摘要
日内的金融高频数据轨迹呈现显著的连续函数特征,由此构成的日度函数序列间往往存在相依性。为了提升预测的准确率和稳健性,综合考虑金融高频数据的日内波动模式和日度相依结构,文章提出了以自适应分类为基础的混合期望预测方法(cDFPC)。首先,基于函数型自适应聚类分析识别日内金融高频数据潜在的波动模式类别,并构建波动模式类别归属的函数型判别准则;其次,重构长期协方差算子对相依函数序列的协方差进行纠偏,并给出相依函数型回归模型的参数估计;最后,对于给定的待预测对象,基于相依函数型回归模型预测其在每一个类别的取值,使用函数型判别模型计算其类别隶属的后验概率,并以此为权重汇总每一个类别的预测值。数值模拟结果表明,考虑波动模式的混合期望预测能够提升预测精度,cDFPC在长期和短期相依函数型数据的预测中均有优势。基于上证指数开盘价预测的实证发现,上证指数的日内波动模式和日度相依结构显著存在,cDFPC的相对预测优势保持稳健。
Intraday financial high-frequency data can be regarded as a continuous function,and there is often interdependency among the resulting function series.In order to improve the accuracy and robustness of forecasting,this paper proposes a classification-based mixed expectation forecasting method(cDFPC),which takes into account the intraday volatility patterns and daily interdependency structure of high-frequency financial data.Firstly,the potential volatility pattern categories of intraday financial high-frequency data are identified based on functional adaptive clustering analysis,and the functional discriminant criterion for volatility pattern category attribution is constructed.Secondly,the long-term covariance operator is reconstructed to correct the covariance of the dependent function series,and the parameter estimation of the interdependency function-type regression model is given.Finally,the values of the predicted objects in each category are predicted based on the interdependency function-type regression model;the posterior probabilities of their category affiliation are calculated by using the functional discriminant model,and the predicted values of each category are aggregated with this weight.The numerical simulation results show that the mixed expectation forecasts considering volatility patterns can improve the forecasting accuracy,and that cDFPC has advantages in forecasting both long-term and short-term interdependency function data.The empirical evidence based on the opening price forecast of the Shanghai Stock Exchange Composite Index finds that the intraday volatility pattern and daily interdependency structure are significantly present,and that the relative forecasting advantage of cDFPC remains robust.
作者
徐妍
郭梦霞
Xu Yan;Guo Mengxia(The Second Clinical Medical School,Xuzhou Medical University,Xuzhou Jiangsu 221004,China;School of Statistics and Data Science,Southwestern University of Finance and Economics,Chengdu 611130,China)
出处
《统计与决策》
北大核心
2025年第22期17-23,共7页
Statistics & Decision
基金
江苏省社会科学基金青年项目(25ZHC024)。
关键词
波动模式
相依结构
自适应分类
混合期望
函数型时间序列
fluctuation patterns
interdependent structure
adaptive classification
mixed expectation
functional time series
