摘要
研究的随机系数回归模型为 y_t=sum from i=1 to s (β_i(t)x_i(t)+e_t) t=0,1,… 其中回归系数β_i(t)是对经典的定常系数的推广,并假定β_i(t),i=1,2,…,s为应用广泛的多元ARMA或ARIMA时序模型或关于时间t的某种函数.并依此给出y_t的多步预报递推公式可进行动态预报.此预报公式仍适用于系数时间序列为非平稳情况.在无初始信息情况下,此预报公式仍具有某种最优性质。
A discussion on random variation coefficient regression model is given:where the regression coefficients are generalized to non- constant , with the assumption that βi(t) i=1,…, 5 are some widely known multivariate time series model (ARMA or ARIMA) or some time dependent functions. This gives rise to N- step ahead iterated prediction formulas of y1. They can be used to give on -time prediction . These formulas are adapted also to non - stationary cases of coefficient series and they keep some kind of optimal qualities in cases without initial information .
出处
《北京理工大学学报》
EI
CAS
CSCD
1992年第1期4-10,共7页
Transactions of Beijing Institute of Technology
关键词
回归模型
随机系数
预测
迭代法
random variation parameter system
multivariate regression analysis
iteration method
predictions
