导　　师： 张泽银

学科专业： 070101

授予学位： 硕士

作　　者： ;

机构地区： 浙江大学

摘　　要： 传统频域分析方法在处理非平稳时间序列时往往受到Heisenberg测不准原理的制约,无法在时频两域同时达到很高的精度;传统的时域分析方法所基于的平稳性假设、正态分布的假设、线性性质等假设在实际的金融时间序列中往往是不成立的,因此在对一些非平稳时间序列建模时可能会产生严重的失真。 HHT方法是1998年由美籍华人N.E.Huang提出的一种数据分析方法,该方法由两个步骤组成:EMD/(empirical mode decomposition,经验模式分解/)和HSA/(Hilbert spectrum analysis,Hilbert谱分析/)。由于该方法具有完全的自适应性,能处理非线性非平稳数据,能够不受Heisenberg测不准原理制约,并且对经过EMD得到的IMF进行Hilbert谱变换能产生具有物理意义的瞬时频率,这是传统谱分析法很难做到的,因此能够弥补传统方法在处理具有极强非线性非平稳特性的金融时闻序列分析上的不足。 复杂系统如证券市场、气象系统等,系统变量之间的相互关系十分复杂,甚至可能找不出比较固定的规律性,这类系统的具有一个最大的特征即:时间序列具有极强的非线性、非平稳性;关于系统的重要变量具有大量的时序模式。因此,传统的定性预测方法,包括Delphi法和目标分解预测法,传统的时间序列模型,包括移动平均模型和指数平滑法,传统的因果模型,包括分解预测法、ARIMA、回归分析等模型均不十分适用于此类系统的建模。而非参数自组织数据挖掘算法——类比合成算法/(AC算法/)在对输出变量进行预测时不需要预先对输入变量的发展趋势进行估计或假设,其预测结果完全依据已知的数据给出,因此非常适合于此类系统的建模。 The traditional frequency analysis method often subject to Heisenberg uncertainty principle when processing the non-stationary time series, so it is unable to achieve high precision in both time and frequency simultaneously; Traditional time analysis method always based on the stable, normal distribution's, linear hypothesis and so on, but in fact it is often untenable, therefore it is likely to have the serious distortion when cheating some non-stationary time series. HHT/(Hilbert-Huang transform/)is a new data analysis method proposed by N. E. Huang in 1998, it consists of two steps: EMD/(empirical mode decomposition/) and HAS/(Hilbert spectrum analysis/). Because this method is totally adaptive; it can process non-stationary data, and not subject to Heisenberg uncertainty principle, and the IMFs which introduced by EMD has the physics significance instantaneous frequency, it can therefore overcome the conventional route's insufficiency which in processing the financial time series which has greatly strengthened non-stationary characteristic. The complicated system like stock market, the meteorological system and so on, the correlation between system variables is often very complex, it can hardly find out the regularity of them, the ultimate characteristic is: the time series have the greatly strengthened misalignment, the non-stationary. Therefore, the traditional qualitative forecast technique, including the Delphi law and the goal decomposition forecast law, the traditional time sequence model, including the running mean model and the index smoothing procedures, the traditional causes and effects model, including decomposition models, forecast law, ARIMA, regression analysis isn't suitable for this kind of system's modeling. But non-parameter, self organization data mining algorithm-- analogy compose algorithm /(AC algorithm/) which carries on the forecast to the output variable does not need to carry on the estimate or the supposition in advance to the input variable trend of development, its forecasting result rests on the completely foregone data, therefore it is very fit for this kind of system's modeling.

关 键 词： 金融时间序列 算法 波动周期 预测

分 类 号： [F224 F830]

领　　域： [经济管理] [经济管理]