基于Adaboost和正则化ELM的混合金融时间序列预测模型及其应用  

A Hybrid Model Based on Adaboost and Regularized ELM for the Forecasting of Financial Time Series and Its Applications

作　　者：陈艳[1] 石智慧[1] 

机构地区：[1]上海财经大学统计与管理学院,上海200433

出　　处：《数理统计与管理》2017年第1期113-125,共13页Journal of Applied Statistics and Management

基　　金：国家自然科学基金资助项目(71271128;71331006;71571113);长江学者和创新团队发展计划(上海财经大学:IRT13077);上海财经大学创新团队支持计划

摘　　要：为提高金融时间序列的预测精度,本文提出了基于MODWT、MCP变量选择方法和RELM_Adaboost的混合预测模型。该模型由三步构成:第一步,收集特征变量,包括MODWT分解得到的特征变量以及常用的技术指标;第二步,利用MCP惩罚方法从上述特征变量中选取重要的作为输入变量;第三步,利用Mnet惩罚正则化ELM,将RELM视作弱预测器,然后用Adaboost算法生成强预测器进行预测。实证结果显示:第一,经过MCP方法的筛选,最终的输入变量中不仅包含常用技术指标,还有小波分解所得的变量。第二,混合预测模型RELM_Adaboost有良好的泛化误差表现。本文提出的模型在量化交易时代具有良好的应用前景。In order to improve the prediction accuracy of financial time series, this paper proposes a hybrid model combining MODWT, MCP variable selection method and RELM_Adaboost. The hybrid model has three stages. In the first stage, collect variables which include the popular technical indices and features extracted through wavelet decomposition. In the second stage, select the significant vari- ables with MCP method as final inputs. Finally, regularized ELMs with Mnet penalty （RELM） are used as weak predictors, and they are composed of strong predictor model with the RELM_Adaboost algorithm. Experimental results show that： Firstly, the important inputs selected by MCP method in- clude both popular technical indices and features obtained by the wavelet decomposition. Secondly, the RELM_Adaboost model possesses higher precision accuracy, which means it is highly promising. The proposed model shows its practical value in the field of quantitative trading.

关 键 词：小波分析 变量选择 正则化ELM Adaboost强预测器 

分 类 号：F222.1[经济管理—国民经济] F832.5