田源.从国家战略意义上认识期货市场.投资与证券,2002(6),7374.
[57][美]詹姆斯D.汉密尔顿.时间序列分析.北京;中国社会科学出版社,1999.
[58]常清.中国期货市场发展的战略研究.北京:经济科学出版社,2001.
[1]BessembinderH,SeguinPJ.Futurestradingactivityandstockpricevolatility[J].Journalo fFinance,1992,47(5):2015-2034.
[2]BessembinderH,SeguinPJ.Pricevolatility,tradingvlumemandmarketdepth:evidencefromfutures market[J].JournalofFinancialandQuantitativeAnalysis,1993,28(1):21-39.
[3]FosterAJ.Volume-volatilityrelationsforcrudeoilfuturesmarkets[J].JournalofFuturesMarkets,1995,15:929-951.
[4]FujiharaRA,MougoueM.Lineardependence,nonlineardependenceandpetroleumfuturesmarkete fficiency[J].JournalofFuturesMarkets,1997,17:75-99.
[5]RonaldGA,RossiPE,TauchenG.Stockpricesandvolume[J].TheReviewofFinancialStudies,1992,5(2):199-242.
[6]WangGHJ,YauJ.Tradingvolume,bid-askspread,andpricevolatilityinfuturesmarkets[J].Journalof FuturesMarkets,2000,20(10):943-970.
[7]Karel Janecek.Futures Trading Model with Transaction Costs[J].Mathematical Sciences,2004.
[8]Dequan He.Modeling Transactions Costs Band and Nonlinear Price Dynamics in ForestCommoditv Markets[J].Economics,2005.
[2]什建玲、贺仲雄。期货价格预测的新方法--反馈灰色Markov模型[A]。1996中国控制与决策学术年会论文集;1996.
[3]张棉,严质彬.属性聚类网络和属性识别准则在期货价格预测中的应用.哈尔并商业大学学报(自然科学版).2003.6.
[4]陈志航程乾生.属性识别方法及其在期货价格预测中的应用.系统工程理论与实践.1999,19(6).
[5]胡斌,朱莉等。遗传算法的期货交易决策辅助系统应用方法[J].电脑开发与应用.2005,18(2):44-45.
[6]张方杰,胡燕京。ARMA模型在期货价格预测中的应用[J]。陕西统计与社会。2005,3:39-41。
[7]王习涛.ARIMA模型在期货交易预测中的应用研究[J]。微计算机信息.2006,15:139-140。
[8]刘轶芳,迟国泰。基于GARCH-EWMA的期货价格预测模型[J].哈尔滨工业大学学报。第9期:1572-1575.
[9]童宛生、胡俞越、冯中越.中国商品期货价格形成理论与实证分析[M].第一版。北京:中国财政经济出版社.1997.12.
[10]贺涛。上海粮食期货市场价格发现功能分析。华东经济管理[J].1999,3(5).
[11]华仁海,仲伟俊.中国期货市场期货价格波动与成交量和空盘量动态关系的实证分析[J].数量经济技术经济研究,2004(7).
[12]曹庆.基于神经网络和模糊理论的股市分析、预测与决策.[D].硕士学位,北京工业大学,2004.
[13]王科明.基于网格神经计算平台的预测模型的研究设计与实现[D].硕士,浙江大学,2004.
[14]王引斌.测定核心期刊的新方法--主成分分析法[J].情报学报.1998,17(5).
[15]陈晓红,朱霞。基于神经网络的期货套期保值决策支持系统[J].管理科学学报。2001,4(8):18-23.
[16]Martin T.Hagan,Howard B.Demuth,Mark H.Beale.神经网络设计[M].机械工业出版社,2002。9.
[17]伍华健.MATLAB/VC++组合编程方法及其应用.软件时空[J].2007.
[18]祖彦柱.时间序列ARCH模型在期货市场中的应用研究[D].硕士,河北工业大学,2005。
[19]杨行峻,郑君里.人工神经网络.高等教育出版社.1992
[20]袁曾任.人工神经网络及其应用.清华大学出版社.1992.
[21]闻新,周露等.Matlab神经网络应用技术.科学出版社.2000.
[22]黄文梅,杨勇等.系统仿真分析与设计。国防科技大学出版社.2001.
[23]韩力群.人工神经网络理论、设计及应用[M].化学工业出版社,2002.1.
[24]王科明,杨建刚,周炜彤.利用人工神经网络技术预测期货行情[J]。计算机工程与设计,2004,25(7):1164.
[25]李焕荣、王树明.一种改进的BP神经网络预测方法及其应用[J].系统工程,2000.9,18(5):86.
[25]禹建丽,孙增圻,Valgi.Kroumov,成久洋之,刘治军.基于BP神经网络的股市建模与决策[J]。系统工程理论与实践,2003.5,5;15.
[27]Karel Janecek.Futures Trading Model with Transaction Costs[D].Doctor,Carnegie Mellon University,2004.
[28]馀秉淨,张百灵,韦岗.神经网络理论及应用[D].华南理工大学出版社1994.
[29]王引斌.测定核心期刊的新方法--主成分分析法[J].情报学报1998,17(5)。
[30]徐自祥,周德云,罗奕然.基于主成分的模糊神经网络[J].计算机工程与应用2006.5.
[31]刘家学,陈世国.基于主成分分析的投资决策[J].运筹与管理2006,15(2)。
[32]魏广芬,唐祯安、余隽.基于主成分和BP神经网络的气体识别研究[J]2001.12 4.
[33]姬春煦,张骏.基于主成分分析的股票指数预测研究[J]。2006,28(8).
[34]王文剑。BP神经网络模型的优化.计算机工程与设计。2000,21(6):8-10.
[35]王晖,何新贵。BP网络泛化能力改进研究.系统工程与电子技术.2001,23(3):85-88。
[36]王继成.一种新的神经网络学习算法.计算机工程与科学.2000,22(3):6-11。
[37]江虹,曾立波,胡继明.优化的BP神经网络分类器的设计与实现.计算机工程与应用.2001(5).
[38]何晓群.回归分析与经济数据建模.中国人民大学出版社。1997.
[39]范剑波,杨建刚.宏观经济智能化预测性的研究,2001.
[40]丁雪梅。基于BP算法的证券预测技术研究[D].硕士,哈尔滨工程大学,2002.
[41]刘豹、胡代平.神经网络在预测中的一些应用研究[J].系统工程学报,1999.
[42]黄小原.神经网络预测系统[J].预测,1995.
[43]陈小前、罗世斌、王振国.BP神经网络应用中的前后处理过程[J]。系统工程理论与实践,2002.
[44]吕瑜、马良华.中国农产品期货价格与期货市场功能研究[D].硕士学位,浙江大学,2003.
[45]刘志新,贾福清.证券市场有效性理论与实证,航空C业出版社,2001.
[46][英」特伦斯C.米尔斯.金融时间序列的经济计量学模型(2).北京:经济科学出版社,2002.
[47]王耀东,张德远,张海雄.经济时间序列分析.上海:上海财经大学出版社,1996.
[48]张永东,毕秋香.上海股市波动性预测模型的实证比较.管理工程学报,2003(2),16-9.
[49]张定胜.计量经济学武汉:武汉大学出版社,2002.
[50]戴国强,吴林祥.金融市场微观结构理论.上海:上海财经大学出版社,1999.
[51]刘波.中国证券市场实证分析.北京:学林出版社,1997.
[52]中国期货业协会.期货市场教程(2).北京:中国财政经济出版社,2003.
[53]倪加勋,袁卫.应用统计学.北京:中国人民大学出版社,1992.
[54][美]纳坦恩伯格.期权价格波动率与定价理论.北京:经济科学出版社,2000.
[55]姬广坡.期货市场风险控制论.北京:中国财政经济出版社,2000.
[56」田源.从国家战略意义上认识期货市场.投资与证券,2002(6),7374.
[57][美]詹姆斯D.汉密尔顿.时间序列分析.北京;中国社会科学出版社,1999.
[58]常清.中国期货市场发展的战略研究.北京:经济科学出版社,2001.
[1]BessembinderH,SeguinPJ.Futurestradingactivityandstockpricevolatility[J].Journalo fFinance,1992,47(5):2015-2034.
[2]BessembinderH,SeguinPJ.Pricevolatility,tradingvlumemandmarketdepth:evidencefromfutures market[J].JournalofFinancialandQuantitativeAnalysis,1993,28(1):21-39.
[3]FosterAJ.Volume-volatilityrelationsforcrudeoilfuturesmarkets[J].JournalofFuturesMarkets,1995,15:929-951.
[4]FujiharaRA,MougoueM.Lineardependence,nonlineardependenceandpetroleumfuturesmarkete fficiency[J].JournalofFuturesMarkets,1997,17:75-99.
[5]RonaldGA,RossiPE,TauchenG.Stockpricesandvolume[J].TheReviewofFinancialStudies,1992,5(2):199-242.
[6]WangGHJ,YauJ.Tradingvolume,bid-askspread,andpricevolatilityinfuturesmarkets[J].Journalof FuturesMarkets,2000,20(10):943-970.
[7]Karel Janecek.Futures Trading Model with Transaction Costs[J].Mathematical Sciences,2004.
[8]Dequan He.Modeling Transactions Costs Band and Nonlinear Price Dynamics in ForestCommoditv Markets[J].Economics,2005.