马腾 汪晶 丁绍纹 潘佳铭 朱家明



【摘 要】在“互联网+大数据”的背景下,搜索引擎为人类提供了多源的瞬时信息。在预测中,由于预测系统的复杂性,区间数作为刻画事物随机阶段性信息的一种表现形式,蕴含信息较时点序列更加丰富。而传统的区间组合预测模型并不能很好地处理非线性时間序列,因此,论文研究多尺度视角下区间组合预测模型及其在金融时间序列中的应用。首先利用改进的BEMD算法对区间金融时间序列进行多尺度分解,其次利用三种区间型单项预测方法对分解后的序列进行单项预测,最后组合单项预测的结果得到最优组合预测结果,通过对上证指数的实证,验证了论文所提多尺度区间组合预测模型的有效性。
【Abstract】Under the background of "internet + big data", search engines provide human with instantaneous information of multiple sources. In forecasting, because of the complexity of the forecasting system, interval number, as a form of expression to describe the random periodic information of things, contains more rich information than the time point series. However, the traditional interval combination forecasting model can not deal with the nonlinear time series well. Therefore, this paper studies the interval combination forecasting model from the multi-scale perspective and its application in financial time series. The paper firstly performs a multi-scale decomposition of interval financial time series using the improved BEMD algorithm, and then uses three interval single forecasting methods to perform single forecasting on the decomposed series, and finally combines the results of the single forecasting to obtain the optimal combination forecasting results. The validity of the multi-scale interval combination forecasting model proposed in the paper is verified through the empirical evidence of the Shanghai Composite Index.
【关键词】多尺度分解;组合预测;区间预测;金融时间序列
【Keywords】multiscale decomposition; combination forecasting; interval forecasting; financial time series
【中图分类号】F224;F832 【文献标志码】A 【文章编号】1673-1069(2021)09-0059-04
1 引言
金融市场是中国市场经济体制中的一个极其重要的组成部分,而股票指数则是金融市场的核心。近年来,金融市场价格波动频繁,不仅为投资者们带来经营风险,也对金融管理部门决策造成一定不良影响。准确地预测股票指数不仅有助于建立稳定有效的金融市场定价机制,为政府制定合理金融市场交易政策提供帮助,同时,有利于金融市场的稳定和健康发展。
近年来,随着计算机网络技术的广泛和深入发展,金融数据出现和使用的频率越来越密集,其不确定性在不断地增加,时间序列数据在量上要更庞大,从特征上看要更为复杂,因此,利用传统的预测方法对非线性的时间序列进行预测,效果较差。……