Some properties of the distribution of an operational ridge estimator (Q789848)

Consider the linear regression model \(Y=X\beta +U\) where Y (\(n\times 1)\) represents a vector of observations, \(X(n\times p)\) is a matrix of n observations on p variables such that \(X^ TX=diagonal\) matrix \(\Lambda\) with \(\lambda_ i (i=1,2,...,p)\) as its diagonal elements, \(\beta(p\times 1)\) is an unknown vector of regression coefficients, and \(U(n\times 1)\) is such that \(E(U)=0\) and \(E(UU^ T)=\sigma^ 2I_ n\), \(\sigma^ 2\) being the unknown disturbances variance. It is well-known that the least squares estimator of \(\beta\) is \({\hat \beta}=(X^ TX)^{-1}X^ TY=\Lambda^{-1}X^ TY. {\hat \beta}\) is such that \(E({\hat \beta})=\beta\), and \(E({\hat \beta}-\beta)({\hat \beta}- \beta)^ T)=\sigma^ 2\Lambda^{-1}.\) The general ridge estimator of \(\beta\) is defined by \({\tilde \beta}=(X^ TX+K)^{-1}X^ TY=(\Lambda +K)^{-1}X^ TY=(\Lambda +K)^{-1}\Lambda {\hat \beta}\) where \(K=diag (k_ 1,k_ 2,...,k_ p)\) and it characterizes the estimator. The optimum value of \(k_ i\) obtained by minimizing the total mean squared error involves \(\sigma^ 2\) and \(\beta_ i\). If \(\sigma^ 2\) and \(\beta_ i\) are replaced by their least squares estimators, then the obtained estimator is called operational general ridge estimator \(\beta^*.\) In this paper the properties of the sampling distribution of \(\beta^*\) are discussed. Under the assumption of normality of the disturbances, the exact expressions for the moments of \(\beta^*\!_ i\) are given. Under the assumption that the non-centrality parameter is large, same approximations have been worked out which are used to study kurtosis and skewness of the sampling distribution.