DOI: 10.1002/j.1521-4036.1976.tb00011.x ISSN: 0323-3847
Minimax‐linear and MSE‐ estimators in generalized regression
H. ToutenburgSummary
One method to combine the MSE technique with restrictions on the coefficient vector in regression is to minimize the maximum of the mean square error risk, whereas the maximum is taken according to a convex set of the E
uklid
ean space containing the coefficients. The so‐called Minimimax‐linear estimators are compared with the MSE‐estimators according to the two risk functions. Further prior information on
σ
2
is used to give practicable estimators those are better than the G
auss
‐M
arkoff
‐estimator. A numerical example is given.