【作者】 徐刚；

【导师】 张伟平；

【作者基本信息】 中国科学技术大学 ， 统计学， 2017， 硕士

【摘要】 检验高频金融数据跳跃点和研究它的波动性在应用中是必要的,例如衍生品定价和风险管理。虽然近些年学者们提出大量的检验跳跃的方法,但这些方法依赖跳跃点数量已知或多重假设检验。这导致了这些检验方法表现出不稳健性以及在实证研究中检验出虚假的跳跃点。另外降噪算法可以清洗数据并估计系统整体趋势,因此对有跳跃点的高频金融数据做降噪处理也是研究中很重要的一部分。基于局部线性尺度逼近(LLSA)和极大重叠离散小波变换(MODWT),本文提出了基于小波的跳跃信息准则(WJIC),它可以同时对数据降噪和识别跳跃点,并且我们构造得分函数优化选择参数。我们通过模拟实验对比WJIC和其它方法的表现,并且把我们的算法应用到美国全国证券交易商协会自动报价表(NASDAQ)。我们证明了 WJIC得到的估计量有良好的渐近性质,模拟及实证研究表明了 WJIC得到的估计量在数值计算中表现很好。更多还原

【Abstract】 Detecting jumps in high-frequency financial series data and studying its dynamics is imperative in applications such as derivatives pricing and risk management.Although numerous methods to detect the presence of jumps have been introduced recently,they rely on the assumption that the number of jumps is known or multiple test procedure.Consequently,they are non-robust in the presence of jumps and could lead to spuri-ous detections in empirical studies.In addition,denoising high-frequency financial se-ries data with jumps is also significant because the denoising algorithm can clean the data and get the estimator of systematic pattern.Based on local linear scaling approx-imation(in short,LLSA)algorithm and the linear maximal overlap discrete wavelet transform(MODWT),we propose a wavelet-based jump information criterion(WJIC)in this paper for denoising and jump identification simultaneously and we optimally de-termine the parameters by using a score function.We conduct a simulation to compare the performance of WJIC and other methods and apply our algorithm with National As-sociation of Securities Dealers Automated Quotations(NASDAQ)Index.Asymptotic properties of the proposed estimators are derived and simulations and empirical studies demonstrate that the proposed estimator has good properties in terms of mean squared error.更多还原