Second-order risk structure of GLSE and MLE in a regression with a linear process (Q1083807)

The paper deals with estimation of \(\beta \in R^ p\) in the model \[ y_ t=x_ t'\beta +\sum^{\infty}_{j=0}g_ j(\theta)\epsilon_{t- j}, \] where \(\{x_ t\}\) is fixed, and \(\theta \in R^ 1\) is unknown. To obtain a GLSE, \(\theta\) is estimated by the minimizing value \({\hat \theta}\) of a Whittle functional for \(\tilde u=y-X{\hat \beta}\), \({\hat \beta}\) being the LSE for \(\beta\). With the covariance matrix V(\(\theta)\) of the linear error-process \(\{u_ t\}\), the GLSE is \[ {\hat \beta}_{\hat w}=\{X'V^{-1}({\hat \theta})X\}^{-1}X'V^{- 1}(\theta)y. \] \({\hat \beta}_{\hat w}\) is shown to be unbiased and the leading term of the estimation effect of \(\theta\) on the covariance matrix of \({\hat \beta}_{\hat w}\) is evaluated.