Abstract: The Best Linear Unbiased Prediction for Function Continuation(FCBLUP) is introduced into high-frequency financial data mining, through which this paper constructs discrete observation sequence into functional data model to predict. Adopting high-frequency Shanghai stock index as test data to establish the FCBLUP model, meanwhile, in order to have an effective evaluation and positioning, the prediction model based on Auto Regression Moving Average(ARMA) is built as another group. Experimental results show that the prediction effect of FCBLUP is better than ARMA, its error is slightly larger than ARMA only in small interval, but it is significantly lower than the ARMA at the other times.

Key words: functional continuation, functional data, Best Linear Unbiased Prediction(BLUP), Auto Regression Moving Average(ARMA), high-frequency data

摘要： 将函数延拓最优线性无偏估计(FCBLUP)引入高频金融数据挖掘中，对离散观测值序列建立函数数据模型，并进行预测。选取上证收盘价格为实验数据，建立FCBLUP模型。为能对预测效果进行有效的评价与定位，设立基于ARMA模型的预测组。实验结果表明，FCBLUP预测效果较ARMA模型更理想，FCBLUP预测误差除在小段预测区间略大于ARMA外，其余时刻均低于ARMA预测。

关键词: 函数延拓, 函数数据, 最优线性无偏估计, 自回归移动平均, 高频数据

CLC Number: 

• F830.91