The error bounds of combined forecasting (Q1410660)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | The error bounds of combined forecasting |
scientific article |
Statements
The error bounds of combined forecasting (English)
0 references
14 October 2003
0 references
Let \(f_{it}\) be a forecast of the observation \(y_t\) \((1\leq t\leq m)\) using \(i\)th forecasting method \((1\leq i\leq n)\). If \(\mu_1, \dots,\mu_n\) are nonnegative numbers such that \(\sum \mu_i=1\), then \(f_t=\sum_{i=1}^n \mu_i f_{it}\) is a combined forecasting (CF) value. Define \(e_{it}=y_t-f_{it}\), \(e_t=y_t-f_t\), \(E_{ij}=\sum_{t=1}^m e_{it}e_{jt}\), \(J=\sum_{t=1}^m e_t^2\), \(\mu=(\mu_1,\dots,\mu_n)'\), \({\mathbf E}=(E_{ij})\), \({\mathbf 1}=(1, \dots,1)'\). If \({\mathbf E}\) is regular, then the minimal value \(J^*\) of \(J\) is reached when \(\mu=\mu^* =({\mathbf 1'}{\mathbf E}^{-1}{\mathbf 1})^{-1} {\mathbf E}^{-1}{\mathbf 1}\). Let \(\lambda_1\leq \dots \leq \lambda_n\) be eigenvalues of \({\mathbf E}\). Then \(n^{-1}\lambda_1\leq J^*\leq n^{-1} \lambda_n\). Further the authors prove that \(\mu=\mu_A= n^{-1}{\mathbf 1}\) is optimal iff \({\mathbf E}{\mathbf 1}=C{\mathbf 1}\) where \(C\) is a constant. Some relationships among errors of a general CF method and errors of each forecasting model are given.
0 references
optimal combined forecasting
0 references
simple average method
0 references
forecasting error information matrix
0 references