This paper considers the estimation of a dynamic linear regression model after a pretest of exact linear restrictions on the coefficient vector. Monte Carlo evidence illustrates that pre-testing can be risk-superior to both ordinary and restricted least squares.