导　　师： 杨建辉

学科专业： 1201

授予学位： 硕士

作　　者： ;

机构地区： 华南理工大学

摘　　要： 针对金融时间序列的复杂性，本文将经验模态分解（EMD）引入金融时间序列预测框架中进行研究。EMD多应用于通信、气象领域的数据处理，而应用于金融领域则不多，但是它具有明显的优点：能根据数据本身的时间尺度特征准确反映原时间序列的物理特性而不造成信号损失，而且无需预先设定任何基函数，这与小波分析、傅里叶变换等方法有本质的区别。但EMD存在模态混叠问题，因此需要对EMD方法进行改进和优化以提高预测的有效性。 本文首先构建了总体经验模态分解（EEMD）模型，基于获取的原始数据，模拟产生多条路径的修正数据，每一次修正的数据中加入不同的白噪声以抵消原始数据中的噪声，对每个修正的序列进行EMD分解，而后取多次分解的平均值作为最后的分解序列，从而升了序列的信噪比，解决模态混叠问题。之后，将支持向量回归（SVR）模型引入到金融时间序列分析，同时，采用多种群遗传算法（MPGA）进行SVR的参数寻优。MPGA引入多个种群同时进行遗传进化搜索，对不同的种群设置相应的控制参数，并在不同种群之间依靠移民算子完成信息交互，最终在多个种群协同进化下得到最优解，可以有效地防止早熟，大大提高收敛速度。 最后，基于前文构建的EMD与SVR的改进模型，在趋势交易中进行应用。构建EEMD-MPGA-SVR预测模型。应用的结果表明：其一，对比不同参数寻优的SVR模型发现，不论是否进行EEMD分解，与网格搜索法SVR、标准遗传算法SVR相比，多种群遗传算法SVR的参数估计及其预测效果都是最好的。其二，对比进行EEMD分解前后的预测效果发现，基于EEMD分解的SVR预测效果明显优于直接采用原始序列的SVR预测（偏差较小），而且能较快地捕捉市场信息，由此所触发交易的累计收益率也更高，平均累计收益率也更高，从而收益的稳定性更强。 Consider of the complexity of financial time series, we introduce the empiricalmode decomposition /(EMD/) model into the framework of financial time seriesprediction. EMD is mostly utilized on data processing in fields like communicationand weather, while the application in finance is rare, but it owns distinct advantages:according to the time-scale features, the original data can be decomposed into a seriesand accurately reflect the physical characteristics of the original series without signallosses, what’s more, there is no need of any pre-set basis functions, which differsEMD from other methods such as wavelet decomposition and Fourier transformationessentially. But, mode mixing may appear during EMD decomposition, so we shouldoptimize the this model to improve the effectiveness of prediction. Firstly, we introduce the ensemble EMD /(EEMD/) model. Based of the originaldata, it firstly generates a set of multi-path revised data, and that on each path containsa distinct white noise series from any others in order to cancel the noise in the originaldata, then perform EMD to each revised series and then take the average ofdecompositions as the final decomposed series. It has enhanced the signal-to-noiseratio of the decomposed series, thus solving the problem of mode mixing.Secondly,weintroduce the support vector regression /(SVR/) model into the financial time seriesanalysis, and optimize the parameters using multi-population GA model/(MPGA/). TheMPGA model introduces multiple populations when performing GA searching, at thesame time, it makes the parameters of GA different between populations.What’smore,the MPGA model depends information exchange between populations onimmigration operator, and finally gets the optimal solution after co-evolution ofmultiple populations, avoiding prematurity and improving the convergence rate. Finally, based on the modified EMD and SVR models, we build the predictingmodel called EEMD-MPGA-SVR, and apply into the trend trading. The result shows:First, after comparing different parameter optimization methods of SVR, we find thatwhether or not executing EEMD decomposition, the parameters estimation and predictions of MPGA-SVR are best while comparing with the other two of GS-SVRand GA-SVR. Second, after comparing predictions by original data with those byEEMD decomposition, we find that the latter are significantly better /(smallerdeviation/), and can capture market information more quickly, thereby the cumulativereturn of triggered transaction is higher, and the same to the average cumulative return,meaning more stability is the return.

关 键 词： 经验模态分解 支持向量回归机 遗传算法 股指期货 趋势交易

分 类 号： [F830.91 F224 TP18]

领　　域： [经济管理] [经济管理] [自动化与计算机技术] [自动化与计算机技术]