Reduced rank regression with autoregressive errors (Q579823)

The multivariate regression model \[ Y_ t=CX_ t+U_ t,\quad U_ t=RU_{t-1}+\epsilon_ t,\quad t=1,2,...,T, \] is considered where \(Y_ t\) is an \(m\times 1\) vector of dependent variables, \(X_ t\) an \(n\times 1\) vector of independent variables, C an \(m\times n\) regression coefficient matrix of lower rank \(r\leq \min (m,n)\), R an \(m\times m\) matrix of unknown parameters with all eigenvalues less than one in absolute value, and \(\epsilon_ t\) are independently distributed random vectors with mean zero and positive definite covariance matrix \(\Sigma_{\epsilon}.\) Two methods of estimation of the component matrices A and B of \(C=AB\) are discussed subject to the normalization conditions \(A'\Gamma A=I_ r\), \(B(xx'/T)B'=\Lambda^ 2\), where \(\Gamma ={\tilde \Sigma}_{\epsilon}^{-1}\), \({\tilde \Sigma}_{\epsilon}\) being the sample covariance matrix of the residuals \({\tilde \epsilon}_ t\). Asymptotic theory and iterative computational procedures for the two estimators are also presented.