农产品期货市场波动率的预测以及预测精度评价研究

Agricultural futures play an important role in price discovery and risk coverage, and scientifically and accurately predicting its volatility play a key role in full realization of its hedging and other multiple functions, but the research on the volatility prediction of agricultural futures market is very little at present..The project is to (1)estimate the the volatility of agricultural futures markets by kernel-based estimators and scale transform based on the high-frequency data, and test the possibility of structural breaks, long memory and asymmetry in the daily realized volatility series as the basis of modeling.(2) propose a new ARFIMA model to predict the volatility of agricultural futures market. The new model will corret the shortage of the current models which ignored the characteristic of structural breaks for volatility, and the new model is estimated by MCMC method which overcome the shortcomings that the previous studies rely solely on MLE and GMM.(3) propose a robust prediction accuracy evaluation method based on Bootstrap method to evaluate and compare the prediction accuracy of various volatility forecasting model. The new method will overcome the shortcomings that the previous studies about prediction accuracy evaluation is not rigorous enough..The realization of the project, will extend the research about volatility prediction for agricultural futures markets, and provide theory for agricultural producers, investors and consumers of agricultural products to obtain accurate forward price information, asset pricing and arbitrager to avoid the risk effectively, and financial supervision departments to improve the agricultural futures market risk supervision level.

农产品期货在价格发现和风险防范过程中扮演重要角色，科学准确地预测其波动率对充分实现其避险等多重功能是决定性的，而目前对农产品期货市场波动率的预测研究尚较少。. 本项目拟：(1)基于高频数据的非参数核估计和尺度变换估计农产品期货市场波动率，拟检验波动率的长记忆性、结构突变和不对称性，作为建模的基础；（2）构建新型的ARFIMA模型预测农产品期货市场波动率，改进现有模型忽略结构突变的不足，拟采用MCMC方法估计模型参数，克服以往单纯依赖MLE和GMM估计的缺陷；（3）实证评价模型的预测性能，拟基于Bootstrap构建稳健的预测精度评价方法，改进现有评价方法不严谨的缺陷。. 本项目的实现，将丰富农产品期货市场波动率预测的研究，为农业生产者、投资者和农产品消费者获得准确的远期价格信息，资产定价者和套利者有效地规避风险以及金融监管部门提高农产品期货市场的风险监管水平提供理论依据。

本课题通过构建MS-ARFIMA模型、TVS-HAR模型和自适应的不对称性HAR-CJ-D-FIGARCH模型探讨了已实现波动率的动态特征，并对波动率进行了预测。MS-ARFIMA模型采用MCMC的方法估计，克服了现有模型未考虑区制转换的缺陷；TVS-HAR模型将所有潜在预测因子纳入到时变HAR模型中，同时考虑了预测因子的时变性和预测模型的时变性，克服了现有HAR族模型面临预测因子预测能力的时变性、模型选择风险以及预测模型的时变性等方面的问题；自适应的不对称性HAR -CJ-D-FIGARCH 模型扩展了HAR-GARCH模型和HAR-CJ模型，同时考虑了已实现波动率的结构突变、长记忆性、不对称性、周内效应和已实现波动率的不同成分特征。此外，本课题还采用修正的MCS检验实证评价和比较了新构建的预测模型同其他现有预测模型的预测性能，同时深入探讨了结构突变条件下如何预测农产品期货已实现波动率的问题。. 分析结论表明：我国农产品期货的已实现波动率具有高波动和低波动两种状态，状态之间的转换概率较小，纯粹的分数阶积分过程或区制转换模型都不能全部解释已实现波动率的强长记忆性；已实现波动率同时具有长记忆性、结构突变、不对称性和周内效应等特征，结构突变只能部分解释其长记忆性，其连续性成分的长记忆性很强，而跳跃性成分的长记忆性非常弱，不同期限的跃性成分对未来波动率都有显著的正效应；针对不同的市场和不同的预测期，任意潜在预测因子对农产品期货市场波动率的预测的重要性都具有时变性；同现有波动率预测模型相比，新模型的样本外预测效果是最好的，且基于NKR的非参数组合和基于NRLS和SIC的线性组合预测模型是在结构突变条件下预测农产品期货市场波动率尤其有效的方法。. 本研究丰富了波动率预测的研究方法，进一步完善了预测理论，对于各类投资者优化资产组合的配置，测量、预测和防范资产价格变动的风险，以及提高金融监管部门对金融市场的风险监管水平都具有重大的现实意义。