本文报告一种金融时间序列预测的信号分析、信息融合与智能计算组合模型，简称FEPA，由针对金融时间序列（FTS）信号分析的经验模态分解（EMD）、用于数据降维的主成分分析（PCA）和用于非线性建模的人工神经网络（ANN）三部分组成。该模型首先应用滑动窗口截取原始金融时间序列最近期数据集，应用EMD分解算法把数据集分解成不同尺度的本征模态函数（IMF），然后通过主成分分析将分解后的数据降维，提取最有信息量的特征；然后将这些特征输入到神经网络进行组合预测。本文提出的组合预测模型FEPA是基于分解-提优-合成的信息融合思想，有效提高了预测可靠性。其创新点在于：1）首次给出了EMD算法的结构化表达，提供了今后融合更多信息的算法接口；2）通过多步长预测输出深入研究EMD分解的有效信息结构；3）通过切换到更细时间框架来处理EMD的端点效应，并探索了两级时间框架下的预测效果；4）给出了金融时间序列组合预测模型的一般性架构，具有可升级性和可扩展性。并且通过滑动窗口EMD使得实证更能切近实际。通过在沪深300股指和澳大利亚股指上的实证，结果表明FEPA预测模型在沪深300股指日线和15分钟线上的预测命中率高达78%和82%，在澳大利亚股指日线上也达到了74%的命中率，经比较，明显高于文献中常见的5种模型。
In this paper, an adaptive model is documented for predicting financial time series integrating signal processing, information fusion and computational intelligence. The model consists of financial time series (FTS)-specific Empirical Mode Decomposition (EMD) for signal processing, Principal Component Analysis (PCA) for dimension reduction, and Artificial Neural Networks (ANN) for nonlinear prediction. The model uses a sliding window to capture the most recent time series data, applies EMD to transform the data into multilevel Intrinsic Mode Functions (IMF's). PCA is then used to reduce the dimension of IMF's and to generate a set of information-rich features which are input into an ANN to generate the output as prediction. This novel model of prediction implements an information fusion process consisting of signal decomposition, dimension reduction and nonlinear synthesis. This model lifts the prediction capability to a new level. The originality of this work exhibits in fouraspects:1) a structural reformulation of EMD algorithm, providing an interface to more information fusion, 2)deepening into finer time frames for tackling the end effect of EMD andimplementation and testing on two levels of time frameimplementation,3) investigation of multi-step prediction, 4) a generic framework of prediction models for financial time series with upgradability and extensibility. The use of sliding window for EMD also gets the test closer to the reality.The new model is tested on the historical data of two stock indices-Chinese HS300 and Australian AORD, the performance, achieving a hit rate of 78% and 82% on HS300 D1 and M15, and 74% on AXJO D1 respectively, significantly higher than 5 existing models after comparison.
[1] Pan Heping, Sornette D, Kortanek K. Intelligent finance-An emerging direction[J]. Quantitative Finance, 2006,6(4):273-277.
[2] Pan Heping. A basic theory of intelligent finance[J].New Mathematics and Natural Computation, 2011,7(2):197-227.
[3] Huang N E, Shen Zheng, Long S R, et al. The empirical mode decomposition and the Hilbert spectrum for nonlinear and nonstationary time series analysis[J].Proceedings:Mathematical Physical and Engineering Sciences, 1998,454(1971):903-995.
[4] Yu Lean, Wang Shouyang, Lai K K. Forecasting crude oil price with an EMD-based neural network ensemble learning paradigm[J]. Energy Economics,2008,30(5):2623-2635.
[5] 周德群,鞠可一,周鹏,等. 石油价格波动预警分级机制研究[J].系统工程理论与实践,2013,33(3):585-592.
[6] Lin C S, Chiu S H, Lin T Y. Empirical mode decomposition-based least squares support vector regression for foreign exchange rate forecasting[J]. Economic Modelling, 2012,29(6):2583-2590.
[7] 王书平,胡爱梅,吴振信. 基于多尺度组合模型的铜价预测研究[J].中国管理科学,2014,22(8):21-28.
[8] 阮连法,包洪洁.基于经验模态分解的房价周期波动实证分析[J].中国管理科学,2012,20(3):41-46.
[9] 张秀艳,徐立本. 基于神经网络集成系统的股市预测模型[J]. 系统工程理论与实践,2003, 9(9):67-70.
[10] 秦宇. 应用经验模态分解的上海股票市场价格趋势分解及周期性分析[J].中国管理科学,2008,16(S1):220-225.
[11] 王文波,费浦升,羿旭明. 基于EMD与神经网络的中国股票市场预测[J].系统工程理论与实践,2010,30(6):1027-1033.
[12] 张承钊,潘和平. 基于前向滚动EMD技术的预测模型[J].技术经济, 2015,34(5):70-76.
[13] Dragomiretskiy K, Zosso D.Variational mode decomposition[J]. IEEE Transactions on Signal Processing,2014,62, 531-544.
[14] Wei Liangying.A hybrid ANFIS model based on empirical mode decomposition for stock time series forecasting[J].Applied Soft Computing, 2016,42:368-376.
[15] Lahmiri S.A variational mode decomposition approach for analysis and forecasting of economic and financial time series[J].Expert Systems with Applications, 2016,55:268-273.
[16] Krogh A, Vedelsby J. Neural network ensembles, cross validation and active learning[C]//Proceedings of the 7th International Conference on Neural Information Processing Systems, Cambridge, MA, USA, 1994.
[17] Rilling G, Flandrin P, Goncalves P. On empirical mode decomposition and its applications[C]//Proceedings of IEEE-EURASIP Workshop on Nonlinear Signal and Image Processing. 2003,NSIP-03, Grado.
[18] 潘和平. 深度智能投资-智能投资组合理论、强势投资法与深度学习智能交易策略的统一[C]//第四届全国金融大数据战略与应用研讨会特邀报告,中科院大学主办, 2016年10月9-11日,北戴河.
