Adaptive ridge-type predictors in finite population (Q1179343)

This paper considers a finite population with \(N\) units labelled \(1,2,\dots,N\). Associated with unit \(i\) is a fixed known \(p\times 1\) column vector \(x_ i\) and an unknown number \(y_ i\). The paper considers the linear model \(y=X\beta+e\), where \(X=(x_ 1,\dots,x_ N)'\) is an \(N\times p\) known matrix, \(y=(y_ 1,\dots,y_ N)'\), \(\beta\) is a \(p\times 1\) unknown superparameter, \(e\) is an \(N\times 1\) random vector with zero mean vector and dispersion matrix \(\sigma^ 2V\), \(\sigma^ 2\) being unknown and \(V\) being a known diagonal matrix. The objective is to predict a linear function \(\ell'y\) from a sample where \(\ell\) is an \(N\times 1\) known vector. Under the assumption of normality of the distribution of \(e\), two adaptive versions of ridge-type predictors are considered. The superiority of these predictors to the best linear unbiased predictor under the stated model has also been shown.